<?xml version="1.0" encoding="UTF-8"?>
<rss  xmlns:atom="http://www.w3.org/2005/Atom" 
      xmlns:media="http://search.yahoo.com/mrss/" 
      xmlns:content="http://purl.org/rss/1.0/modules/content/" 
      xmlns:dc="http://purl.org/dc/elements/1.1/" 
      version="2.0">
<channel>
<title>Topos Institute</title>
<link>https://topos.institute/blog/opinion/</link>
<atom:link href="https://topos.institute/blog/opinion/index.xml" rel="self" type="application/rss+xml"/>
<description>An **opinion essay** shares the personal perspective of a Topos community member, as part of opening up our research visions and culture to the world.
</description>
<generator>quarto-1.8.26</generator>
<lastBuildDate>Fri, 18 Sep 2026 00:00:00 GMT</lastBuildDate>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Evan Patterson</dc:creator>
  <link>https://topos.institute/blog/2026-09-18-changing-the-frame-1/</link>
  <description><![CDATA[ 




<div id="post-content" class="page-columns page-full">
<section id="frames-and-formal-models-today" class="level2 page-columns page-full">
<h2 data-anchor-id="frames-and-formal-models-today">Frames and formal models today</h2>
<p>A wide swathe of technical creative activity, from creating scientific models and engineering designs to writing computer programs to making spreadsheets, is aimed at expressing a model of something precisely enough to be operationalized on a computer. The model, or at least the part of it given algorithmic content, is thus in principle mathematizable. We will think of scientific models, engineering designs, computer programs, and spreadsheets all as kinds of <em>formal models</em>, even if the degree to which they are formalized mathematically varies in the current state of practice.</p>
<p>Any formal model exists within some <em>frame</em>,<sup>1</sup> defining the concepts and language within which the model is expressed and the rules to which this expression must conform. A typical frame for scientific modeling might be “a parameterized system of ordinary differential equations,” specific enough to unambiguously describe a class of models and to support uniform algorithms for fitting and simulation. The frame for a computer program is, to a first approximation, the programming language in which it’s written along with the ecosystem of libraries supporting it.</p>
<div class="no-row-height column-margin column-container"><div id="fn1"><p><sup>1</sup>&nbsp;We will not attempt to be too precise about what is a “frame”—for us, a pre-theoretical concept—but the choice of word is intended to evoke both a “frame of reference,” meaning a chosen coordinate system or, more abstractly, a chosen point of view, and a “framework,” defined by <a href="https://en.wikipedia.org/wiki/Framework">Wikipedia</a> as an “essential supporting structure on which other things are built.”</p></div></div><div class="pull-quote-flow-around">
<div class="pull-quote pull-quote-left">
<p>Builders and users of computing technology seem to have revealed an overwhelming preference for unconstrained frames.</p>
</div>
<p>Frames can be more or less constrained. For example, a paperwork form like a tax return is fairly constrained, prescribing a rigid structure for the requested information, while a survey posing short-answer questions is much less so. A relatively unconstrained frame has the obvious benefit of allowing creator flexibility to formally express their intent. A relatively constrained frame carries a more definite and opinionated viewpoint, stipulating stronger structure and assumptions. This makes it easier to say <em>some</em> things, while prohibiting the expression of other things. That is good when it blocks expressions that would be outright mistakes or, what is more subtle, would deviate from settled design decisions or conventions. It is bad when it prevents useful and desirable expressions not anticipated by the choice of frame.</p>
</div>
<p>Builders and users of computing technology seem to have revealed an overwhelming preference for unconstrained frames. From the beginning, the field of computing has, in its theory, emphasized the <em>universality</em> of programming languages; in its practice, language designers herald something more elusive, the “expressivity” of their languages. Meanwhile, the species of mathematized software most widely used by people who are not mathematicians or programmers—the spreadsheet—is highly unconstrained, a blank canvas for tabular data with no further requirements on data layout or typing. Advanced users can introduce their own logic, of arbitrary complexity, within cell formulas, which can vary across cells in arbitrary ways.</p>
<p>It is not hard to imagine what might lie behind this preference for unconstrained frames. Working in a constrained frame, if a creator reaches the boundary of the frame, they may be unable to express their intent at all within the formal model, an experience at best frustrating and at worst intolerable. Think of the anxiety caused by being unsure what to write in your tax return. In the most unconstrained frames, like a general-purpose programming language, it is usually possible to find <em>some</em> way to proceed, even if it involves blurring or subverting what were once crisp abstractions. Put simply, most people would rather muddle through than risk hitting an immovable wall.</p>
<p>Nevertheless, the costs borne under the status quo are severe. With the frame for the formal model having few hard constraints, the typical software project of even moderate complexity relies on a web of semi-formal and informal abstractions described by a mix of code and documentation but not fully specified by either. Rather, the abstractions are maintained—if they are maintained—by a shared understanding among the engineering team, created and renewed through careful communication. Competent programmers create bugs not by trivial logic errors but by failing to understand intricacies of what software abstractions assume or how they interact with each other.</p>
<p>As for spreadsheets, a notorious weakness of the spreadsheet frame is in transitioning from individual use to reliable organizational processes. Because the assumptions behind a spreadsheet are highly implicit, hidden within a web of cell dependencies and in the mind of the spreadsheet’s designer, it is all too easy to make ad hoc changes that silently break those assumptions. The lack of type constraints in spreadsheets is a frequent source of errors, including the perennial joke that Excel interprets every input it sees as a date. In finance and office work, the imperative of reliability has for decades spurred expensive consulting efforts to shift processes from spreadsheets to bespoke software systems—often indecisively, since the replacement systems tend to be less flexible and adaptable for their users.</p>
<p>While different in many respects, what unites the standard paradigms for programming and spreadsheets is that by situating the formal model in a generic and unconstrained frame, the domain-specific abstractions that inevitably support the model possess little formal content; to compensate, disciplined human processes are required to safely use and change the model at any scale beyond a single person.</p>
</section>
<section id="the-effect-of-emerging-llm-technology" class="level2 page-columns page-full">
<h2 data-anchor-id="the-effect-of-emerging-llm-technology">The effect of emerging LLM technology</h2>
<div class="pull-quote-flow-around">
<div class="pull-quote pull-quote-right">
<p>The process of <em>creating</em> models forces us to sharpen and clarify our ideas.</p>
</div>
<p>Such is the status quo, an equilibrium beset by familiar problems but so far ultimately workable. New technology, in the form of generative machine learning based on LLMs, threatens the status quo by changing the tradeoffs involved. It also presents an opportunity to rethink how we design technologies for formal modeling.</p>
</div>
<p>In programming, the emergence of LLMs is decreasing the cost of generating more or less functional code, while exacerbating the challenge of maintaining a coherent set of abstractions and assumptions. The experienced programmer, when attempting to make a change, will notice when the change seems to depart from existing structures or assumptions and take the opportunity to rethink their approach, which can ultimately lead to revising past design decisions or changing the abstractions. The LLM, ever compliant, proceeds along the path of least resistance. More fundamentally, while adept at mirroring surface-level patterns present in the codebase, the LLM is not privy to the web of shared mental representations that enable the effective engineering team to function.<sup>2</sup> Thus, at present, LLM coding agents appear to enable significant short-term productivity gains while gradually eroding the codebase’s overall concision and coherence. As LLMs become more prolific in office work, the lack of explicit structure and typing constraints in tools like spreadsheets will only exacerbate the fragility already present. In both cases, the distinction between <em>creating</em>, an active process that engages the mind and forces it to confront problems, and <em>reviewing</em>, a passive activity often experienced as boring, means that opportunities to catch mistakes or revisit assumptions are more likely to be missed when LLMs are heavily involved.</p>
<div class="no-row-height column-margin column-container"><div id="fn2"><p><sup>2</sup>&nbsp;The use of Markdown files like <code>AGENTS.md</code> is a primitive means to mitigate this problem.</p></div></div><p>With LLMs having become so fluent in formal languages, a mainstream view now holds that the problems around sharing formal artifacts between humans and agents are best solved by having humans cease to directly interact with formal models at all. A formal model is useful, the thinking goes, only insofar as it fulfills some externally validatable purpose; if it does so adequately, it does not matter how the model was constructed or how messy it might be internally. The LLM will mediate the interaction between the human and the formal model exclusively through natural language, so that the human never has to look inside.</p>
<p>Whether or not it is technically feasible, this vision misses much of the point of using formal models in the first place. Be they scientific models or computer programs or spreadsheets or abstract philosophical schemes, mathematical (or at least mathematizable) models have been among humanity’s most successful tools to augment our capacity for thinking. Not only does <em>having</em> good formal models increase our powers of deduction, prediction, and counterfactual reasoning, but the process of <em>creating</em> models forces us to sharpen and clarify our ideas, both for ourselves and to communicate with others. To abdicate a direct role in formal modeling would be to weaken human capability and agency.</p>
<p>The question then is not <em>whether</em> we should continue to directly interact with formal models but <em>how</em> the process of model building should adapt to a potent new source of automation and augmentation.</p>
</section>
<section id="the-future-of-formal-modeling" class="level2 page-columns page-full">
<h2 data-anchor-id="the-future-of-formal-modeling">The future of formal modeling</h2>
<p>All of the foregoing points to the same conclusion: that safe and reliable formal modeling pairing humans with LLM agents requires more constrained frames than are common today. Under our meaning of “frame,” this entails more than a Markdown file making vague suggestions; a frame specifies concepts, language, and assumptions that set the rules for formal modeling and are enforced when type checking a putative model.</p>
<div class="pull-quote-wrap-around">
<div class="pull-quote pull-quote-right">
<p>Insist on more constrained frames, while making it possible to <em>change the frame</em>.</p>
</div>
<p>How can we adopt more constrained frames while retaining the flexibility that has made the present paradigm of unconstrained modeling dominant? The question may seem paradoxical but the answer is, in a sense, simple: by insisting on more constrained frames, while making it possible to <em>change the frame</em> when needed.</p>
</div>
<p>This claim is more easily asserted than realized. Existing languages and tools that serve as frames for formal models tend to be very rigid. Changing them is costly and impractical. There are two reasons to believe that now is the right time to build more malleable tools for formal modeling.</p>
<p>First, as we have seen, LLMs have made it cheaper to generate code and other formal artifacts; consequently, they have made precise specification both more useful as input and more necessary to prevent abstraction drift. There is a growing recognition that existing practices must change. This recognition should create the will to explore new paradigms rather than tinker around the margins of existing ones.</p>
<p>Second, there is now a mature enough body of theoretical knowledge to build the new infrastructure that will allow fluidly defining and changing frames. This body of knowledge is broadly <em>categorical logic</em>, the intersection of category theory with mathematical logic and type theory.</p>
<p>That logic and type theory are the mathematical underpinning for <em>syntax</em> in formal modeling is generally appreciated, and the use of categories as denotational <em>semantics</em> for programming languages is familiar to experts. Much less appreciated, even among experts, is how category theory provides a meta-mathematical toolbox to enable modularity and reuse within and between frames. We refer to specific technical ideas but also to cultural attitudes about how to design formal languages. As we have noted, language designers have overwhelmingly sought to create ever more general and expressive languages, the ideal frame in which to say everything that can be said. This attitude is essentially reductionist. Category theory encourages us to see logic not as something prior to mathematics and science, but as a part of it, and to search for well-adapted mathematical universes in which to create models.<sup>3</sup> These universes can look familiar, like functional programming or first-order logic, but can also natively encompass non-standard features crucial to certain domains, such as resource sensitivity, reversibility, temporality, nondeterminism, probability, and fallibility. Categorical logic offers a uniform technical toolbox to describe these paradigms without privileging any of them.</p>
<div class="no-row-height column-margin column-container"><div id="fn3"><p><sup>3</sup>&nbsp;The evocative phrase “mathematical universe” we borrow from Ingo Blechschmidt’s paper on exploring <a href="https://doi.org/10.1007/978-3-030-84706-7_4">“custom-tailored mathematical universes”</a>, where it means a purpose-built <em>topos</em>. For technically minded readers, we remark that our usage is intended to encompass categorical-logical structures generally, not only toposes.</p></div></div><p>In the <a href="https://catcolab.org/">CatColab</a> project, we are exploring this vision for scientific and cyberphysical modeling, programming in domain-specific logics, and tabular and relational data. It is beyond the scope of this post to describe these efforts in detail, nor do we claim to have a complete account of any of these vast topics. Rather, in the next post in this series, we will hone in on one of the paradigms highlighted above—spreadsheet-like data modeling—as a case study in how to build tools, underpinned by categorical logic, that set constrained frames for formal modeling while also allowing the frame to be changed fluidly.</p>


</section>

</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>LLMs</category>
  <category>modeling</category>
  <category>CatColab</category>
  <guid>https://topos.institute/blog/2026-09-18-changing-the-frame-1/</guid>
  <pubDate>Fri, 18 Sep 2026 00:00:00 GMT</pubDate>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Brendan Fong</dc:creator>
  <link>https://topos.institute/blog/2026-03-18-quanta-can-the-most-abstract-math/</link>
  <description><![CDATA[ 




<div id="post-content">
<div class="card" style="border: 1px solid black;">
  <div class="card-body">
    <h5 class="card-title">Can the Most Abstract Math Make the World a Better Place?</h5>
    <h6 class="card-subtitle mb-2 text-muted">Columnist Natalie Wolchover explores whether applied category theory can be "green" math.</h6>
    <p class="card-text"><i>Like these mathematicians, I yearn to make the world a better place while doing what I love. (Don't we all?) Philosophically, I see the promise in applied category theory. Time will tell whether the approach will genuinely help humanity or the planet. But for those who feel called to do good and to do math, it's worth a try.</i></p>
    <a href="https://www.quantamagazine.org/can-the-most-abstract-math-make-the-world-a-better-place-20260304/" class="card-link">Full post</a>
  </div>
</div>


</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>crosspost</category>
  <category>micro</category>
  <guid>https://topos.institute/blog/2026-03-18-quanta-can-the-most-abstract-math/</guid>
  <pubDate>Wed, 18 Mar 2026 00:00:00 GMT</pubDate>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Kevin Carlson</dc:creator>
  <link>https://topos.institute/blog/2025-05-05-the-sun-is-good/</link>
  <description><![CDATA[ 




<div id="post-content">
<section id="introduction" class="level1">
<h1>Introduction</h1>
<p>What is the world? Is it “everything that is the <a href="https://en.wikipedia.org/wiki/Tractatus_Logico-Philosophicus">case?</a>” Does it have different strata, whose realities are grounded in mutually incompatible ways, or just one? When we describe the world to each other, don’t our choices of language privilege certain answers to these questions, and foreclose on the possibility of certain others?</p>
<p>I won’t waste much time with monism (i.e.&nbsp;the answer “just one”), even when many fellow Western tech-and-science types seem to take it as not only beyond the pale but actually laughable to quibble with the monistic assumption of rank materialism. As Aristotle saw long ago, “things are said to be in many ways”; pace him, for reasons I can’t so easily explain, the following millennia of careful research into the “ways in which things are said to be” have recently been overwritten with nothing much more than, well…a quick snicker at Cartesian dualism followed by a jump to, at best, Dan Dennett, or maybe early Wittgenstein.</p>
<p>Yet, the simple fact remains that things are said to be in many ways, and eliminativisms and materialisms of all kinds are foolish, harmful when taken seriously, and ineffective in following and influencing the unfolding of the world.</p>
</section>
<section id="schopenhauers-view" class="level1">
<h1>Schopenhauer’s View</h1>
<p>A favorite answer of mine to the questions up top is Schopenhauer’s. He says things are said to be in two ways: first, as Will (capitalized <em>á l’Allemand</em>), in which there are no <em>things</em> at all, really, just the surging, chaotic flow of pure existence that lies behind all appearances.</p>
<p>The second, which Schopenhauer calls the world as Representation, is the familiar world of subjects-regarding-objects. What’s interesting is that, even in this world of Representation, Schopenhauer subdivides into four distinct ways in which things are grounded, or in which things can be: cognitive objects, grounded by reason (if all men are pigs and Socrates is a man, then Socrates is a pig), material objects (through causation), moments and places in time and space (grounded by other nearby places and times), and the ground of motive (the connection between intention and action).</p>
</section>
<section id="models-for-real" class="level1">
<h1>Models, for real</h1>
<p>Why am I inflicting this recapitulation of Schopenhauer on you? Well, let me get back to that…First: I’m interested in world modeling—what it is, how to do it, and whether AI can help. Here are some examples of world models:</p>
<ul>
<li><img src="https://latex.codecogs.com/png.latex?%5Cmathrm%7BGDP%7D%20=%20C+I+G+X"></li>
<li>If <img src="https://latex.codecogs.com/png.latex?%5Cmathrm%7BGDP%7D'%20%3E%200">, then all in all the economy is doing well.</li>
<li>God is love.</li>
<li>The vibes are off with that guy.</li>
<li>The rich are, by and large, talented, hardworking, and probably way cooler than you.</li>
<li>The rich are, by and large, greedy, lucky, and probably ugly.</li>
<li><div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="regnet.png" class="lightbox" data-gallery="quarto-lightbox-gallery-1" title="A large regulatory network describing many human cancer pathways"><img src="https://topos.institute/blog/2025-05-05-the-sun-is-good/regnet.png" class="img-fluid figure-img" alt="A large regulatory network describing many human cancer pathways"></a></p>
<figcaption>A large regulatory network describing many human cancer pathways</figcaption>
</figure>
</div></li>
</ul>
<p>A world model is at bottom nothing more or less than a representation in the sense of Schopenhauer: a network of objects: a subject organizing (part of) the world to itself. Modernity is the state of a civilization that can develop precise language and intrepid norms of discourse to permit more-or-less free competition between publicly expressed world models. Postmodernity is the observation that grand public world models seem to often throw the baby (the illegible, fractally complex, organic system determining individual flourishing) out with the bathwater (naive adherence to tradition).</p>
<p>The fundamental problem of today’s civilization is finding what comes after postmodernity, which correctly teaches that excessive reliance on legible public models leads to runaway harm beyond the edges of the model, but then gets mired in inescapable negativity. Some individuals find a way out of nihilism to what might be called a meta-modern attitude of understanding and using modern systems (plural) without identifying with them, but our civilization hasn’t advanced to this way of being at scale. (If you like this paragraph, you’ll love David Chapman’s writings on <a href="https://meaningness.com">meaning</a> and <a href="https://metarationality.com">metarationality</a>.)</p>
</section>
<section id="the-world-is-not-an-object" class="level1">
<h1>The World Is Not An Object</h1>
<p>My point in juxtaposing Schopenhauer’s metaphysics and the dream of meta-modern world modeling is this: a world model is a claim, made by some particular subject, about what is to be attended to (perhaps, what is to receive <em>care</em>), which is pragmatically indistinguishable from a claim about what is real. The middlebrow attitude is that a model approximates the world. This is wrong, because the world is not an object. There is no “ground truth” to compare models to, not because we’re solipsists, but because the world-in-itself is a chaotic surging mess in which no grounds are admitted.</p>
<p>The world is not an object. The only objects are our models of the world. This doesn’t mean models can’t be wrong—if you believe “vaccines cause autism,” I can produce observations that (defeasibly!) contradict it. But it wouldn’t be by “looking at the ground truth”; I’d observe that your qualitative model is inconsistent with what seem to me to be reasonable quantitative models, and we’d debate, presumably, until exhaustion.</p>
</section>
<section id="a-story-about-the-sun" class="level1">
<h1>A Story About the Sun</h1>
<p>Let me tell a story about a day when Alice and Bob, a couple, were discussing the ideal frequency of sunscreen use. Alice thought people should use sunscreen every day. Bob opposed this, citing studies showing most deadly skin cancer risk comes from childhood sunburns, with moderate sun exposure possibly beneficial (via Vitamin D). Alice responded citing anecdotes of contacts who tanned and later died of skin cancer.</p>
<p>This seemingly rational debate concealed deeper models. Alice’s unstated world model factors included “if Bob really cares about me, he’ll listen without being annoying about statistical details.” Bob’s hidden model factors included “I’m highly sensitive to the perception of being forced into something, even when the point is correct” and deeper still, something like “the Sun is Good”—a quasi-religious view.</p>
<p>This pattern—where legible, objective modeling on the surface masks emotional, relational, and spiritual paradigms below—is not special to small-scale, intimate interactions, but characterizes human disagreements generally. Consider NAFTA debates: quantitative economic models may mask deeper motivations about the decay of childhood hometowns, manufacturing jobs’ reality-grounding nature, or—of course—attitudes toward foreigners.</p>
<p>We often have this cargo cult of public conversation where we, complicated bundles of stories and feelings, drape ourselves in black cloth and hold out crisp graphs as the only focus of discussion, hoping no one will look under the drape at the messy, emotional, willful grounds of our complete world model. One clear sign of reaching into meta-modernity would be when the sad, angry, hopeful creatures under the cloth start explaining their real models in all their incommensurability, and only then seek common ground as necessary.</p>
<p>A case at larger scale: in problems like “design a zoning policy for Berkeley,” I want everyone with a stake to potentially contribute. There are serious challenges:</p>
<ol type="1">
<li>Normal people would rather socialize than specify policy preferences</li>
<li>Models range from “Berkeley should never change” to complex quantitative projections</li>
<li>There are many stakeholders with stakes of many, contested, sizes</li>
</ol>
<p>Sortition (jury-duty style participation) could help with problems 1 and 3, but number 2 is deeper. How do you funge the infungible, commensurate the incommensurable? Current options are:</p>
<ol type="1">
<li>Establish a top-down modeling paradigm producing scalar ratings that can be mechanically aggregated</li>
<li>Contributions are in plain text, recombined unpredictably and illegibly by decision-makers to produce a conclusion</li>
</ol>
<p>Both have problems: Option 1 excludes those who can’t express themselves in the approved language, leads to overconfidence, and lacks consensus mechanisms to update. Option 2 determines the course of civilization by who yells loudest nearest the leader.</p>
</section>
<section id="what-is-to-be-done" class="level1">
<h1>What Is To Be Done?</h1>
<p>People haven’t really learned to communicate their world models to each other; modernity was built on narrow exceptions that postmodernity has shown unsatisfactory. How do we get people explaining honestly, fully, how they see the world? This problem has spiritual aspects—getting people to notice how they see the world may require a lot of meditation, for instance. But that part really isn’t my brief.</p>
<p>What do I offer, then? Structures for models: organizing, translating, comparing them. In the sunscreen example, the conversation would have improved if Alice and Bob had made their models explicit. Both could have seen the sensitive emotional points they were approaching instead of debating factual details. I think this would really help, modulo the UX problem of figuring out how to develop these models in the midst of a real, human conversation.</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="bob.png" class="lightbox" data-gallery="quarto-lightbox-gallery-2" title="Bob’s world model (click hamburger menu → new analysis → new cell symbol in new pane → visualization to view graph scrollably)"><img src="https://topos.institute/blog/2025-05-05-the-sun-is-good/bob.png" class="img-fluid figure-img" alt="Bob’s world model (click hamburger menu → new analysis → new cell symbol in new pane → visualization to view graph scrollably)"></a></p>
<figcaption>Bob’s world <a href="https://catcolab.org/diagram/01955a7b-9d05-7c92-a3c6-e5c36b05383b">model</a> (click hamburger menu → new analysis → new cell symbol in new pane → visualization to view graph scrollably)</figcaption>
</figure>
</div>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="alice.png" class="lightbox" data-gallery="quarto-lightbox-gallery-3" title="Alice’s world model"><img src="https://topos.institute/blog/2025-05-05-the-sun-is-good/alice.png" class="img-fluid figure-img" alt="Alice’s world model"></a></p>
<figcaption>Alice’s world <a href="https://catcolab.org/diagram/01955a60-e098-72f1-8bde-f520baf9b2fd">model</a></figcaption>
</figure>
</div>
<p>For those UX problems at the very least, AI support is a critical activator, smoothing over what would otherwise often become a grind of manual model refinement disincentivizing the kind of deep iterated reflection that seems worthwhile even for this kind of quotidian discussion. We need low barriers to entry but high legibility and reliability.</p>
<p>My suggestion, instantiated in the <a href="http://catcolab.org/help">CatColab</a> software being built at Topos and used to make the models shown above, is not to pick a single modeling language but a single way of constructing languages, formal enough to describe translations between approaches. AI can help by scanning possibilities and showing candidate jumps in the space of models. If candidate models have to be expressed in a well-specified language, that solves some AI-slop problems. If humans pick from candidate refinements, rather than relying on the AI to provide a consensus world model out of its broad averaging over humans-whose-writing-is-on-the-internet, that solves more problems.</p>
<p>The key to all this is to use AI to help clarify people’s world models, not press people to align with algorithm-generated models. We must preserve and enrich individuals’ and communities’ abilities to see the world clearly and express their vision as it is, enhanced, never constrained, by new technological affordancesssss</p>
<p><em>Thumbnail credit: Jan Buchholtz on <a href="https://www.flickr.com/photos/jansgate/16926167745">flickr</a></em></p>
<p><em>If you liked this and want to full-length and slightly unhinged version, including rambling footnotes on the dating of the beginning of modernity and all, it’s <a href="https://docs.google.com/document/d/1qyWbE3s1oe30Ufrw56iWIHU_sUa8s114KIYUQqx2clI/edit?tab=t.0">here</a>, below the short version.</em></p>


</section>
</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>informal</category>
  <category>vision</category>
  <category>CatColab</category>
  <category>LLMs</category>
  <category>philosophy</category>
  <guid>https://topos.institute/blog/2025-05-05-the-sun-is-good/</guid>
  <pubDate>Fri, 09 May 2025 00:00:00 GMT</pubDate>
  <media:content url="https://topos.institute/blog/2025-05-05-the-sun-is-good/lemur.jpg" medium="image" type="image/jpeg"/>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>B. Scot Rousse</dc:creator>
  <link>https://topos.institute/blog/2025-04-29-care-at-the-edge-of-automation/</link>
  <description><![CDATA[ 




<div id="post-content">
<div class="callout callout-style-simple callout-none no-icon callout-titled" title="Communicated by [Brendan Fong](/people/brendan-fong)">
<div class="callout-header d-flex align-content-center">
<div class="callout-icon-container">
<i class="callout-icon no-icon"></i>
</div>
<div class="callout-title-container flex-fill">
<span class="screen-reader-only">None</span>Communicated by <a href="../../people/brendan-fong">Brendan Fong</a>
</div>
</div>
<div class="callout-body-container callout-body">
<p>I’m delighted to welcome our first philosopher-in-residence, <a href="https://topos.institute/people/b-scot-rousse/">B. Scot Rousse</a>! In getting to know B over the past few months, I’ve been struck by his insights into the ways technologies can centre and marginalise human care and meaning-making, as well as his deep commitment to serving others and building a world that supports this care. We’re very excited to bring B’s rich, distinctive intellectual tradition into Topos. Moreover, I’m excited to explore how this perspective can be a fresh ingredient in new, Topos-led technologies that empower human communities in this technological era.</p>
<p>Here’s a first post from B, in which he shares a bit more about his intellectual lineage and the questions that drive him.</p>
</div>
</div>
<section id="philosophy-ontology-and-design" class="level2">
<h2 data-anchor-id="philosophy-ontology-and-design">Philosophy, Ontology, and Design</h2>
<p>Technologies are not just the application of scientific knowledge to practical problems. They reshape our space of possibilities, altering how we live, act, and understand ourselves. The design of new technologies is, often quietly, the design of new ways of being human. Our inventions invent us in return.</p>
<p>This insight captures the notion of “ontological design,” introduced by Terry Winograd and Fernando Flores in their 1986 book <em>Understanding Computers and Cognition: A New Foundation for Design.</em> Rapid advances of AI and other technologies today demand that we grapple anew with this startling realization.</p>
<p><a href="image2.jpg" class="lightbox" data-gallery="quarto-lightbox-gallery-1"><img src="https://topos.institute/blog/2025-04-29-care-at-the-edge-of-automation/image2.jpg" class="img-fluid"></a></p>
<p>Take the smartphone. It didn’t merely make telecommunication more convenient. It placed us into a new condition of <em>constant connectivity</em>—reshaping how we learn about events, navigate the physical world, seek social connection, and even become the people we are. The repercussions of this transformation for our collective well-being are still coming into view.</p>
<p>Today is an exhilarating and disorienting time to be alive, and to be thinking about and building new technologies. Advances in AI have reignited fundamental questions about the human predicament: What is language? Intelligence? Communication? Flourishing? What kind of human beings are we becoming? What understanding of our predicament should guide the design and use of AI and other technologies?</p>
<p>In <em>Understanding Computers and Cognition</em>, Winograd and Flores showed that philosophical questions are always at stake in technological design. Every new system, they argued, carries a tacit or explicit stand on fundamental issues: what cognition is, what agency is, what communication is, and so on.</p>
<p>They named the guiding assumptions of the AI research of their time “the rationalistic tradition”: a view that human intelligence consists in formal operations (such as search and inference over explicit representations); that agency is the solving of discrete problems by selecting between definite alternatives; and that communication is the transmission of information.</p>
<p>Winograd and Flores issued a threefold challenge to the rationalistic tradition: (1) to call attention to the hidden philosophical assumptions shaping AI research; (2) to show how these assumptions can both limit our technological capabilities and thwart our imagination for the possibilities of human-machine interaction; and (3) to offer an alternative ontology to guide future design. The urgency of this threefold challenge has been renewed today.</p>
<p><em>Understanding Computers and Cognition</em> argued, quite presciently, that computer systems would become woven into human life as conversational technologies. But not all conversations are alike. Sometimes, for example, we are simply speculating about possibilities; other times we are directly coordinating action in requests, offers, and promises.</p>
<p>Adequately designing software to assist in the execution of such <em>conversations for action</em>, Winograd and Flores showed, required rethinking the nature of communication itself—not as the transmission of information, but as the <em>coordination of commitments</em>.</p>
<p>A promise is not a piece of information. It is a way of shaping and bringing forth the future, together. For example, ride-sharing apps work when a request (“pick me up”) and a promise (“driver arriving”) coordinate action; both hinge on mutual commitment, not just clarity of information.</p>
</section>
<section id="our-need-for-renewed-ontological-reflection" class="level2">
<h2 data-anchor-id="our-need-for-renewed-ontological-reflection">Our Need For Renewed Ontological Reflection</h2>
<p>While the AI paradigm has shifted since the 1980s—from symbolic rules-based AI to neural networks and machine learning—the need for the kind of philosophical reflection exemplified by Winograd and Flores has only deepened. Today’s systems operate differently, but often rest on similarly narrow assumptions about human intelligence, communication, and agency.</p>
<p>We must ask: What are the ontological assumptions guiding AI research today? How might they be limiting not only technical development, but also how we live and interact with AI systems? What alternative conceptions of intelligence, agency, and communication might better orient the future?</p>
<p>These are the kinds of questions that animate my research. I am enthusiastic to explore, alongside the Topos community, how philosophical reflection and technological invention can mutually inform each other, especially as we seek to create technologies that expand, rather than constrict, our capacity for shared sense-making in these turbulent times.</p>
<p>Philosophical reflection becomes especially vital in times of transition and upheaval, when our settled certainties begin to fracture. The world needs bold and rigorous philosophical reflection in our times more than ever. Philosophy helps us see the world anew by helping us question what seems obvious.</p>
<p>We live in an era shaped by a largely unquestioned commitment to efficiency, control, optimization, and problem-solving. Human life itself increasingly appears as a series of problems to solve; and instrumental rationality, the capacity to identify and pursue efficient means to an end, is treated as the highest human excellence.</p>
<p>AI systems today reflect and amplify this logic. We see it in the long-running dream that machines might soon automate all human work. We see it in the hope—expressed again and again—that a superintelligent AI might one day answer all human questions and solve all human problems.</p>
<blockquote class="blockquote">
<p>Machines will be capable within twenty years of doing any work that a man can do.</p>
<p>— Herbert Simon, <em>The Shape of Automation for Men and Management</em>, (1965)</p>
</blockquote>
<p>But in this relentless drive to optimize and solve, we risk forgetting a more basic question: <em>For the sake of what?</em></p>
</section>
<section id="toward-an-ontology-of-care" class="level2">
<h2 data-anchor-id="toward-an-ontology-of-care">Toward an Ontology of Care</h2>
<p>This brings us to the human propensity to care. Caring orients us toward questions of worth: What is worth doing? What tasks are worth automating? What kind of life is worth living? What kind of future is worth creating?</p>
<p>These are questions without technical answers.</p>
<p>Caring shows up in our sense of what matters. It draws our attention, solicits our action, and binds our lives to people, places, and projects that hold meaning for us.</p>
<p>Our propensity to care is likely grounded in our fundamental fragility and interdependence as the peculiar social, mortal, and biological beings that we are.</p>
<p>Traditions from philosophy (especially phenomenology), nursing, education, and sociology have long emphasized that we cannot adequately tend to what matters if we try only to optimize or control it. Indeed, one of my sources in thinking about care is Patricia Benner, a nursing expert who put care at the center of the activity and education of nurses.</p>
<p><a href="image1.jpg" class="lightbox" data-gallery="quarto-lightbox-gallery-2"><img src="https://topos.institute/blog/2025-04-29-care-at-the-edge-of-automation/image1.jpg" class="img-fluid"></a></p>
<p>What if we cannot adequately tend to what matters while trying to optimize and control it?</p>
<p>Think of the breakdowns in friendships or romantic relationships managed as a series of transactions; the ecological damages wrought by industrial agriculture and factory farming; the failures of medical care when patients are reduced to biochemical aggregates instead of being treated as a whole person.</p>
<p>A maniacal apotheosis of efficiency over care lurks behind one of the most famous thought experiments in AI-safety. I am referring to the so-called “paperclip maximizer,” as described in Nick Bostrom’s <em>Superintelligence</em>. Imagine an AI designed simply to produce paperclips as efficiently as possible undergoes an intelligence explosion. Without constraints, it would set about converting the resources of the entire planet (including human lives) into paperclips.</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="image4.jpg" class="lightbox" data-gallery="quarto-lightbox-gallery-3" title="“Paperclip Embrace,” by The Pier Group (Kevan Christiaens, Hillary Clark, Matthew Schultz) at the Misalignment Museum in San Francisco. Photo by B. Rousse."><img src="https://topos.institute/blog/2025-04-29-care-at-the-edge-of-automation/image4.jpg" class="img-fluid figure-img" style="width:60.0%" alt="“Paperclip Embrace,” by The Pier Group (Kevan Christiaens, Hillary Clark, Matthew Schultz) at the Misalignment Museum in San Francisco. Photo by B. Rousse."></a></p>
<figcaption>“Paperclip Embrace,” by The Pier Group (Kevan Christiaens, Hillary Clark, Matthew Schultz) at the Misalignment Museum in San Francisco. Photo by B. Rousse.</figcaption>
</figure>
</div>
<p>The paperclip maximizer is a monster of unconstrained instrumental rationality! A system relentlessly optimizing but impervious to the question of <em>why</em> its goal matters, or for whose sake it should be pursued.</p>
<p>In this sense, the paperclip maximizer is not a mere fantasy; it is a mirror held up to our own technological age. As Martin Heidegger observed, modern technology tends to reveal the world (and we human beings ourselves) as resources to be optimized, ordered, and consumed.</p>
<p><a href="image5.jpg" class="lightbox" data-gallery="quarto-lightbox-gallery-4"><img src="https://topos.institute/blog/2025-04-29-care-at-the-edge-of-automation/image5.jpg" class="img-fluid"></a></p>
<p>When care is thoughtlessly subordinated to efficiency, we risk forgetting to ask the questions that make us human and connect us with our care.</p>
<p>How might a closer attention to the dynamics of care—rather than intelligence narrowly understood as efficient problem-solving—help orient the design and deployment of AI systems and other technologies? My work contributes to such questions by retrieving and extending a tradition of thought centered on the embodied, skillful, and caring dimensions of human life, dimensions often overlooked in dominant conceptions of AI.</p>
<p>This tradition runs through the work of Hubert Dreyfus (<em>Alchemy and Artificial Intelligence</em>, 1965; <em>What Computers Can’t Do</em>, 1972), Stuart Dreyfus (<em>Mind Over Machine</em>, 1986, co-authored with Hubert Dreyfus), Patricia Benner (<em>From Novice to Expert</em>, 1984), Terry Winograd and Fernando Flores (<em>Understanding Computers and Cognition</em>, 1986), to my recent <a href="https://cacm.acm.org/opinion/can-machines-be-in-language/">piece</a>, “Can Machines Be in Language?” (<em>Communications of the ACM</em>, Feb 2024, co-authored with Peter Denning).</p>
<p>The enduring message of <em>Understanding Computers and Cognition</em> still holds: technology design is guided by often unexamined assumptions about cognition, action, communication, and other fundamental human phenomena. Today, we are called to articulate and explore an ontology of care, and to do so in a dynamic interplay with the design and development of emerging technologies.</p>
<p>This orientation aligns closely with the mission of the Topos Institute, where I am honored to be a visiting researcher. I look forward to collaborating with the team at Topos in their mission to “research new technologies that increase capacity for collective sense-making, while also creating a culture of use that is mindful of where technology cannot substitute for what must be done by people.”</p>
<p>To continue exploring the issues raised in this blog post, see B’s Topos Colloquium talk, “<a href="https://www.youtube.com/live/9iO0nEVzWGI">Language, Technology, and Care</a>”.</p>


</section>
</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>technology</category>
  <category>LLMs</category>
  <category>vision</category>
  <guid>https://topos.institute/blog/2025-04-29-care-at-the-edge-of-automation/</guid>
  <pubDate>Tue, 29 Apr 2025 00:00:00 GMT</pubDate>
  <media:content url="https://topos.institute/blog/2025-04-29-care-at-the-edge-of-automation/image4.jpg" medium="image" type="image/jpeg"/>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Brendan Fong</dc:creator>
  <link>https://topos.institute/blog/2025-03-11-category-theory-inspired-bbn/</link>
  <description><![CDATA[ 




<div id="post-content">
<p>You can find a link to the full interview on Eric’s blog below:</p>
<div class="card" style="border: 1px solid black;">
  <div class="card-body">
    <h5 class="card-title">A Category Theory-Inspired BBN</h5>
    <h6 class="card-subtitle mb-2 text-muted">How the Topos Institute tackles problems in complex systems</h6>
    <p class="card-text"><i>In putting together this piece, Brendan and I had over three hours of messy, wide-ranging conversations and interviews. We've distilled them into this final, more readable, interview format.</i></p>
    <a href="https://www.freaktakes.com/p/a-category-theory-inspired-bbn" class="card-link">Full post</a>
  </div>
</div>


</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>crosspost</category>
  <category>micro</category>
  <guid>https://topos.institute/blog/2025-03-11-category-theory-inspired-bbn/</guid>
  <pubDate>Tue, 11 Mar 2025 00:00:00 GMT</pubDate>
  <media:content url="https://topos.institute/blog/2025-03-11-category-theory-inspired-bbn/thumbnail.jpg" medium="image" type="image/jpeg"/>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Hamidah Oderinwale</dc:creator>
  <link>https://topos.institute/blog/2025-02-27-codified-context/</link>
  <description><![CDATA[ 




<div id="post-content" class="page-columns page-full">
<p><em>I’d like to thank David Spivak for his contributions to this post and for his guidance and supervision for this project.</em><sup>1</sup></p>
<div class="no-row-height column-margin column-container"><div id="fn1"><p><sup>1</sup>&nbsp;David Spivak played a key role in refining the questions, thinking through the overall structure, and providing thoughtful edits on both wording and content. He originally suggested this undertaking and helped shape and sharpen the final version.</p></div></div><section id="library-reform" class="level1">
<h1>Library reform</h1>
<p>Libraries collect and organize human knowledge, equipping people with the ability to access, learn, and build upon others’ ideas. In turn, today’s society is equipped to handle problems that are more complex than at any point in history. Still, there is more to document and the scope of our libraries could be meaningfully larger. By expanding what we preserve, we could solve even more complex problems than we do today.</p>
<p>A weakness of today’s knowledge systems is their inability to capture and integrate <em>procedural knowledge</em>. Much of it remains undocumented, locked in our minds, or scattered across sources—thereby going to waste. Debugging, the act of fixing an existing engineered system, is a necessary but tedious step of R&amp;D. Debugging is not only about fixing bugs to make the system functional, it’s about inching towards a vision that likely fuzzily exists in our minds. This vision is made increasingly clear and concrete by tinkering.</p>
<p>Additionally, capturing procedural knowledge helps reveal trails of thought, or even better: <a href="https://oecs.mit.edu/pub/vgigt1aq/release/1">mechanistic explanations</a> behind a conclusion, gives us the capacity to check conclusions and “choose our own adventure” in trying to arrive to our own conclusions in research. Citation is a virtue in traditional academic research, but means to tracing the provenance of ideas could be far more ubiquitous.</p>
<p>As an example of a unique modern form of “library,” take GitHub. Modern platforms like GitHub demonstrate how libraries can evolve to preserve not just code, but the entire development process. Developed rather recently, it simultaneously serves as a platform, a collaborative development environment, and a history tracker for developers.</p>
<p>Researchers used to solely write papers to present their findings. Now, they may link to drafts, data, and code offering a more complete picture of what it took to arrive at a final set of results. GitHub simultaneously supports both the act of conducting research and our ability to share and communicate it with others. As a reformed library itself, it has expanded our ability to document and share current research processes. We can take GitHub as a model of a new kind of library—one that we can apply in many more contexts to preserve and understand the (uniquely human) path to discovery.</p>
</section>
<section id="conserving-thought" class="level1">
<h1>Conserving thought</h1>
<p>The challenge of preserving thought extends beyond storing information—it requires capturing the context, reasoning, and evolution of ideas. Moreover, the purpose of documentation isn’t just to share knowledge with future generations or to fortify institutions; it also helps us preserve and build upon our past work, ensuring that human ingenuity can progress and evolve rather than be lost to time. As I’ve edited, copied, pasted, and deleted text from previous versions of this document, even the version history on my text editor (Overleaf) can’t fully capture the context, goals, and nuances that were present when I first engaged with the project. Arguably, neither can the most sophisticated version control systems. In light of this: <em>How can we effectively collaborate on long-term projects if we struggle to understand the intentions driving the artifacts of our own past?</em></p>
<p>Our memories are the imperfect recall of imperfect snapshots of the past, the “ground truth.” Preserving the integrity and authority of our sources is worth striving for. Doing so will require rethinking the tools that support researchers and developers who value provenance, documentation and then ensuring it is usable and accessible.</p>
<p>To conclude, I will share some questions that emerged from grappling with the points discussed above. I am making them public not only for others with similar curiosities, but also for my own record as I explore them. Together, these questions form an early-stage research agenda.</p>
</section>
<section id="directions-for-future-work" class="level1">
<h1>Directions for future work</h1>
<ol type="1">
<li><p>In his book <em>Libraries of the Future</em> (1965), Licklider presents a vision for a new age of libraries. How do the above ideas align with his vision, and in what ways do they differ?</p></li>
<li><p>How can we design information storage systems that detect and flag errors without altering the original content, ensuring both historical integrity and clarity for future understanding? How do we decide what and when to correct, and how must these corrections be made? How do we keep track of the edits that have been made in the order they’ve been made? And what storage mechanisms allow us to update works while maintaining cost-effectiveness (e.g., by reducing redundancy whilst keeping accuracy)?</p></li>
<li><p>At what point do edits transform a source into something fundamentally different, and how do we measure and express this difference?</p></li>
<li><p>As inspiration for reformed knowledge representations, what analogies can we draw to gain insights from other fields? Biological systems offer powerful models for how knowledge can be preserved and transmitted across generations. As an example, DNA could be considered a generational log of what worked for survival, including the processes and mechanisms for replication for a given cell.</p>
<ul>
<li>Given these analogies (and additional ones we could generate), what can we learn from existing attempts to disclose procedural knowledge in various complex systems?</li>
<li>What fundamental properties allow certain systems—biological, cultural, and social—to preserve and transmit procedural knowledge over long periods, even without explicit design?</li>
<li>How do different <a href="https://en.wikipedia.org/wiki/Code">encoding</a> methods (such as physical, procedural, and symbolic) and various forms of media (e.g.&nbsp;written and auditory) work together to enhance our understanding and retention of information?</li>
<li>How do these systems balance redundancy with efficiency?</li>
</ul></li>
<li><p>How should we catalog know-how? What would a schema or notational system for procedural explanations look like? How could we detail them in a way that includes the necessary context?</p></li>
<li><p>Just as <a href="https://www.geeksforgeeks.org/what-is-hashing/">hashing</a> and <a href="https://en.wikipedia.org/wiki/Checksum">checksums</a>—algorithms used to verify data integrity—ensure that a piece of data is accurately copied or downloaded, they’re also a way to verify data integrity. Similarly, when humans reproduce knowledge—whether through paraphrasing or replicating a textbook solution—we perform a kind of mental “checksum” to verify that an idea or concept has been accurately understood and transferred. <em>Given this, can verification-via-reproducibility help us maintain the integrity of information that been passed down over time?</em></p></li>
<li><p>How might temporal metadata help us track the continuous evolution of intellectual artifacts and thought processes over time? How do we design identifier systems for the artifacts we develop? How might we move beyond commit-based, discrete-time version control systems?</p></li>
<li><p>How might the concepts of authorship and contribution evolve when cataloging research artifacts that take unique forms or are in progress? How do we take account of contributions that don’t rise to the level of citation, such as hearing an interesting thought from a friend? Should we broaden our definition of authorship, or should we create more space to cite unplanned, non-traditional, and perhaps informal academic interactions?</p></li>
<li><p>How can we support the R&amp;D needed to develop these technologies? Would academia be a feasible place for this work? How have relevant research projects been supported in the past?</p></li>
<li><p>How can we (better) capture the “negative space" of knowledge work: the dead ends, failed attempts, and pivot points that led to final works?</p></li>
<li><p>What does it look like to have dynamic interfaces that adapt to individuals? What user information is required to achieve this, and how can it be acquired in a <a href="https://people.csail.mit.edu/henrycg/pubs/tiptoe/.">privacy-preserving</a> <a href="https://arxiv.org/abs/2412.13678">way</a>?</p></li>
</ol>


</section>

</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>technology</category>
  <category>vision</category>
  <guid>https://topos.institute/blog/2025-02-27-codified-context/</guid>
  <pubDate>Thu, 27 Feb 2025 00:00:00 GMT</pubDate>
  <media:content url="https://topos.institute/blog/2025-02-27-codified-context/library.jpg" medium="image" type="image/jpeg"/>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>David Spivak</dc:creator>
  <link>https://topos.institute/blog/2024-08-27-plausible-fiction/</link>
  <description><![CDATA[ 




<div id="post-content">
<section id="introduction" class="level1">
<h1>Introduction</h1>
<p>Each of us can see problems in the world, cases where something we care about is in need of attention. On the global scale, I am deeply concerned about rampant pollution, countries engaged in hostile acts of aggression, political corruption, cultural fragmentation, the effects of modern technology on human lives and livelihoods, and more. To myself and I believe many other people, the world seems to be cracking under some sort of immense pressure, with “no one at the wheel” to steer the ship.</p>
<p>At the same time, we humans are imaginative. According to historian Yuval Harari, our superpower lies in creating fictions that can become real through our collective belief and action. This manifests in several ways:</p>
<ul>
<li>Social constructs: The value of money is a fiction that becomes real because we collectively believe in and act on it.</li>
<li>Technological innovation: Many ideas from science fiction, such as mobile phones or virtual reality, have eventually been realized.</li>
<li>Institutional change: New laws or social movements often start as ideas, and gain traction until they are eventually part of the social fabric.</li>
</ul>
<p>These examples show how our “fictions” can shape the flow of events.</p>
<p>Rather than waiting for someone else to steer our ship, perhaps we can harness this imaginative power to solve our problems. In this post, I’ll begin by considering how it is that what we propose in language can become reality. Then I’ll propose an idea I’m calling plausible fiction, based on this consideration, and how implementing this idea could lead to a good future for humanity.</p>
<p>As we’ll see, plausible fiction is a lightly-structured approach to imagining and implementing viable solutions to real-world problems. It works by encouraging people to outline a path from our current reality to a desired future, and inviting collective effort to fill in the gaps along the way. This collaborative gap-filling process refines and realizes ideas that would otherwise remain abstract or incomplete, and in so doing move them from fiction to reality. I believe that this approach could be applied to a wide range of issues, at a wide range of scales, and potentially lead to a more participatory world.</p>
</section>
<section id="when-are-proposals-enacted" class="level1">
<h1>When are proposals enacted?</h1>
<p>Language clearly has the power to shape the dynamics of reality. Indeed, if I say “pass the salt,” it is empirically the case that with high confidence I can expect that some 10^25 atoms of salt-in-shaker will move through space and that subsequently a few grains of salt will land on my soup. At another scale, the symbolic language of DNA nucleotides codes for complexly-orchestrated chemical processes, whereby proteins are made and bodies are formed. It’s clear that language works; if it didn’t do anything in material reality, no one would use it.</p>
<p>So why not solve our problems by saying “abracadabra, microplastics be gone!”? Obviously, not just any linguistic utterance will result in the reality we want. In order to “come true” or obtain, the utterance needs to be part of a process that starts now, proceeds through the outcome, and obeys the laws of nature throughout.</p>
<p>If someone were to say “pass the salt” in an empty room, or with only their cat to hear them, would it work? Maybe the cat knows English and can decode what the person meant, decides to help out, and gets off the comfy couch to slide the salt shaker over with her paw. Yes, maybe, but it doesn’t sound plausible to me. Similarly, when you think someone is very angry at you, you might be less likely to ask them to pass the salt, because you’re less sure they’ll do it.</p>
<p>The better we understand reality as it is, the more we can live in accordance with it. Some might say we “shape it”, but I think it’s important to see that our understanding of “reality” means that something has already shaped us, e.g.&nbsp;how our brains encode our relationship to it. If we understand quantum mechanics well enough, we can invent a laser; otherwise, we simply can’t. Our stories about Edison developing the practical light bulb involve him making hundreds of attempts: only when he sufficiently understood both the material world and the social world did his proposal catch on. Thus “connecting the dots” between what was wanted by society and what was possible with available resources, the lightbulb and the laser could be reified. This ability to connect the dots between have and want arises from the work of attentive proposing and listening, but the connection itself can stand as a path of low resistance for achieving the result.</p>
<p>What I’m suggesting may be reminiscent of “following the Tao,” or listening to the way of things, but it is not meant to be mysterious at all. The nonaction spoken of as “Wu Wei” is probably the noticing of the low resistance that comes from proposing something that is in accordance with how things work naturally. Common law—the codification of what already exists as customary norms—is the easiest for a government to enforce, a friend proposing an activity they know the group enjoys is the easiest way to get its buy-in, and a hypnotist acting with, not against, their subject is the best way to get results. That is, when we act so as to usher in something that is already supposed to happen, by which I mean already poised or potentiated to happen, it is easier to be successful.</p>
<p>This is also reminiscent of active inference or predictive processing. A successful proposal should be pretty close to a prediction. My outsider understanding of active inference is that we can replace “I want X to happen” with something like “given that I exist and have the needs and abilities that I do, and given that the world is the way that it is, I predict that X will happen”. Wanting and guessing come together in a story.</p>
<p>The last image I’ll give is of lightning. Suppose there is a difference of electrical potential between the ground and the sky. Then nature “wants” to—is drawn to—equalize the two sides by making a spark between them. And if we put up a lightning rod, we propose a path of lower resistance, which tends to attracts such sparks to pass through it.</p>
<p>While a lightning rod can be seen as a kind of proposal, the proposals I mostly want to consider here are spoken or written using language. I believe that language attracts activity too, in some ways just like a lightning rod does. Saying “pass the salt” calls up all sorts of neural and then motor activity within my friend. Inventors use notebooks filled with symbols to attract their own attention to useful way points, and they use other semiotic schemas to attract certain behavior by the factory that makes the product, i.e.&nbsp;they lead the factory to successfully follow their design. Computer programs, written in highly-controlled languages, attract the voltage changes that cause all our applications to run.</p>
<p>And mathematics, from which computer science emerged, has for millenia been in the business of creating high-fidelity language for engineering and navigating—attracting outcomes in—our world. It compresses our ideas and makes them legible and unambiguous, and hence attractive to people. As <a href="https://youtu.be/kdUkqW7m5-A?si=AUjAqRxauzff2fbc&amp;t=6581">Barry Mazur said</a> in 2009,</p>
<blockquote class="blockquote">
<p>Imagination needs something [in order] to apply itself directly to the largest realm it can possibly apply itself, and for that it needs conveyances, like bridges. And perhaps mathematics is that from my point of view: it’s a way of conveying the imagination to its broadest reach.</p>
</blockquote>
<p>(Thanks to TJ for directing me to the Mathematics and Beauty roundtable, and for other advice throughout the article, including suggesting relevance with the work of the Frankfurt critical theory and separately of Michael Bratman.)</p>
<p>By naming an idea concisely and then showing how it unpacks, we can attract broader attention to it, enabling coordination among those who think it is worth implementing.</p>
</section>
<section id="plausible-fiction" class="level1">
<h1>Plausible fiction</h1>
<p><em>Plausible fiction</em>, as I can define it here, is a story that attempts to satisfy the following criteria:</p>
<ol type="1">
<li>It begins with the world as it is now, including all relevant details and without distortion;</li>
<li>It ends with a future that the author desires;</li>
<li>It follows a trajectory that obeys natural laws and social dynamics; and</li>
<li>It is <em>memetically fit</em>, i.e.&nbsp;it invites participation by those who read it.</li>
</ol>
<p>While it may seem paradoxical to call these narratives “fiction” when our aim for them is to manifest in reality, using the term “fiction” lightens the burden on the author and reminds us that this is an imaginative process of envisioning a future that hasn’t yet materialized. “Pass the salt” exists in our minds as something like “my friend will pass the salt,” and this story still exists in our own imagination—even if it is quite plausible—until it is reified.</p>
<p>Telling plausible fiction is supposed to be a fun and social activity, but with a functional purpose. The way it works is that one crafts a proposal that aims to influence events by supplying a path of low resistance from the present moment to a preferred future.</p>
<p>But what makes a proposal more likely to work is the process of gap-filling. Every plausible fiction has gaps in its narrative, and each gap is either plausibly fillable, or implausible. Once we find an implausible gap, the fiction becomes implausible and we stop working on it. But when the fiction is actually plausible and compelling, the work to making it real is simply to fill in each of its gaps with more plausible fiction: start where the story leaves off, end where the story picks up, and follow a plausible trajectory throughout. The gap-filling process is recursive, like zooming into a fractal, because even a filled gap leaves smaller gaps—more detail—to be filled in. But as smaller and smaller gaps are filled in, real material conditions will start to move. As my brain fills in the fact that you want me to pass the salt, and then fills in how I’m going to maneuver to do it, and then what it’s going to feel like to grab the salt shaker, my muscle cells actually become activated and the salt actually gets passed.</p>
<p>In other words, I’m making the following conjecture:</p>
<blockquote class="blockquote">
<p><em>The only thing it takes to enact plausible fictions is to recursively and plausibly fill the gaps.</em></p>
</blockquote>
<p>Imagine that one person proposes a difficult but plausibly-true math theorem, but that they don’t know a proof. If a few other people suggest plausible results that would combine to prove the theorem, this would be what I’m calling filling the gaps. The biggest gaps get filled first; if people start to think that filling them is implausible, the proposal is dismissed. But as gaps get filled, the theorem slowly becomes proved.</p>
<p>Similarly, imagine that a restaurant chef suggests a meal to be made. Then the sous chefs fill in the details. No one needs to explicitly tell the sous chef “to cut the carrot, move your arm like this”; but their brain does need to fill in that gap. Eventually the gap-filling descends beyond the level humans need to think about, but I claim that creating the meal is gap-filling all the way down.</p>
<p>Very loosely, the mathematical infrastructure I’m imagining to help manage all this collective gap filling is based on applied category theory. Category theory provides a language for describing and connecting diverse mathematical structures, so we may be able to use it to create a formal system for relating and composing different plausible fictions. Exactly how this might work is a gap which remains to be filled within my own fiction here, but luckily creating formal systems to relate and compose things is precisely the kind of problem we try to solve at Topos Institute, so it’s not implausible to me. If a reader has ideas, please let me know!</p>
<p>But supposing we had mathematical infrastructure to facilitate collective gap-filling, then all we’d need are the plausible fiction writers. As long as our vision starts from the present, is compelling as a possibility, and doesn’t require any magical thinking, I believe people will be motivated to fill in the gaps.</p>
<p>A collaborative process of imagining and refining plausible futures, especially one that seeks varied perspectives, could lead to a more participatory world, where more people become inventors of reality. Much like how Edison’s understanding of both material possibilities and social needs allowed him to develop a commercially viable light bulb, plausible fiction aims to bridge our current reality to a desired future. The more people consider and work out what process would actually bring about a good future, the more likely we are to get there. This approach has the potential to democratize the shaping of our collective future, enabling more people to contribute meaningfully to solving global challenges.</p>
<p>Of course, not all proposed futures will be deemed plausible or beneficial, and so we need a process by which people can examine narratives for viability and desirability. In order to steer <em>away</em> from a plausible fiction you don’t like, try to identify a gap you deem either unfillable or for which you can provide a plausible gap-filling story that contains a negative arc. As such a “defeater” fiction gains traction, the original story may become dormant, with fewer people engaging with or developing it, or it could even become a center around which there is active work to counter it.</p>
<p>What I’m proposing is not very different from the situation we have now. We already tell each other plausible fictions in a dialectic of proposing and critiquing that helps refine our collective vision. But I think we can do this more deliberately and effectively. And like our current system, if this system were effective, it would involve risk. As my wife points out, risk is a tenet of life: one cannot truly live without risk. Nevertheless, I believe that clarifying a structure for this ongoing dialogue and refinement of ideas is crucial for the development of robust, well-considered futures.</p>
<p>I imagine a future where humans do labor that is less stressful because it’s closer to the work they want to be doing, and that a good plausible fiction platform might help us get there. In fact, this vision serves as an example of plausible fiction: it starts with our current reality, where work is often stressful, ends with a desirable future where work is more fulfilling, and suggests that a plausible fiction platform makes that trajectory plausible. But there is a big gap, already identified before: realizing this future depends on developing effective mechanisms for collectively sharing and refining such plausible fictions.</p>
<p>So what exactly is a positive future? I’d love to hear your responses and ideas. For example, here are some ways to engage:</p>
<ul>
<li>Share a plausible fiction of your own in the comments, starting with a current challenge and ending with a positive future.</li>
<li>Find gaps in the plausible fictions I’ve presented here, or those in the comments, and possibly propose ways to fill them.</li>
<li>Offer technical suggestions for how we might formalize the process of connecting and composing plausible fictions.</li>
</ul>
<p>By collaborating on these plausible fictions, we can cultivate a better future. My proposal is that plausible fiction may offer a functional, memetically-fit system to bring about what I consider a good future.</p>
<hr>
</section>
<section id="epilogue" class="level1">
<h1>Epilogue</h1>
<blockquote class="blockquote">
<p><em>David:</em> Hi Claude. Please read the above and identify 3 gaps. Then choose one of them and try to fill it with plausible fiction of your own.</p>
</blockquote>
<blockquote class="blockquote">
<p><em>Claude:</em> I’ve read the document about plausible fiction by David Spivak. Here are three gaps I’ve identified in the proposal:</p>
<ol type="1">
<li>The specific mathematical infrastructure based on applied category theory that would help manage collective gap-filling.</li>
<li>The mechanism for determining which plausible fictions are most beneficial or desirable to pursue.</li>
<li>How to incentivize and facilitate widespread participation in creating and refining plausible fictions.</li>
</ol>
<p>I’ll attempt to fill the third gap with some plausible fiction of my own:</p>
<h2 id="incentivizing-widespread-participation-in-plausible-fiction" data-anchor-id="epilogue">Incentivizing Widespread Participation in Plausible Fiction</h2>
<p>In 2025, a group of tech entrepreneurs, social scientists, and game designers collaborate to create “FutureForge,” a gamified platform for developing and refining plausible fictions. The platform combines elements of social media, collaborative writing tools, and prediction markets to engage a wide audience in the process of imagining and shaping potential futures.</p>
<h3 id="key-features-of-futureforge">Key Features of FutureForge:</h3>
<ol type="1">
<li><p><strong>Intuitive Interface</strong>: Users can easily create, view, and contribute to plausible fictions through a user-friendly web and mobile app.</p></li>
<li><p><strong>Reputation System</strong>: Contributors earn “Foresight Points” based on the quality and impact of their contributions, as rated by peers and expert reviewers.</p></li>
<li><p><strong>Collaborative Editing</strong>: Wiki-style tools allow for seamless collaboration and version control of plausible fictions.</p></li>
<li><p><strong>Visualization Tools</strong>: AI-powered visualization tools help users create compelling infographics and animations to illustrate their ideas.</p></li>
<li><p><strong>Integration with Education</strong>: Schools and universities incorporate FutureForge into their curricula, teaching students to think critically about the future and contribute to solutions.</p></li>
<li><p><strong>Corporate Challenges</strong>: Companies and organizations sponsor challenges on the platform, offering rewards for the most promising solutions to specific problems.</p></li>
<li><p><strong>Prediction Market</strong>: Users can invest virtual currency in the likelihood of various plausible fictions coming true, adding a layer of crowd-wisdom to the process.</p></li>
<li><p><strong>Real-World Impact Tracking</strong>: The platform monitors and highlights real-world developments that align with or stem from plausible fictions created on the site.</p></li>
</ol>
<p>As FutureForge gains traction, it becomes a go-to platform for policymakers, innovators, and citizens looking to shape the future. By 2030, millions of users worldwide regularly contribute to and discuss plausible fictions, leading to accelerated problem-solving and more participatory governance at local, national, and global levels.</p>
<p>The success of FutureForge inspires similar platforms and methodologies, creating a cultural shift towards more proactive and collaborative approaches to addressing societal challenges. This widespread engagement with plausible fiction helps to bridge ideological divides, as people from diverse backgrounds work together to imagine and create better futures.</p>
<p>This plausible fiction attempts to address the gap of incentivizing and facilitating widespread participation by proposing a gamified platform that makes contributing to plausible fictions engaging, rewarding, and impactful. It starts with our current technological capabilities and social media engagement, and progresses towards a future where collaborative future-thinking becomes a widespread and influential practice.</p>
<p>The fiction aims to be plausible by building on existing technologies and social trends, such as gamification, collaborative platforms, and the increasing integration of technology in education and policy-making. It also addresses potential challenges, such as maintaining quality contributions and ensuring real-world impact.</p>
<p>Would you like me to elaborate on any part of this plausible fiction or discuss how it might be further refined?</p>
</blockquote>
<hr>
<p>This work was funded by AFOSR grant FA9550-23-1-0376, “Structure and dynamics of working language” and a grant from the Mathematics and Metaphysics Institute, “Accountable metaphysics”.</p>
<p><em>Thumbnail by <a href="https://unsplash.com/@cdd20?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">愚木混株 cdd20</a> on <a href="https://unsplash.com/photos/a-bunch-of-boxes-with-a-star-on-top-of-them-fehk0T5g9Ew?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Unsplash</a></em></p>


</section>
</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>informal</category>
  <category>vision</category>
  <guid>https://topos.institute/blog/2024-08-27-plausible-fiction/</guid>
  <pubDate>Tue, 27 Aug 2024 00:00:00 GMT</pubDate>
  <media:content url="https://topos.institute/blog/2024-08-27-plausible-fiction/cdd20-fehk0T5g9Ew-unsplash.jpg" medium="image" type="image/jpeg"/>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Sophie Libkind</dc:creator>
  <dc:creator>Priyaa Varshinee Srinivasan</dc:creator>
  <link>https://topos.institute/blog/2024-08-02-rethink-math-talks/</link>
  <description><![CDATA[ 




<div id="post-content">
<p>Here’s a classic math joke.</p>
<blockquote class="blockquote">
<p>In an one-hour math talk, most follow the first 15 minutes, only the experts follow the next 15 minutes, only the speaker follows the next 15 minutes, and only God and Gauss follow the last 15 minutes.</p>
</blockquote>
<p>Of course God and Gauss cannot contribute to a discussion, so I’ve been to many math talks that have ended with an awkward silence, and a Q&amp;A that only engages one or two audience members. Can we build a tradition of math talks which invites the audience to engage not only with the speaker but also with each other? That is, I dream of a seminar that starts as a lecture and ends as a conversation. The speaker, novices, experts, God, and Gauss — everyone comes together by the end of the talk.</p>
<p>Picture this:</p>
<blockquote class="blockquote">
<p>It’s 10:55 and there are 10–15 people sitting in a rough semi-circle facing a trio of whiteboards. There’s the happy buzz of conversation and cookie munching. They’ve arrived early because they value the seminar and are excited to hear what the speaker has to say. You, the speaker, are ready, relaxed, and excited to share a cool story from your research.</p>
<p>When the clock chimes 11, the host introduces you, and you take charge of the room. Your role for the next hour is to be a storyteller and also a moderator. It’s a tall order! But the audience is friendly and willing to give you their attention.</p>
<p>For the first 40 minutes you present your talk, spaciously introducing each concept piece by piece. You often pause to check if the audience is present and has any questions.</p>
<p>As is the custom, the audience asks a question only when it is necessary to follow your talk, to clarify concepts, or if there is some mistake to be rectified. You bookmark the exploratory and technical questions for the end. Once or twice, the audience is suspiciously silent. You are nervous that you have lost them, and so you explicitly ask a couple of audience members to share where their attention is.</p>
<p>You finish presenting your talk, and the discussion begins. Because you made sure the audience were with you throughout the talk, everyone — from experts to novices — participates in the discussion. At first, all of the questions are aimed at you. But, whenever the question is beyond your expertise, you welcome contributions from the audience who happily share their insights. Some questions go into technical details while others are more exploratory. Each question often leads to a few minutes of back-and-forth that engages multiple people in the audience.</p>
<p>At noon, the group breaks to go to lunch. But the conversations continue. The next day, a student asks you about a specific detail they were interested in. You feel quite satisfied that you have created a ground for rich discussion and are reaping the rewards.</p>
</blockquote>
<p>Now, does that sound like a good talk?</p>
<hr>
<p><strong>Sophie</strong>: This model of a talk was based on a recent experience in Tallinn, right? Can you say a little bit more about it? What felt remarkable about it?</p>
<p><strong>Priyaa</strong>: Yes! So in that seminar I gave a 45 minute talk on “Polynomial comonoids are categories” which was followed by 45 minutes of impromptu questions. At first I was overwhelmed by the questions and felt the need to answer them all. However, that was simply not possible because the questions were diverse and each one was coming from the asker’s unique background. However, eventually the Q&amp;A turned into a participatory discussion, and my job was more to moderate than to have all the answers ready. I realized later in the day that I may not have appeared impressive (since I didn’t tackle each question by myself) but I felt happy that I’ve given everyone from experts to novices something to walk away with. In fact, I think that was what made the seminar go so well: my not-knowing everything and my willingness to listen.</p>
<p><strong>Sophie</strong>: I’m excited about this idea. As you know, I’ve been refreshing the Berkeley Seminar here at Topos and I am curious to try out a format like this. One thing that I’m nervous about is that it takes a lot of individual and group-level skills to track and maintain the attention of 10–15 people. This sort of attention tracking seems important both (1) for the speaker to make sure they are not losing the audience during their talk and (2) to have a friendly whole-group discussion afterwards. What sort of attitude do you think will help us gain these skills?</p>
<p><strong>Priyaa</strong>: Confidence and care. I think it’s important that the speaker approaches the talk not as “I know everything” or even “I have to know everything”. If we walk in with the attitude (both the audience and the speaker) that we are here to share and learn then we can have a more relaxed and inviting setting. The second thing that seems really important is patience. It’s important for the speaker to be considerate with how much they are asking the audience to digest in the 40 minute time frame.</p>
<p><strong>Sophie</strong>: Yeah, I was thinking about the time, and I’ve tried to give a 40 minute talk but it’s just so short! Especially if I want everyone to follow, and so I’m including a lot of background material.</p>
<p><strong>Priyaa</strong>: Yes, it is indeed a difficult task. However, if you have a core audience that is there each week, then you can take time to first solidify their knowledge over time. And then reap its benefits when you present more complex talks as the seminar evolves.</p>
<p><strong>Sophie</strong>: That basically answers my next question which was what setting can you see this working in vs.&nbsp;in what setting does the traditional math talk make more sense.</p>
<p><strong>Priyaa</strong>: Yes, I think a traditional talk is probably great in a conference setting where there’s a large, ephemeral audience that you don’t know deeply. On the other hand, I think this sort of talk would be excellent in a setting where there is a recurring audience who are ready for conversations and who want to get to know each other over a period of time.</p>
<p><strong>Sophie</strong>: Just like the Berkeley seminar.</p>
<p><strong>Priyaa</strong>: Haha, yes!</p>
<p><strong>Sophie</strong>: Thanks so much for sharing this! I’m excited to see how the Berkeley seminar evolves. Join us at Topos Berkeley most Tuesdays at 11 am. You can see the schedule and RSVP at <a href="../../events/berkeley-seminar/">topos.institute/events/berkeley-seminar</a>.</p>
<p><a href="invitation.png" class="lightbox" data-gallery="quarto-lightbox-gallery-1"><img src="https://topos.institute/blog/2024-08-02-rethink-math-talks/invitation.png" class="img-fluid"></a></p>


</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>outreach</category>
  <category>informal</category>
  <category>events</category>
  <guid>https://topos.institute/blog/2024-08-02-rethink-math-talks/</guid>
  <pubDate>Fri, 02 Aug 2024 00:00:00 GMT</pubDate>
  <media:content url="https://topos.institute/blog/2024-08-02-rethink-math-talks/invitation.png" medium="image" type="image/png" height="170" width="144"/>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Owen Lynch</dc:creator>
  <link>https://topos.institute/blog/2024-03-13-a-retrospective-on-the-oxford–topos-meeting/</link>
  <description><![CDATA[ 




<div id="post-content">
<p><em>You can find a copy of the contents of this post on the <a href="https://forest.localcharts.org/ocl-001O.xml">LocalCharts Forest</a>.</em></p>
<section id="overview" class="level2">
<h2 data-anchor-id="overview">Overview</h2>
<p>Sam Staton, Paolo Perrone and I organized a small meeting between Topos Institute and the Oxford CS department. It was kind of like a workshop, but because of the short time frame that we planned it on and limited funding, we didn’t invite all the people we’d like to invite or have it open to applications, so we called it a “meeting.” Anyways, we managed to make some good progress on some open problems within categorical systems theory, and it was a lot of fun, so this is a short retrospective on what worked, what didn’t work, and directions to pursue in the future.</p>
<p>The meeting was fairly loosely structured. We had two talks at the beginning, one from David Jaz on double categorical systems theory, and another from Sam Staton on LazyPPL. The rest of the time was spent on:</p>
<ul>
<li>Talking in small groups about math.</li>
<li>Writing up what we talked about on the LocalCharts forest.</li>
<li>Explaining things we talked about to the whole group.</li>
</ul>
<p>I opened the meeting by asking participants to do three things.</p>
<ol type="1">
<li>As much as possible, attempt to ground any new theory with concrete examples.</li>
<li>Write things down on the forest on the day that you discussed them, so that you won’t forget them, and so that people interested in the topics of the meeting who didn’t attend the meeting wouldn’t be too left out.</li>
<li>Be comfortable with the idea that you might spend three hours teaching existing theory to people who aren’t familiar with it yet. Transmitting knowledge is a good outcome of the workshop and not at all a waste of time.</li>
</ol>
<p>The last request was I think the most successful idea; people were pretty happy at the end of the workshop about things that they had learned. For the second request, I tried to set aside time at the end of each day to write, but it was very tempting to continue conversations into this time instead of writing, and also it was somewhat hard to write at the end of the day, when everyone was tired from doing math all day and looking forward to dinner. The first request I think was a good idea, but very easy to forget when you are a category theorist! So I think that it’s worth asking people to do in the future, even though we weren’t very good at living up to it.</p>
</section>
<section id="content" class="level2">
<h2 data-anchor-id="content">Content</h2>
<p>We gathered together things that people wrote at the workshop into <a href="https://forest.localcharts.org/oxford-topos-meeting-2024-output.xml">Oxford-Topos Meeting 2024 - Outcomes</a>. Some people ended up writing a lot, others none at all. I ended up not writing very much because I was hovering around helping people install forester. Hopefully in future events, everyone will have forester installed and be used to forester syntax before designated writing times.</p>
<p>So my dream of having all of the discussions captured on paper for those who weren’t present didn’t quite materialize. But fortunately I can talk a little bit now about some of the topics of the discussions that I participated in.</p>
<hr>
<section id="port-hamiltonian-systems" class="level3">
<h3 data-anchor-id="port-hamiltonian-systems">Port-Hamiltonian Systems</h3>
<p>I ended up in two discussions on port-Hamiltonian systems. Both of these discussions were somewhat one-sided, in that they mostly consisted of me explaining what I did in <a href="https://forest.localcharts.org/lynch-relational-2022.xml">my masters thesis</a>. I want to emphasize that I did not set out for this meeting to consist of me shilling for my own work, but it seemed like people were interested and enjoyed learning about it.</p>
<p>However, I was especially pleased that after I went through some of the big gaps in my thesis which had to do with my lack of knowledge of differential geometry: Paolo Perrone was kind enough to teach me some intuition about integrable forms. Specifically, the kernel of an integrable 1-form <img src="https://latex.codecogs.com/png.latex?%5Comega%5Cin%5CGamma(T%5E*X)"> is the tangent bundle of a codimension-1 foliation. He told me to imagine this like sedimentary rocks: the manifold is divided up into layers, and the kernel of <img src="https://latex.codecogs.com/png.latex?%5Comega"> consists of directions that travel along a single layer.</p>
<p>Then, as far as I understand it, the idea Paolo was proposing was to replace the relations that I use in my thesis with something like forms which vanish on the relations. The problem that I was running into in my thesis is that, when thought of as relations, linear subbundles of vector bundles don’t necessarily compose because of constant-rank issues. Perhaps moving to forms would allow me to talk about non-constant-rank linear subbundles? Anyways, I’m excited to investigate this direction, and not having a good intuition for integrable forms was something that had bothered me for a while so I was happy to learn about that.</p>
</section>
<section id="stochastic-behaviour" class="level3">
<h3 data-anchor-id="stochastic-behaviour">Stochastic Behaviour</h3>
<p>Another group I participated in tackled the problem of stochastic behavior of dynamical systems. There is a good story for “representable behaviors” within categorical systems theory, but it was unknown how to generalize this to talk about behaviors of a Markov chain.</p>
<p>We were able to come up with a definition for representable stochastic behavior which mimicked the classical notion of “a stochastic process adapted to a filtration” using some techiniques from <a href="https://ncatlab.org/nlab/show/quasi-Borel%20spaces">quasi-Borel spaces</a>. I wrote up some preliminary notes on this <a href="https://forest.localcharts.org/ocl-001N.xml">here</a>, but that does not capture where we ended up going on this topic, and hopefully there may end up being a paper on this.</p>
<p>Funnily enough, our group was originally interested in trying to make a categorical systems theory for stochastic differential equations, but we ended up getting sidetracked after we slogged through an hour of half-remembering functional analysis. There were some promising directions here that I hope we circle back around to though.</p>
</section>
<section id="double-operads" class="level3">
<h3 data-anchor-id="double-operads">Double Operads</h3>
<p>I was not in the group that discussed double operads, and in fact to a certain extent, I don’t think it was a group, it was a one-man show of Kevin Arlin sitting down and grinding out higher category theory, and the result are here: <a href="https://forest.localcharts.org/kda-0003.xml">Double operads</a>.</p>
<p>This was especially cool because in David Jaz’s opening talk he said that he’s wanted a good definition for double operad for years.</p>
<p>I think the lesson from this is that sometimes it’s OK to have a group of 1! Working with other people can spark ideas that it is best to work out individually.</p>
</section>
<section id="combinatorial-type-theory" class="level3">
<h3 data-anchor-id="combinatorial-type-theory">Combinatorial Type Theory</h3>
<p>It was a lucky coincidence that <a href="http://www.cs.nott.ac.uk/~psztxa/">Thorsten Altenkirch</a> happened to have been scheduled to give a talk during the meeting, because I learned about the concept of observational type theory and higher observational type theory from this talk.</p>
<p>Or rather, what really happened is that Thorsten gave a talk, and then later on, David Jaz explained to me why it was really cool.</p>
<p>As far as I can understand, the idea of higher observational type theory is that each type constructor in a type theory (i.e.&nbsp;sigma, pi, etc.) should come along with a definitional equality for what the equality type on that type is equivalent to. For instance, equality for the universe type should be definitionally equal to isomorphism, so univalence becomes definitional instead of propositional.</p>
<p>This seems really cool to me, because it is exactly what I want for combinatorial type theory. Namely, if I write down a combinatorial type, I want to automatically compute a definition for identifications between two elements of that combinatorial type: I want to automatically derive from the definition of a graph a definition of graph isomorphism!</p>
<p>I also want to take this one step further: from the definition of a graph, I want to automatically derive a notion of edit of a graph!</p>
<p>Unfortunately, it seems like there hasn’t been much published on Higher Observational Type Theory yet: it’s being kept somewhat under wraps as it develops.</p>
<p>I think I only really need a fragment of the full power of Higher Observational Type Theory to do what I want, so David Jaz and I discussed some ways of doing HOTT “on the cheap”.</p>
<hr>
<p>But these were not all of the topics discussed at the meeting: these are only the topics that I know about and participated in! I encourage anyone who attended the meeting but didn’t get much of a chance to write during the meeting to write up thoughts while the thoughts are still fresh, and if they feel comfortable, share those thoughts!</p>
</section>
</section>
<section id="lessons-learned" class="level2">
<h2 data-anchor-id="lessons-learned">Lessons learned</h2>
<ol type="1">
<li>When you are organizing something for academics, it’s good to have a half-hour buffer at the beginning and tell people to show up at the beginning of it, so that when they inevitably show up late, the rest of the schedule doesn’t have to be shifted.</li>
<li>Keeping to a schedule is hard. But that’s OK: you don’t necessarily need to keep to a strict schedule in order to get things done!</li>
<li>Small is powerful. Generally the most productive discussions involved 2-3 participants, even if there were more than 3 in the group. That being said, there is a balance between “everyone contributes” and “people who don’t have the same level of experience get to be a fly on the wall and see how people with more experience handle a subject”, and I think that sometimes it is more productive to slow down a discussion and keep everyone following, to accomplish the “learning things” objective. All that being said, I think that it is very hard to do math in a group of &gt;5.</li>
</ol>
<p>Overall, people said that they had a good time at the meeting, so I hope to do this again some time. I honestly think both the small discussion groups and the overall small number of people were both assets, and in fact the small number of days was also somewhat of an asset because it forced people to focus. So I think “scaling this up” doesn’t look like inviting more people for a longer time: I think it looks like inviting different groups of people semi-frequently. Of course, this is only practical when the groups of people happen to be in the same place, but perhaps this is possible if small meetings like this can “piggyback” over other events like conferences. And I encourage other people to organize similar small events and not invite me, but still write up the results on localcharts: I think the ideal number of this kind of event is much larger than would be practical for me to attend!</p>


</section>
</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>events</category>
  <category>systems theory</category>
  <category>category theory</category>
  <guid>https://topos.institute/blog/2024-03-13-a-retrospective-on-the-oxford–topos-meeting/</guid>
  <pubDate>Wed, 13 Mar 2024 00:00:00 GMT</pubDate>
  <media:content url="https://topos.institute/blog/2024-03-13-a-retrospective-on-the-oxford–topos-meeting/localcharts-forest.png" medium="image" type="image/png" height="108" width="144"/>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Sophie Libkind</dc:creator>
  <link>https://topos.institute/blog/2023-12-05-trust-reification/</link>
  <description><![CDATA[ 




<div id="post-content">
<section id="some-vignettes" class="level1">
<h1>Some vignettes</h1>
<p>A few weeks ago, I started at Topos as a postdoc. For the next while I’ll be working with David under a grant called <em>The Structure and Dynamics of Working Language</em>. The mystery at hand is how an abstract specification compiles into a tangible reality. For example, how a high-level recipe for pie turns into a literal pie. One that is created with particular eggs and apples by a particular person who rolls the rolling pin in a particular way. That language really works — in other words, that it is sufficiently useful that its practitioners carry it forward — hinges on a mutual relationship of trust and adaptability between the one who outlines the abstraction (the specifier) and the one who executes it (the reifier). Below is a series of vignettes that elaborate this theme.</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="rosetta_stone.png" class="lightbox" data-gallery="quarto-lightbox-gallery-1" title="A Rosetta Stone for our vignettes"><img src="https://topos.institute/blog/2023-12-05-trust-reification/rosetta_stone.png" class="img-fluid figure-img" alt="A Rosetta Stone for our vignettes"></a></p>
<figcaption>A Rosetta Stone for our vignettes</figcaption>
</figure>
</div>
<section id="peanut-butter-and-jelly-sandwich" class="level2">
<h2 data-anchor-id="peanut-butter-and-jelly-sandwich">Peanut butter and jelly sandwich</h2>
<p>When I was in 5th grade, Mrs.&nbsp;Parker told us to write the instructions for making a peanut butter and jelly sandwich. She then set about constructing sandwiches based on our instructions, but at every step she deliberately misinterpreted any ambiguity in the instructions. For example, if I wrote “put the pieces of bread together”, then she would put them together with the peanut butter and jelly facing outwards. The results were absurd and hilarious for her audience of 10 year olds. This exercise is a classic introduction to computer science for kids. The lesson is that computers — unlike people — don’t have the shared context or capacity to reasonably interpret ambiguous instructions. Hence, giving instructions to a computer requires a much more rigorous level of precision than giving instructions to a fellow human.</p>
</section>
<section id="hiring-a-professional" class="level2">
<h2 data-anchor-id="hiring-a-professional">Hiring a professional</h2>
<p>Now that I’m back at work, my baby is looked after by a wonderful caregive rather than one of her parents. Imagine if I gave our caregiver hyper specific instructions:</p>
<blockquote class="blockquote">
<p>Put on her left sock, then put on her right sock, then read her <em>Dinosaur Dance</em>. When you get to page three, point to the wings on the Pterodactyl and say “those are its wings!”.</p>
</blockquote>
<p>Not only would that be absurd, but it would also make for a terrible working relationship and poor care that doesn’t respond to my kiddo’s actual needs and moods. Instead I share my high-level hopes and expectations (keep my baby safe, be attentive, and support her independence), and then trust our caregiver to translate those into a specific reality (the low-level details of how she interacts with and makes decisions regarding my child). I imagine a similar story with hiring an interior decorator. In the hiring process, I select someone who’s designed spaces that I admire and who shares or at least understands my taste. I give a brief (the requirements, goals, and preferences for the space), but ultimately I trust that they will honor my aesthetics in carrying out the vision. In both of these examples, it’s critical that both trusting the professional and honoring the client is responsive rather than blind. I don’t want to work with someone who blindly honors my preferences even if they think they won’t lead to a healthy baby or usable space. I’ve hired an expert because I want their expertise! Conversely, I don’t want to blindly trust that expertise. Since the expert isn’t me, we may discover missing shared context that we need to flesh out.</p>
</section>
<section id="pass-the-salt" class="level2">
<h2 data-anchor-id="pass-the-salt">Pass the salt</h2>
<p>One of David’s canonical examples of working language goes like this:</p>
<blockquote class="blockquote">
<p>When I say ‘pass the salt’, <img src="https://latex.codecogs.com/png.latex?10%5E%7B20%7D"> atoms including your arm and a salt shaker move through space, resulting in some sodium chloride crystals arriving in my bowl of soup. It is as though language transmits momentum: it makes objects move.</p>
</blockquote>
<p>Language works in the sense of physics but also works in the sense of social expectations. “Pass the salt” doesn’t specify precisely where the salt should land, and it wouldn’t work if an utterance of “pass the salt” usually caused salt to end up in my hair or on my lap. I trust that the reifier (you, and notably not an alien or optimal opponent) compiles the language into physics that aligns with my implicit expectations and our shared context in addition to the letter of the request. The semantics of “pass the salt” is deeply rooted in the existing social and physical dynamics of the world.</p>
</section>
<section id="policies" class="level2">
<h2 data-anchor-id="policies">Policies</h2>
<p>At Topos we have a Flexible Work Policy which says, “The underlying principle of this policy is that we expect a commitment to working together in the office, but are not dogmatic about it and the work comes first.” It gives some specific expectations regarding general working hours and office presence, but it is fairly abstract and outlines the purpose of each expectation rather than establishing hard and fast rules. This level of abstraction is reflected in how I account for my time at Topos. I let my manager and colleagues know when I’ll be out of the office for a vacation, but I don’t punch a timecard. All of this language works at Topos because of the high levels of trust between the policy-makers and the policy-reifiers. Topos trusts that I understand the spirit of the policy. I trust that Topos policies will be responsive to my particular situation. Topos trusts me to honor the mission of the organization and to do the work towards it. I trust that Topos is not self-serving at the cost of my own well-being. In systems where there is less trust both given to the workers and earned by the organization, the accounting systems and the policies are more specific. Runners at an Amazon warehouse are expected to account for their minutes rather than for their days as I do at Topos.</p>
<p><a href="guidelines.gif" class="lightbox" data-gallery="quarto-lightbox-gallery-2"><img src="https://topos.institute/blog/2023-12-05-trust-reification/guidelines.gif" class="img-fluid" style="width:60.0%"></a></p>
</section>
</section>
<section id="reification-and-technology" class="level1">
<h1>Reification and technology</h1>
<p>Trustworthy reification is particularly salient at this moment in technological history, because the lesson that I learned in fifth grade about giving computers specific instructions is no longer true. I can give ChatGPT a vague prompt like “Write an algorithm that sorts a list of names by last names” and it produces specific code that respects not only the intentions of my request but also the context of modern day programming. It uses meaningful variable names, adds comments, and flags edge cases.</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="chatgpt.png" class="lightbox" data-gallery="quarto-lightbox-gallery-3" title="The output of a ChatCPT prompt."><img src="https://topos.institute/blog/2023-12-05-trust-reification/chatgpt.png" class="img-fluid figure-img" alt="The output of a ChatCPT prompt."></a></p>
<figcaption>The output of a ChatCPT prompt.</figcaption>
</figure>
</div>
<p>As technology speeds up the reification process — in large part because it can reify more and more abstract instructions — we (the specifiers) want to be in a virtuous cycle of trust with it (the reifier). These vignettes illustate how developing and maintaining trust relies on:</p>
<ul>
<li>Communication and responsiveness between specifier and reifier.</li>
<li>The nature and dynamics of the shared context.</li>
</ul>
</section>
<section id="mathematical-breadcrumbs" class="level1">
<h1>Mathematical breadcrumbs</h1>
<p>As a first-pass at mathematicizing the relationship between high-level specifications with low-level realities, my instinct is to construct a coarse-graining map from low-level to high-level representing which realities implement which specifications. For example, a map that sends collections of pixels to the labels “cat”, “dog”, “sandwich”, etc. From the perspective of this formalism, all low-level implementations of a given high-level specification are equivalent. However, the vignettes show precisely that this is not the case in how we humans expect our instructions to be reified. I wasn’t happy with how Mrs.&nbsp;Parker put the pieces of bread together, even though it strictly adhered to my instructions. And not every interpretation of “keep my baby safe and happy” is how I want my baby to be cared for. These mathematics of coarse-graining fails to capture the shared context and trust between the specifier and the reifier.</p>
<p>In contrast, I am excited about how frameworks like <a href="https://arxiv.org/abs/2205.03906">dynamic operads</a> foreground the two way communication between high-level and low-level actors. As David and I delve more into the structure and dynamics of working language, I plan to keep my eye on the idea of shared context, how it directs the way that language is compiled, how it forms the language itself, and how it is the medium upon which trust is built.</p>
<p><strong>Acknowledgements</strong> This blog post is the product of a conversation with David Spivak about why I care about working language. Kris Brown helped me see the role of shared context and the fourth vignette is due to Brendan Fong. Thanks!</p>


</section>
</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>language</category>
  <guid>https://topos.institute/blog/2023-12-05-trust-reification/</guid>
  <pubDate>Tue, 05 Dec 2023 00:00:00 GMT</pubDate>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Bartłomiej Skowron</dc:creator>
  <dc:creator>David Spivak</dc:creator>
  <dc:creator>David Corfield</dc:creator>
  <link>https://topos.institute/blog/2023-07-11-math-and-goodness/</link>
  <description><![CDATA[ 




<div id="post-content">
<blockquote class="blockquote">
<p><em>This post is a dialogue and consists of two parts. In the</em> <em>first part, Bartłomiej Skowron and David Spivak consider what</em> <em>mathematics and ethics have in common. Though some will think these</em> <em>notions have nothing in common, these authors share the intuition that</em> <em>the opposite is true. In particular, they propose that sense-making is</em> <em>always good and that mathematics is significantly useful in modern</em> <em>sense-making. In the second part, David Corfield responds, pointing out</em> <em>the importance of the good life, argues for strengthening cognitive</em> <em>science research on well-being by scaffolding it with mathematics and</em> <em>category theory in particular.</em></p>
</blockquote>
<section id="part-i" class="level1">
<h1>Part I</h1>
<p><strong>Bartłomiej Skowron, David Spivak</strong></p>
<p>According to legend, one of the ancient philosophers gave a public lecture entitled “<a href="https://www.jstor.org/stable/4182081">On the Good</a>”. This topic attracted many listeners, who probably wanted to listen to considerations about how to live a healthy life, earn money successfully, and gain honorable positions. Nevertheless, this philosopher surprised everyone: instead of talking about these essential goods, he spoke of mathematics, numbers, shapes, etc. Few understood, and many lost interest and left. Reportedly, this lecture concluded that goodness is <em>one</em>, further obscuring the whole matter rather than illuminating it. Why was this philosopher talking about mathematical structures when he was supposed to analyze goodness? And why was the story of this lecture recalled in the 20th century by Alfred North Whitehead and later <a href="https://link.springer.com/chapter/10.1007/978-1-4613-9435-8_13">Felix Browder &amp; Saunders Mac Lane</a>? Because doing mathematics can be seen as a path to goodness, and mathematical structures as structures of goodness itself. Strange, isn’t it? Whitehead wrote: “The lecturer was competent - he was Plato.” It is from Plato that the idea that goodness and mathematics have much in common originates.</p>
<p>Aristotle, who disagreed with Plato, pointed out in his major ethical work <em>The Nicomachean Ethics</em> that the good is understood in different ways. Moreover, even the same person can understand the good differently. When a person is COVID sick, good seems to be health; when she is tired, good seems to be rest; when she acts cowardly, good seems to be courage; when she does not understand how to compose morphisms in the category Poly, good then appears to be understanding the simpler cases of natural transformations. But what should we do when the answer to the question of goodness seems so case-dependent? Is it possible to say something that isn’t empty but instead meaningful and relevant about the good? Following Plato’s intuition, we claim that mathematics may help us do this.</p>
<p>Let us assume, laboriously and for a moment, that goodness is not only what, for example, <em>seems</em> good to you, your friend, or me. Let the good be an <em>idea</em>, a term which we don’t intend in the everyday casual sense, but instead along the lines of Plato’s <em>form</em> or more precisely “idea” in the sense of <a href="https://www.peterlang.com/document/1067462">Roman Ingarden</a>: an idea is something that has some <em>content</em>, some rich ontological structure. For example <em>linearity</em> is an idea—a mathematical one—captured by the ontological structure of the category of vector spaces. In the case of the good, this content is whatever structure allows us to classify certain objects (here, you can replace <em>object</em> with <em>action</em>, <em>person</em>, or <em>value</em>; the object need only be an entity with a desired form or manner of existence) as being instances of this idea. You can think of this content as a specific shape, and of instances of the idea, say goodness, as objects in which this shape is present. To put it yet another way, it is a certain pattern of patterns, and as such can be realized in many objects. We want to know what this pattern called goodness—which remains the same, regardless of how many (if any) instances it has—is.</p>
<p>Nevertheless, caution is needed here: ideas are characterized first of all by the fact that they are general entities (sometimes they are called universals, sometimes they are called abstracts), which means that this shape or pattern has many empty slots, which can be filled in various ways. For example, individual actual real objects do not have such gaps. The act of filling in these gaps is called concretizing qualities, but discussing this would take us into too metaphysical an area. So let’s go back to the basic idea: make a working assumption that when we want to talk about the good, we will try to speak in essence about the idea of the good, or more precisely, about the content of the good as an idea. In this view, the mathematics of goodness becomes the mathematical study of the content of goodness: understanding these “shapes,” patterns and their dynamics. For the purposes of this blog post, let’s assume you can make some sense of what’s being said here.</p>
<p>In fact, <a href="../../blog/2022-06-03-an-account-of-sense-making/">sense-making</a> is central to our story of how goodness and mathematics are related. To see this, let’s return to everyday life on Earth. When I need to shop, I need to find the right store where I can find the products in question. I can get to that store guided by Google Maps or just glance at the map and use my sense of direction, as long as I have that sense developed. When I stroll through the park, I can distinguish the balmy, woody notes of lavender or the earthy notes of iris, as long as I have my sense of smell sharpened. When I’m feeling lonely, I can use social sense to make some friendships that seem to respond to that loneliness. If I’m doing philosophy, and I don’t want to be naïve, I can use my sense of irony, like Socrates, and thus for an exercise, distance myself from my own philosophical beliefs. When I make a decision, dividing some goods among people, I can do it fairly if I have a sense of justice. In general, when I make a decision of a moral nature, something attracts me approvingly to good choices, and something disapprovingly outrages me when I imagine bad decisions—the moral sense is responsible for this attraction or repulsion. This is probably the case with every sense: good compositions attract us, and bad ones repel us. Fitting things together is, in fact, sense-making. In general, it seems that the art of living is the art of making sense and avoiding nonsense.</p>
<p>Sense-making is multi-sensory, dynamic, and complex. Our senses of things and, above all, the corresponding qualities must fit together in order to for us to act proficiently. Otherwise, there will remain a lack of meaning and an inability to navigate. In order to divide goods equitably, I must be able to both divide and also (sometimes) use my sense of sight to see the objects I am dividing, as well as my moral sense to see the merits of those subjects, which would be the basis for equitable distribution. When experienced florists arrange a bouquet, they compose the qualities of the different senses: both the color of the flowers and their scent must be well composed. These qualities must fit together so that the bouquet is good and so that we can say that this florist is good at being a florist. A good florist is a sense-composer. Certain fragrance notes may not fit together, and such combinations repel us because they don’t make sense together, whereas certain combinations are almost perfect and attract many sense-makers. Sometimes these combinations even intoxicate and please, such as that of orange, grapefruit, nutmeg, and vanilla. One might ask: on what is this sense-making founded? Behind each sense are the qualities of that sense. For example, to the sense of sight, there are qualities such as redness, linearity, and blurriness; to the sense of smell, there are qualities such as the notes of nutmeg and vanilla. The sense of irony includes qualities such as perversity. To the metaphysical sense, there are qualities such as sublimity. And to the moral sense there are moral qualities, such as the quality of justice.</p>
<p>How does all this relate to mathematics? Simple. Certain sense qualities fit together, making sense, and others don’t; they don’t make sense because they don’t fit together well. Sense-makers do the work of finding fitting compositions daily: we compose senses, and as long as we do it well, we make sense of the world. Then, in interaction with other sense-makers, we exchange our notions of sense, influencing and changing each other: sometimes we all match each other perfectly, and sometimes the group can’t make sense of each others’ points of view at all. Sometimes we all find the same qualities, and in this case we hook up and compose our senses together, making a synergistic group sense that allows the group to navigate the situation better than any member alone could have. This is how consciousness—not necessarily experience, but coherent sense—is created with many multi-level interactions, both individual and collective. Individual senses can come together to create a new higher-level sense, and this process is formed through an active and creative process. This way senses combine may be governed by multi-level laws of composition or fitting together. It seems that no one currently knows the general laws of sense composition: they still need to be discovered. But perhaps these laws have something to do with <a href="../../blog/2022-06-03-an-account-of-sense-making/">proper accounting</a>, the culmination of which is mathematical and which we can try to approach with category theory.</p>
<p>Plato had the intuition that goodness is oneness and order. To separate good from evil is to separate order, the cosmos, from disorder. We understand order in this sense as meaning structure: that which can be mathematically grasped. This means that our notion of the good could in fact be mathematized by articulating multiscale compositional dynamics that occur within the network of sense-makers immersed in a world of sensory qualities. This way of mathematizating goodness—or at least some nearby notion—means formulating the dynamic process by which sense is made.</p>
<p>Plato assumed that ideas are static—timeless—and that they exist separately from material things. The considerations here indicate that all sense-making is in fact dynamic. Nevertheless, this dynamism still has a mathematical pattern, the stability of which does not consist in being static, as Plato and Ingarden think, but consists in the dynamic structure of how senses are sensibly composed. Thus, although we started from Plato’s intuition, we conclude that Plato’s theory of ideas is like Wittgenstein’s ladder that, having climbed, must be discarded.</p>
<p>You might assume that the good cannot be analyzed because it is simple and thus inaccessible to cognition. Or you might claim that it is a heterogeneous conglomerate without any internal uniformity or structure and that one can only randomly hit parts of it. We do not claim either of these. The shape of the good is cognitively graspable: it can be learned, and it is essentially compositional in nature. To know the good is to understand the laws of composition. All this, of course, does not mean that it is easy to see the good. Nevertheless, it does mean that it can be done in some way, an expression of epistemological, axiological optimism. However, to reiterate, it is not easy and probably will never be complete. The way leads through categorifying the content of goodness, a project which we propose to begin. Does it make sense?</p>
</section>
<section id="part-ii" class="level1">
<h1>Part II</h1>
<p><strong>David Corfield</strong></p>
<p>Thank you both for your reflections on how we might use mathematics to address goodness. There are some helpful ideas contained in your post to get us started. In particular I agree that something like sense-making plays an important role in what we take to be good.</p>
<p>As for my own starting point to approach the question, I’d like to consider first what might appear to be the hardest part of the problem. When Ancient Greek philosophy considers the good, it is always as part of the particular question of what it is for a human to lead a good life. This is a central concern of each of Socrates, Plato and Aristotle. Now of course the good life has you enjoy good things. It can have you appreciating via the senses, say, a florist’s bouquet, the florist having done a good job in composing the scents and colors of the bunch. It can have you appreciating a good coffee at a convenient store. It can have you communing with good friends for an evening. But the good life involves so much more than these. Rather than a series of satisfactory experiences, it encompasses a certain way of living. Naturally it involves a life that makes sense of features of its environment, that allows for a good fit between an individual and their niches. But the good life is a life rich with meaning, and a life full of meaning-making.</p>
<p>‘Making sense of’ is a phrase generally applied to an existing phenomenon, making sense of the scene before you, making sense of what someone is saying to you, making sense of a mathematical argument. Making <em>meaning</em>, on the other hand, is not making meaning <em>of</em> anything, it’s a creative process. It may be the setting up of a project, like we’re doing here. It may be planning to lead a life together with someone and enacting this plan. Leading a good life is living a life full of meaning, following a vocation, sustaining a flourishing community, creating meaning for yourself and others.</p>
<p>Well that only seems to have complicated the problem. Perhaps we might have hoped to be able to say something about how goodness applies to entities of particular types - a good apple, a good book, a good sheepdog, a good soccer match. Now we have to deal with the thorny issue of characterizing the good life, as though there could be something mathematically tractable about this.</p>
<p>Do we see existing work out there to learn from? Well, of course plenty has been said about the good life. Indeed, the wisdom traditions of the world, from the East to the West, have a great deal to say. ‘Know thyself’ advises Socrates. ‘Inculcate the virtues’, says Aristotle. And Eastern traditions of Buddhism and Daoism have their own say: Follow the Noble Eightfold Path, Follow the Way.</p>
<p>So plenty of verbal descriptions of what the good life looks like, and discussion of the means to transform oneself to pursue it. Can we really hope that category theory can have something to say on the subject? Well, there’s plenty of work at the moment from the interface of cognitive science and wisdom studies, which looks to see whether the former can make sense of the Western and Eastern traditions. Here we see a strong convergence at the moment with the following three approaches:</p>
<ul>
<li>Predictive processing/active inference on well-being – Mark Miller, Brett Andersen;</li>
<li>Relevance realization and the meaning crisis – John Vervaeke;</li>
<li>Hemispherical balance, the left serving the right – Iain McGilchrist.</li>
</ul>
<p>Notably these people are talking to each other (e.g., <a href="https://www.youtube.com/watch?v=XzT4tcC-aag">here</a> and <a href="https://www.youtube.com/watch?v=x4bO-3SJp1s">here</a>). They’re even publishing together, see, e.g., the <a href="https://doi.org/10.1007/s11097-022-09850-6">joint work</a> of Vervaeke with Miller and Andersen. The unanimous conclusion is that we face a problem today and that we need to change the way we live in the world, in particular we need to develop our meta-cognitive capacities to equip us with the ability to apportion wisely appropriate care to our meaning-making projects, and the ability to seek the right challenges to push our boundaries.</p>
<p>These approaches also allow a great deal to be said on what goes wrong in a life. From the predictive processing side, the optimisers that we are get stuck in feedback loops, the loop of addiction, of depression, and so on. Or we rely too heavily on left-hemispherical thinking and become too instrumental in our dealings with the world. Departures from the right balance are involved in mental disorders.</p>
<p>Now perhaps here we have a window of opportunity, since in particular the predictive processing approach commences from an explicitly mathematical point of view in its characterization of the “Bayesian Brain”. To do this there needs to be mathematics that deals with recursive systems. One where layers of the hierarchy adapt according to bottom-up and top-down processing. We need to have our higher-level cognitive systems engage in self-monitoring, modeling the way we model the world to predict if that modeling is working or needs correction. We need the cognitive apparatus to apply itself to itself.</p>
<p>Take Vervaeke’s <a href="https://www.meaningcrisis.co/ep-49-awakening-from-the-meaning-crisis-corbin-and-jung/">description</a> of Jung’s idea of the Self:</p>
<blockquote class="blockquote">
<p>“What’s the self? Well, it’s kind of the archetype of the archetypes. It’s like Plato’s notion of the good, which is the form for how to be a form. The eidos of the eidos. It is the virtual engine regulating the self organization of the psyche as a whole. It is the principle—the self is the principle of autopoiesis itself. It’s the ultimate virtual engine that constellates all the other virtual engines so that the psyche can continue its process of autopoietic self-organization. Remember when a system is self-organizing its function and its development are completely merged. It develops by functions and it functions by developing.”</p>
</blockquote>
<p>Readers of this blog will know that we have at least one claim for a fruitful place where ACT meets such a complex system, and that is in the work of one of us, <a href="https://arxiv.org/abs/2005.01894">David Spivak</a>. Another hint that we could take this further is that another Topos Institute member, <a href="https://arxiv.org/abs/2212.12538">Toby St Clere Smithe</a>, is looking to use ACT precisely to capture the workings of the Bayesian Brain. Further relevant work on <a href="https://arxiv.org/abs/2105.06332">category-theoretic cybernetics</a> may also help us address the question of the well-being of a complex system. My proposal then is that we take up such work and see if ACT can help us articulate what the cognitive scientists are telling us about well-being.</p>


</section>
</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>ethics</category>
  <category>category theory</category>
  <category>philosophy</category>
  <guid>https://topos.institute/blog/2023-07-11-math-and-goodness/</guid>
  <pubDate>Tue, 11 Jul 2023 00:00:00 GMT</pubDate>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>davidad </dc:creator>
  <dc:creator>Owen Lynch</dc:creator>
  <link>https://topos.institute/blog/2023-06-15-compositional-world-modeling/</link>
  <description><![CDATA[ 




<div id="post-content" class="page-columns page-full">
<section id="the-problem" class="level1">
<h1>The problem</h1>
<p>In scientific applications (biology, ecology, epidemiology, chemistry, physics, economics, etc.) and related engineering domains, full modeling of a complex system frequently involves the following considerations:</p>
<ol type="1">
<li><p><strong>Multi-disciplinarity</strong>. Problems may show up in a certain domain that cannot be solved with the customary mathematical tools and software frameworks of that domain, and the mathematical, scientific, technical, and cultural assumptions of another scientific domain may be illegible from the perspective of the original domain. For instance, modeling hydraulic systems often involves both rigid body physics and fluid physics.</p></li>
<li><p><strong>Openness</strong>. One is not always lucky enough that one part of the world can be fully isolated from the rest. Rather, systems tend to have interfaces or boundaries, on which (and beyond which) one’s model does not make endogenous predictions—but on which its predictions must depend. For instance, this is typically the case of electronic circuits; they have terminals that are intended to be connected to other systems.</p></li>
<li><p><strong>Continuity in time</strong>. The real world is typically not parceled into discrete time steps. While a sufficiently small time step can generally be chosen, this comes at a great computational cost. We would prefer to be able to reason flexibly about timescales, with a guarantee that for any nonnegative rationals <img src="https://latex.codecogs.com/png.latex?s,t%20%5Cin%20%5Cmathbb%7BQ%7D%5E%7B%5Cgeq%200%7D">, the forward propagation <img src="https://latex.codecogs.com/png.latex?T_%7Bs+t%7D"> is compatible with <img src="https://latex.codecogs.com/png.latex?T_s%20;%20T_t"> in the sense that, for any initial knowledge <img src="https://latex.codecogs.com/png.latex?K">, <img src="https://latex.codecogs.com/png.latex?T_s%20T_t%20K%20%5CRightarrow%20T_%7Bs+t%7D%20K">.</p></li>
<li><p><strong>Continuity in space</strong>. We often are interested in systems whose states have continuous extent. i.e.&nbsp;electric fields, fluids, deformable bodies, etc., meaning that the state space is naturally an infinite-dimensional function space.</p></li>
<li><p><strong>Stochasticity</strong>. We often observe systems at a scale where many degrees of freedom are not visible, and even when they are visible, it is frequently advantageous to ignore them. However, these hidden degrees of freedom affect the larger scale behavior in the form of <em>noise</em>, and cause the system to behave stochastically rather than deterministically. The weather is one good example of a system which has some amount of regularity, but also a good deal of noise.</p></li>
<li><p><strong>Nondeterminism</strong>. Even stochasticity makes strong claims about uncertain behavior, by asserting precisely known probabilities of any observable. In many cases, we only know that probabilities lie within a certain range, or we do not know them at all. This motivates “imprecise probability,” which we believe is best understood as a monad which naturally generalizes both the probability monad and the nondeterminism monad (such as the monad of convex subsets of probability-distribution space). The behavior of your opponent in chess is nondeterministic; it doesn’t make sense to have a full probability model for the state of your opponent’s brain.</p></li>
<li><p><strong>Partiality</strong>. Probabilistic programs may in general be nonterminating, or they may reject inconsistent inputs; these both motivate a general notion of partiality, or the ability of processes to result in <img src="https://latex.codecogs.com/png.latex?%5Cbot"> (a state representing that something went wrong). For instance, you might want to sample from the inside of a disk by sampling from a square, and throwing away points not inside the disk.</p></li>
<li><p><strong>Hybridness</strong>. Complex systems often involve interactions between continuous and discrete (or discretely modeled) components. Discrete components may transition on a fixed timescale, or at times generated by a point process that depends on the outputs of continuous components. Continuous components may depend on the outputs of discrete components. A classical example of this is a hybrid digital-analogue circuit.</p></li>
</ol>
<p>Many of these considerations greatly enlarge the space of solutions: the space of time-varying probabilitity distributions over piecewise-continuous vector fields on a manifold is incredibly “large”, in the sense that <em>multiple</em> levels of discretization are required to convert this into something that can be represented on a computer.</p>
<p>Analytic or numerical solutions to problems which are “that large” are, in general, intractable. However, a logically-specified “full model” of the situation is still useful, as a ground truth that more tractable approximations can be compared to, or that behavioral guarantees can be extracted from.</p>
<p>It is for this reason that we would like a compositional, formal framework for such systems, which could be used to design and analyze complex models. We expect that category theory is a fruitful method for finding such a framework.</p>
<p>Solving such “large systems” is a problem that has been attacked by brilliant mathematical physicists for generations; the usefulness of our approach is not necessarily predicated on the idea that a compositional framework would significantly advance the state of the art with regard to concrete solutions or theorems for such systems.</p>
<p>However, having a formal framework would organize the field in such a way as to make it easier to produce rigorously verified and interpretable computer-aided analyses. Moreover, organizing software implementation makes it possible to take existing techniques much farther in less time and code.</p>
<p>In the next section, we discuss applications of such a framework, and then in the last section we will discuss some concrete criteria we have for such a framework, as well as tentative approaches we are pursuing.</p>
<p>Finally, we should say that by “formal framework”, we do not mean that all “large systems” will necessarily be captured by a single definition. Rather, we expect that we will capture certain classes of large systems with different definitions, but that we will be able to relate and translate these different definitions formally. The reason this is desirable is that we want to be able to take advantage of the special properties of different classes of system.</p>
<p>For instance, we might have some definition for systems whose solutions are smooth, and be able to prove some theorems that wouldn’t hold for systems with non-smooth solutions. However, we should be able to also translate a smooth system into a more general class of systems, which we might want to do in order to compose it with a non-smooth system or stochastic system.</p>
<p>Some examples of types of systems that we would possibly like to fit within our framework are the following.</p>
<ol type="1">
<li>Differential equations
<ul>
<li>ordinary</li>
<li>partial</li>
<li>stochastic</li>
<li>random</li>
<li>jump-diffusion</li>
</ul></li>
<li>Markov processes
<ul>
<li>discrete-time Markov chains</li>
<li>continuous-time Markov chains</li>
<li>Markov decision processes</li>
<li>open games</li>
<li>(open) mean-field games</li>
<li>Markov automata</li>
</ul></li>
<li>“Hybrid systems”</li>
<li>Probabilistic models
<ul>
<li>probabilistic graphical models / Bayesian networks / causalgraphs</li>
<li>corecursive programs in a functional probabilistic programming language (including, notably, large language models based on Transformer neural networks)</li>
<li>generative probabilistic logic programs</li>
<li>score-based generative models (including, notably, diffusion models based on U-Net neural networks)</li>
</ul></li>
</ol>
<p>Classically, these are all different paradigms of system. However, redefining these systems categorically will mean that it will be easier to relate them to each other. This doesn’t necessarily mean that there will be categories for each type of system; we don’t yet know the precise type of categorical gadget we will need for modeling. It could be monoidal categories, it could be pseudo lax virtual triple categories (whatever that is). But morally speaking, there should be some sort of functorial interpretation of certain types of systems into others. And additionally, we should be able to “bottom out” out in a behavioral interpretation in which systems are described merely as joint epistemic states regarding their observables, which can be composed by intersection.</p>
</section>
<section id="applications" class="level1">
<h1>Applications</h1>
<p>Decision-making in complex environments is often critically dependent on coherent predictions, informed by observational data (<img src="https://latex.codecogs.com/png.latex?o%20:%20O">), of how actions (<img src="https://latex.codecogs.com/png.latex?a%20:%20A">) affect the probability of queries <img src="https://latex.codecogs.com/png.latex?Q"> (each Boolean-valued) pertaining to a future trajectory <img src="https://latex.codecogs.com/png.latex?x(t)"> <img src="https://latex.codecogs.com/png.latex?%5Cmathrm%7BPr%7D%5C!%5Cleft(%20%5Cmathrm%7BQ%7D(x(t))%20%5C;%5Cmiddle%7C%5C;%20o%20:%20O,%20%5Ctext%7Bdo%20%7D%20a%20:%20A%20%5Cright)"></p>
<p>Some examples include:</p>
<ol type="1">
<li>Deciding what policy interventions to perform during a pandemic:
<ul>
<li>observations include recorded cases and outcomes</li>
<li>the complex environment includes asymptomatic cases, the details of mobility, the virulence of different strains, etc.</li>
</ul></li>
<li>Deciding when to switch on and off power-generating stations, and when to charge or discharge grid-scale energy storage:
<ul>
<li>observations include voltages (magnitude and phase) and currents</li>
<li>the complex environment includes transmission lines, predictable and unpredictable sources of power demand (at various price levels), weather (affecting renewable capacity), etc.</li>
</ul></li>
<li>Deciding what incentives to provide for reducing the emissions of climate-affecting gases:
<ul>
<li>observations include weather data, sea levels, sea temperatures, energy supply levels, recorded material flows, etc.</li>
<li>the complex environment includes atmospheric advection, cloud processes and albedo, radiative forcing, oceanic eddies, etc.</li>
</ul></li>
<li>Planning a project with many subgoals (and sub-subgoals, etc.) for which various alternative approaches may be viable:
<ul>
<li>observations include status reports, estimates, milestone completion events, etc.</li>
<li>the complex environment includes the changing availability of human resources with varying levels of skills and experience, technical risks and uncertain feasibilities, potential synergies between efforts, etc.</li>
</ul></li>
<li>Deciding which combination of treatments to suggest to a cancer patient:
<ul>
<li>observations include biomarkers, imaging scans, expert assessments, DNA sequencing, proteomics, metabolomics, etc.</li>
<li>the complex environment includes the immune system, cellular regulatory networks, the diffusion of blood in and around tumors, etc.</li>
</ul></li>
</ol>
<p>In many of these cases, we need to integrate information from quantitative models across multiple scientific ontologies and scales (in addition to expert opinions where scientific data is sparse) in a computationally tractable way. In the hopes that machine learning may be increasingly able to provide computationally feasible answers to questions that are sufficiently well-defined, our focus here is on being able to make such boundary-transcending questions well-defined, by defining a sufficiently flexible and rigorous meta-ontology that can encompass all necessary ontologies and adapters between them.</p>
<p>The point is not to figure out how to do traditional-style modeling and simulation of complex systems (although that might be a component), the point is for any model (even when highly complex and multidisciplinary) to have a well-defined semantics to which efficient approximation methods can be rigorously compared. This is as opposed to the common state in current practice in which the best known models are only defined by their custom and informal implementations (especially when machine learning is involved), making it very difficult to reason about the model’s behavior on inputs with which it has never specifically been run. If your model is more than 100,000 lines of mathematically ill-specified Python code, it is hard to have confidence in what it is saying, especially when machine learning is involved in any capacity.</p>
</section>
<section id="litmus-tests" class="level1 page-columns page-full">
<h1>Litmus tests</h1>
<p>In the first section, we expressed our problem somewhat vaguely. To a certain extent, this is the level of detail to which we know our problem, in that our technical approaches may fail, and we would only have these goals to fall back on. However, we have some litmus tests that would indicate we are on the path to a “good” solution.</p>
<p>First of all, David Jaz Myers has constructed a “framework of dynamical systems frameworks” <span class="citation" data-cites="myers_categorical_2022">(Myers 2022)</span>, which he calls “dynamical systems doctrines”, and we expect that a good solution to our problem should fit within one or more dynamical systems doctrines. The formal definition is technical, but informally, a dynamical system doctrine answers the questions</p>
<div class="no-row-height column-margin column-container"><div id="ref-myers_categorical_2022" class="csl-entry">
Myers, David Jaz. 2022. <em>Categorical Systems Theory</em>. <a href="https://github.com/DavidJaz/DynamicalSystemsBook">https://github.com/DavidJaz/DynamicalSystemsBook</a>.
</div></div><ol type="1">
<li>What is a system?</li>
<li>What does it mean to compare systems?</li>
<li>What does it mean to compose systems?</li>
</ol>
<p>When these questions have been answered in a suitably precise manner, the systems under study form a category theoretic structure that has a variety of nice properties, and can be used to organize the study of the systems at hand.</p>
<p>More technically, for example, a jump-drift-diffusion equation models a system with state space <img src="https://latex.codecogs.com/png.latex?%5Cmathbb%7BR%7D%5En"> whose behavior is driven by a combination of</p>
<ol type="1">
<li>Jumps: discontinuous movement from one point to another point</li>
<li>Drift: continuous movement following a vector field</li>
<li>Diffusion: continuous movement that is Brownian in character</li>
</ol>
<p><span class="citation" data-cites="applebaum_semigroups_2019">(Applebaum 2019)</span></p>
<div class="no-row-height column-margin column-container"></div><p>This covers a large class of “stochastic ODEs”. We now make the following conjecture.</p>
<div class="callout callout-style-simple callout-none no-icon callout-titled">
<div class="callout-header d-flex align-content-center">
<div class="callout-icon-container">
<i class="callout-icon no-icon"></i>
</div>
<div class="callout-title-container flex-fill">
<span class="screen-reader-only">None</span>Conjecture
</div>
</div>
<div class="callout-body-container callout-body">
<p>There is a dynamical systems doctrine in which open jump-drift-diffusion equations can be modeled and composed in an analogous way to the dynamical system doctrine for non-stochastic ODEs, where the parameters of one model are determined by the state variables of another.</p>
</div>
</div>
<p>Secondly, models are useful to us in as much as they help us make decisions about the world. And in order to choose between different actions, we need to be able to evaluate <em>counterfactual interventions</em>. The philosophy of causality is complex, but from a mathematical perspective, we hope to model and subsume popular framings for causality such as Pearl’s do-calculus <span class="citation" data-cites="Pearl2009">(Pearl 2009)</span> (see also <span class="citation" data-cites="sepDoCalculus">(Hitchcock 2018)</span>, and likely building on the categorical notion of d-separation developed by <span class="citation" data-cites="Fritz2023">(Fritz and Klingler 2023)</span>).</p>
<div class="no-row-height column-margin column-container"><div id="ref-Pearl2009" class="csl-entry">
Pearl, Judea. 2009. <em>Causality</em>. Cambridge University Press. <a href="https://doi.org/10.1017/cbo9780511803161">https://doi.org/10.1017/cbo9780511803161</a>.
</div><div id="ref-sepDoCalculus" class="csl-entry">
Hitchcock, Christopher. 2018. <span>“<span>C</span>ausal <span>M</span>odels <span>S</span>upplement 2. <span>T</span>he Do-Calculus (<span>S</span>tanford <span>E</span>ncyclopedia of <span>P</span>hilosophy).”</span> Stanford Encyclopedia of Philosophy. <a href="https://plato.stanford.edu/entries/causal-models/do-calculus.html" class="uri">https://plato.stanford.edu/entries/causal-models/do-calculus.html</a>.
</div><div id="ref-Fritz2023" class="csl-entry">
Fritz, Tobias, and Andreas Klingler. 2023. <span>“The d-Separation Criterion in Categorical Probability.”</span> <em>Journal of Machine Learning Research</em> 24 (46): 1–49. <a href="http://jmlr.org/papers/v24/22-0916.html">http://jmlr.org/papers/v24/22-0916.html</a>.
</div></div><div class="callout callout-style-simple callout-none no-icon callout-titled">
<div class="callout-header d-flex align-content-center">
<div class="callout-icon-container">
<i class="callout-icon no-icon"></i>
</div>
<div class="callout-title-container flex-fill">
<span class="screen-reader-only">None</span>Conjecture
</div>
</div>
<div class="callout-body-container callout-body">
<p>We can make a coherent model for actions and observations within a dynamical-systems doctrine, that subsumes the kind of Bayes nets used in Pearl’s work on causality.</p>
</div>
</div>
<p>This conjecture seems plausible because dynamical systems doctrines have a good notion of control, and “observations and actions” are a central part of control. Moreover, dynamical systems are “intensional” systems: because their state spaces are decomposable, they have a native notion of interventional query (one can factor the state space, force one factor to a certain value, and resume time-evolution).</p>
<div class="callout callout-style-simple callout-none no-icon callout-titled">
<div class="callout-header d-flex align-content-center">
<div class="callout-icon-container">
<i class="callout-icon no-icon"></i>
</div>
<div class="callout-title-container flex-fill">
<span class="screen-reader-only">None</span>Conjecture
</div>
</div>
<div class="callout-body-container callout-body">
<p>There is a single dynamical systems doctrine which provides semantics for both of the above and enables some nontrivial forms of composition between them.</p>
</div>
</div>
<p>This is more speculative, but it is the thrust of the direction we want to aim for: constructing a “big tent” in which all such models can be “compatible”.</p>
</section>
<section id="other-related-work" class="level1 page-columns page-full">
<h1>Other related work</h1>
<section id="temporal-type-theory" class="level2 page-columns page-full">
<h2 data-anchor-id="temporal-type-theory">Temporal type theory</h2>
<p><span class="citation" data-cites="Schultz2019">(Schultz and Spivak 2019)</span> also aims to be a “big tent” for both discrete and continuous dynamical systems. It can also model nondeterminism. But it cannot model stochastic processes. Also, because temporal type theory takes a natively “behavioral” or “relational” approach to system composition, specifying that a system has distinct input ports and output ports requires quite a few symbols, including a condition called “inertiality” that is not entirely intuitive. (That said, we consider it possible that a behavioral approach may be best.)</p>
<div class="no-row-height column-margin column-container"><div id="ref-Schultz2019" class="csl-entry">
Schultz, P., and D. I. Spivak. 2019. <em>Temporal Type Theory: A Topos-Theoretic Approach to Systems and Behavior</em>. Progress in Computer Science and Applied Logic. Springer International Publishing. <a href="https://books.google.co.uk/books?id=bs6FDwAAQBAJ">https://books.google.co.uk/books?id=bs6FDwAAQBAJ</a>.
</div></div></section>
<section id="decorated-cospans" class="level2 page-columns page-full">
<h2 data-anchor-id="decorated-cospans">Decorated cospans</h2>
<p><span class="citation" data-cites="Fong2015 Patterson2023">(Fong 2015; Patterson 2023)</span> is a generic framework that can be specialized to many different types of “open systems”, but once specialized, does not provide a “big tent” in the sense that there isn’t a systematic way to compose morphisms from different decorated-cospan categories. That being said, decorated cospans often provide good “syntax” for describing systems, and with the appropriate semantic functors should be able to be integrated into whatever systems theory we end up with.</p>
<div class="no-row-height column-margin column-container"><div id="ref-Fong2015" class="csl-entry">
Fong, Brendan. 2015. <span>“Decorated Cospans.”</span>
</div><div id="ref-Patterson2023" class="csl-entry">
Patterson, Evan. 2023. <span>“Structured and Decorated Cospans from the Viewpoint of Double Category Theory.”</span>
</div></div></section>
<section id="mathbfspanmathbfgraph" class="level2 page-columns page-full">
<h2 data-anchor-id="mathbfspanmathbfgraph"><img src="https://latex.codecogs.com/png.latex?%5Cmathbf%7BSpan%7D(%5Cmathbf%7BGraph%7D)"></h2>
<p><span class="citation" data-cites="Katis1997 DiLavore2023">(Katis, Sabadini, and Walters 1997; Lavore et al. 2023)</span> is (arguably) a canonical “big tent” for discrete-time systems, but does not model stochastic processes or differential equations.</p>
<div class="no-row-height column-margin column-container"><div id="ref-Katis1997" class="csl-entry">
Katis, Piergiulio, N. Sabadini, and R. F. C. Walters. 1997. <span>“Span(graph): A Categorical Algebra of Transition Systems.”</span> In <em>Algebraic Methodology and Software Technology</em>, 307–21. Springer. <a href="https://doi.org/10.1007/bfb0000479">https://doi.org/10.1007/bfb0000479</a>.
</div><div id="ref-DiLavore2023" class="csl-entry">
Lavore, Elena Di, Alessandro Gianola, Mario Román, Nicoletta Sabadini, and Paweł Sobociński. 2023. <span>“Span(graph): A Canonical Feedback Algebra of Open Transition Systems.”</span> <em>Software and Systems Modeling</em> 22 (2): 495–520. <a href="https://doi.org/10.1007/s10270-023-01092-7">https://doi.org/10.1007/s10270-023-01092-7</a>.
</div></div></section>
<section id="multispans-in-mathsfdiag_leftarrowmathsfsh_mathbbr" class="level2 page-columns page-full">
<h2 data-anchor-id="multispans-in-mathsfdiag_leftarrowmathsfsh_mathbbr">Multispans in <img src="https://latex.codecogs.com/png.latex?%5Cmathsf%7BDiag%7D_%5Cleftarrow(%5Cmathsf%7BSh%7D_%5Cmathbb%7BR%7D)"></h2>
<p><span class="citation" data-cites="Patterson2022">(Patterson et al. 2022)</span> provides a “big tent” for both discrete-space and continuous-space differential equations, thereby enabling compositional multiphysics, but does not model stochastic processes.</p>
<div class="no-row-height column-margin column-container"><div id="ref-Patterson2022" class="csl-entry">
Patterson, Evan, Andrew Baas, Timothy Hosgood, and James Fairbanks. 2022. <span>“A Diagrammatic View of Differential Equations in Physics.”</span> <a href="https://doi.org/10.3934/mine.2023036">https://doi.org/10.3934/mine.2023036</a>.
</div></div></section>
</section>
<section id="conclusion" class="level1">
<h1>Conclusion</h1>
<p>What we have outlined in this post is not just a series of conjectures or a simple project outline, but a full research program, which could supply decades of work across a large community of mathematicians, scientists, and engineers. However, that is not to say that we will only see benefits from this research program after decades of work. We expect that progress can be made in many ways along these lines, but more importantly we see that progress <em>is already</em> being made, throughout both the applied category theory and alignment communities, as evidenced by our bibliography.</p>
<p>Finally, there are plenty of mathematical problems that need to be solved in order to lay the foundations that we need for this program. But the larger challenge is to integrate all of the progress being made, not into a seamless edifice, but into a cohesive patchwork whole, where each abstract concept learned is exploited in as many ways as possible across all domains. This efficient use of concepts is demanded by the twin requirements that it is possible for humans to use the fruits of this program, and that this program has a truly <em>global</em> scope. If you have not yet been convinced of the use of category theory, then hear this: Category theory is the only conceptual technology and intellectual tradition of sufficient maturity and breadth to have a chance of solving this larger challenge.</p>
</section>
<section id="appendix-some-initial-threads-of-research" class="level1 page-columns page-full">
<h1>Appendix: Some initial threads of research</h1>
<p>The general hope here is that by suitably clever foundations, we can relegate some of the tricky analysis necessary in dealing with large state spaces into some (large) constant amount of work, which will then allow us to work synthetically when we go to actually model systems. However, this constant amount of work still has to be done; we need to find “well-behaved” categories to have semantics in. In this section, we talk about some approaches we hope would lead to such categories.</p>
<section id="von-neumann-algebra-approach" class="level2 page-columns page-full">
<h2 data-anchor-id="von-neumann-algebra-approach">Von Neumann algebra approach</h2>
<p>One of the original applications of category theory was algebraic geometry, where category theory allowed the duality between algebra and geometry to be written down explicitly with the identification of the category of affine varieties (spaces described by the zeros of polynomials) with the dual to the category of commutative rings. Since then, many other duality results have been discovered, as tabulated at the nlab <span class="citation" data-cites="nlab:gelfand_duality">(nLab authors 2023)</span>. We hope to use similar ideas to get a handle on some of the challenges of stochastic dynamical systems in category theory, using functional analysis.</p>
<div class="no-row-height column-margin column-container"><div id="ref-nlab:gelfand_duality" class="csl-entry">
nLab authors. 2023. <span>“Gelfand Duality.”</span> <a href="https://ncatlab.org/nlab/show/Gelfand+duality" class="uri">https://ncatlab.org/nlab/show/Gelfand+duality</a>.
</div><div id="ref-applebaum_semigroups_2019" class="csl-entry">
Applebaum, David. 2019. <em>Semigroups of Linear Operators: With Applications to Analysis, Probability and Physics</em>. London Mathematical Society Student Texts. Cambridge: Cambridge University Press. <a href="https://doi.org/10.1017/9781108672641">https://doi.org/10.1017/9781108672641</a>.
</div></div><p>The direct connection between stochastic dynamical systems and functional analysis is that jump-drift-diffusion systems are often modeled by Markov semigroups acting on function spaces <span class="citation" data-cites="applebaum_semigroups_2019">(Applebaum 2019)</span>. In this setting, the Hille-Yosida theorem tells us how to write down the “derivative” of such a system as an unbounded operator. With an eye towards formalizing this, we have a couple conjectures related to this setup.</p>
<p>Commutative von Neumann algebras are well-studied objects within functional analysis, known to have deep connections to probability theory. Surprisingly, it seems that although the basic theory of von Neumann algebras and the analogies to probability theory have been known for a long time, a formal proof that the category of commutative von Neumann algebras is actually dual to some category of measurable spaces was not available until quite recently <span class="citation" data-cites="pavlov_gelfand-type_2022">(Pavlov 2022)</span> (specifically, the category of compact strictly localizable enhanced measurable spaces). We conjecture that there should be a generalization of the theorem proved in that paper.</p>
<div class="no-row-height column-margin column-container"><div id="ref-pavlov_gelfand-type_2022" class="csl-entry">
Pavlov, Dmitri. 2022. <span>“Gelfand-Type Duality for Commutative von Neumann Algebras.”</span> <em>Journal of Pure and Applied Algebra</em> 226 (4): 106884. <a href="https://doi.org/10.1016/j.jpaa.2021.106884">https://doi.org/10.1016/j.jpaa.2021.106884</a>.
</div></div><div class="callout callout-style-simple callout-none no-icon callout-titled">
<div class="callout-header d-flex align-content-center">
<div class="callout-icon-container">
<i class="callout-icon no-icon"></i>
</div>
<div class="callout-title-container flex-fill">
<span class="screen-reader-only">None</span>Conjecture
</div>
</div>
<div class="callout-body-container callout-body">
<p>There is a category <img src="https://latex.codecogs.com/png.latex?%5Cmathsf%7BPCVNA%7D"> where the objects are commutative von Neumann algebras and the morphisms are <em>unital positive linear maps</em>, where a unital positive linear map from <img src="https://latex.codecogs.com/png.latex?A"> to <img src="https://latex.codecogs.com/png.latex?B"> is a bounded linear map from <img src="https://latex.codecogs.com/png.latex?A"> to <img src="https://latex.codecogs.com/png.latex?B"> that sends the positive elements of <img src="https://latex.codecogs.com/png.latex?A"> to positive elements of <img src="https://latex.codecogs.com/png.latex?B">, and sends <img src="https://latex.codecogs.com/png.latex?1"> to <img src="https://latex.codecogs.com/png.latex?1">.</p>
<p>Then <img src="https://latex.codecogs.com/png.latex?%5Cmathsf%7BPCVNA%7D"> is equivalent to the dual of the Kleisli category for the Giry monad on compact strictly localizable enhanced measurable spaces.</p>
</div>
</div>
<p>This can be though of as a category-theoretic version of the Riesz-Markov-Kakutani theorem. This states that for a compact Hausdorf space <img src="https://latex.codecogs.com/png.latex?X">, unital positive linear maps between <img src="https://latex.codecogs.com/png.latex?L%5E%5Cinfty(X)"> and <img src="https://latex.codecogs.com/png.latex?%5Cmathbb%7BC%7D"> correspond to probability measures on <img src="https://latex.codecogs.com/png.latex?X">. Then more generally, unital positive linear maps from <img src="https://latex.codecogs.com/png.latex?L%5E%5Cinfty(X)"> to <img src="https://latex.codecogs.com/png.latex?L%5E%5Cinfty(Y)"> correspond to Markov kernels from <img src="https://latex.codecogs.com/png.latex?Y"> to <img src="https://latex.codecogs.com/png.latex?X">.</p>
<p>We further conjecture that one of the tensor products (there are several) on von Neumann algebras gives an equivalent symmetric monoidal structure on <img src="https://latex.codecogs.com/png.latex?%5Cmathsf%7BPCVNA%7D%5E%5Cmathrm%7Bop%7D"> to the one given by the cartesian product on the Kleisli category of the Giry monad.</p>
<p>Then we have a Markov category structure on <img src="https://latex.codecogs.com/png.latex?%5Cmathsf%7BPCVNA%7D%5E%5Cmathrm%7Bop%7D">, because each von Neumann algebra has a multiplication <img src="https://latex.codecogs.com/png.latex?A%20%5Cotimes%20A%20%5Cto%20A">, which behaves like the diagonal in the dual category. A map <img src="https://latex.codecogs.com/png.latex?A%20%5Cto%20B"> is deterministic if and only if it preserves multiplication, which corresponds to the result that a homomorphism of von Neumann algebras <img src="https://latex.codecogs.com/png.latex?L%5E%5Cinfty(X)%20%5Cto%20L%5E%5Cinfty(Y)"> is equivalent to a measurable map <img src="https://latex.codecogs.com/png.latex?Y%20%5Cto%20X">.</p>
<p>The nice thing about this result is that we did not have to “add” anything to the von Neumann algebras; we just had to relax the notion of morphism to not preserve multiplication.</p>
<p>Finally, we conjecture that we can again relax the notion of morphism, to be only convex instead of linear, and then recover a category similar to the Kleisli category of an infradistribution monad (see <span class="citation" data-cites="diffractor_introduction_nodate">(Diffractor and Kosoy n.d.)</span> for an introduction to the idea of infradistributions, and <span class="citation" data-cites="topos_institute_nisan_2023">(Topos Institute 2023)</span> for the relation to convex maps <img src="https://latex.codecogs.com/png.latex?C(X)%20%5Cto%20%5Cmathbb%7BR%7D_%7B%5Cgeq%200%7D">).</p>
<div class="no-row-height column-margin column-container"><div id="ref-diffractor_introduction_nodate" class="csl-entry">
Diffractor, and Vanessa Kosoy. n.d. <span>“Introduction to the Infra-Bayesianism Sequence.”</span> Accessed June 9, 2023. <a href="https://www.alignmentforum.org/posts/zB4f7QqKhBHa5b37a/introduction-to-the-infra-bayesianism-sequence">https://www.alignmentforum.org/posts/zB4f7QqKhBHa5b37a/introduction-to-the-infra-bayesianism-sequence</a>.
</div><div id="ref-topos_institute_nisan_2023" class="csl-entry">
Topos Institute. 2023. <em>Nisan Stiennon: Metagames and Imprecise Probability</em>. <a href="https://www.youtube.com/watch?v=2HF9wJ88shM">https://www.youtube.com/watch?v=2HF9wJ88shM</a>.
</div></div><p>Apart from the technically interesting merits of this thread of research, this also leads us to an “observables-first” viewpoint on dynamical systems, which we explore more in the next thread.</p>
</section>
<section id="feynmankac-approach" class="level2 page-columns page-full">
<h2 data-anchor-id="feynmankac-approach">Feynman–Kac approach</h2>
<p>Functional analysis has some nice results that lend themselves to category theory, but the theory of unbounded operators is both difficult and subtle. Fortunately, we are typically <em>actually</em> working with probabilistic observables on some smooth space, which gives us some handles to make our problem more well-behaved. To stochastic processes on manifolds, there are associated Feynman-Kac PDEs (Owen learned this from <span class="citation" data-cites="spreij_stochastic_2023">(Spreij 2023)</span>, the original theory goes back to <span class="citation" data-cites="kac_distributions_1949">(Kac 1949)</span>). Once we are in PDE land, we have an opportunity to use synthetic differential geometry, which is known to play nicely with category theory. <span class="citation" data-cites="lavendhomme_basic_1996">(Lavendhomme 1996)</span>.</p>
<div class="no-row-height column-margin column-container"><div id="ref-spreij_stochastic_2023" class="csl-entry">
Spreij, P. J. C. 2023. <em>Stochastic Integration</em>. Universiteit van Amsterdam. <a href="https://staff.fnwi.uva.nl/p.j.c.spreij/onderwijs/master/si.pdf">https://staff.fnwi.uva.nl/p.j.c.spreij/onderwijs/master/si.pdf</a>.
</div><div id="ref-kac_distributions_1949" class="csl-entry">
Kac, M. 1949. <span>“On Distributions of Certain Wiener Functionals.”</span> <em>Transactions of the American Mathematical Society</em> 65 (1): 1–13. <a href="https://doi.org/10.1090/S0002-9947-1949-0027960-X">https://doi.org/10.1090/S0002-9947-1949-0027960-X</a>.
</div><div id="ref-lavendhomme_basic_1996" class="csl-entry">
Lavendhomme, René. 1996. <em>Basic Concepts of Synthetic Differential Geometry</em>. Vol. 13. Kluwer Texts in the Mathematical Sciences. Boston, <span>MA</span>: Springer <span>US</span>. <a href="https://doi.org/10.1007/978-1-4757-4588-7">https://doi.org/10.1007/978-1-4757-4588-7</a>.
</div></div><p>More concretely, a well-behaved way of dealing with PDEs on manifolds is via the theory of <em>derivations</em> on <img src="https://latex.codecogs.com/png.latex?C%5E%5Cinfty"> rings. Roughly speaking, <img src="https://latex.codecogs.com/png.latex?C%5E%5Cinfty"> rings are an axiomatization of the properties that the ring of smooth functions <img src="https://latex.codecogs.com/png.latex?X%20%5Cto%20%5Cmathbb%7BR%7D"> has for a manifold <img src="https://latex.codecogs.com/png.latex?X">.</p>
<div class="callout callout-style-simple callout-none no-icon callout-titled">
<div class="callout-header d-flex align-content-center">
<div class="callout-icon-container">
<i class="callout-icon no-icon"></i>
</div>
<div class="callout-title-container flex-fill">
<span class="screen-reader-only">None</span>Definition
</div>
</div>
<div class="callout-body-container callout-body">
<p>A first-order derivation on a <img src="https://latex.codecogs.com/png.latex?C%5E%5Cinfty"> ring <img src="https://latex.codecogs.com/png.latex?R"> is a linear operator <img src="https://latex.codecogs.com/png.latex?%5Cmathrm%7Bd%7D%20%5Ccolon%20R%20%5Cto%20R"> that satisfies the Leibniz rule, <img src="https://latex.codecogs.com/png.latex?%5Cmathrm%7Bd%7D(ab)%20=%20a%20%5C;%20%5Cmathrm%7Bd%7Db%20+%20b%20%5C;%20%5Cmathrm%7Bd%7Da"></p>
</div>
</div>
<p>If <img src="https://latex.codecogs.com/png.latex?R%20=%20C%5E%5Cinfty(X)"> for <img src="https://latex.codecogs.com/png.latex?X"> a manifold, then first-order derivations on <img src="https://latex.codecogs.com/png.latex?R"> are in bijective correspondence with vector fields on <img src="https://latex.codecogs.com/png.latex?R">. We can then compose first-order derivations as operators and get higher-order spatial derivatives. This allows us to do the second-order differential equations necessary for the evolution of probability distributions in an “algebraic” way.</p>
<p>The disadvantage is that this is not so “natively” probabilistic, and thus might not play well with, for instance, Bayesian inference (or infra-Bayesian inference). Additionally, the classical theory of Feynman-Kac PDEs does not cover jump processes; there may be an extension to the case of jump processes but I’m not aware of it.</p>
<p>In both this approach and the previous approach, there have been interesting insights made by moving to the “algebra of observables” point of view. This point of view is also desirable because symbolic computational systems also benefit from taking an “algebraic” as opposed to a geometric point of view, and as we hope to eventually implement these systems, the fact that the theory meshes well with computational practice is important.</p>
</section>
<section id="epistemically-generalized-behavioral-approach" class="level2 page-columns page-full">
<h2 data-anchor-id="epistemically-generalized-behavioral-approach">Epistemically generalized behavioral approach</h2>
<p>The <strong>behavioral approach to systems theory</strong> <span class="citation" data-cites="willems97">(Willems and Polderman 1997)</span> models systems with observable states <img src="https://latex.codecogs.com/png.latex?X"> as subsets of trajectory space <img src="https://latex.codecogs.com/png.latex?C(%5Cmathbb%7BR%7D%5E+,X)%5Crightarrow%20%5Cmathbf%7B2%7D">. This reflects a “nondeterministic” approach to uncertainty: a subset is an epistemic state which picks out certain trajectories as possible. Given a way of relating two interfaces <img src="https://latex.codecogs.com/png.latex?X%20%5Cxrightarrow%7Bf_X%7D%20R%20%5Cxleftarrow%7Bf_Y%7D%20Y">, one can compose systems by pullback.</p>
<div class="no-row-height column-margin column-container"><div id="ref-willems97" class="csl-entry">
Willems, Jan C, and Jan W Polderman. 1997. <em>Introduction to Mathematical Systems Theory: A Behavioral Approach</em>. Vol. 26. Springer Science &amp; Business Media.
</div></div><p>Probabilistic theories do not have such a simple composition, because of the “disintegration” difficulties around normalization after conditioning on equality.</p>
<div class="callout callout-style-simple callout-none no-icon callout-titled">
<div class="callout-header d-flex align-content-center">
<div class="callout-icon-container">
<i class="callout-icon no-icon"></i>
</div>
<div class="callout-title-container flex-fill">
<span class="screen-reader-only">None</span>Conjecture
</div>
</div>
<div class="callout-body-container callout-body">
<p>Credal sets offer a way to translate directed stochastic Markov kernels into an undirected, behavioral-style epistemic state, in such a way that preserves the semantics of <img src="https://latex.codecogs.com/png.latex?(f%20:%20A%20%5Crightarrow%20B);%20(g%20:%20B%20%5Crightarrow%20C)"> style composition.</p>
</div>
</div>
</section>
<section id="generalized-spaces-for-probability" class="level2 page-columns page-full">
<h2 data-anchor-id="generalized-spaces-for-probability">Generalized spaces for probability</h2>
<p>There is a general approach to building nice categories of spaces. What you do is you start with the nice examples of the type of space you care about. For instance, if you want smooth spaces, these are open subsets of <img src="https://latex.codecogs.com/png.latex?%5Cmathbb%7BR%7D%5En">. If you want spaces locally described by polynomial equations, these are affine varieties. If you want spaces suitable for probability theory, these are standard Borel spaces. Then the category of your “nice spaces” with suitable maps may not be nice, in that it might not have all the colimits you want. In order to “complete it”, you put a suitable Grothendieck topology on it, and then consider the corresponding category of sheaves, or possibly the subcategory of concrete sheaves.</p>
<p>This is the approach taken for <a href="https://ncatlab.org/nlab/show/diffeological+space">diffeological spaces</a>, <a href="https://ncatlab.org/nlab/show/scheme">schemes</a>, and <a href="https://ncatlab.org/nlab/show/quasi-Borel+space">quasi-Borel spaces</a>.</p>
<p>We have been thinking about using this approach to make a category of spaces that are suitable for both stochastic behavior and continous behavior. More technically, a category of spaces with both a <a href="https://ncatlab.org/nlab/show/tangent+bundle+category">tangent structure</a> and a <a href="https://ncatlab.org/nlab/show/Giry+monad">Giry monad</a>.</p>
<p>One approach is to work with spaces modeled after convex subsets of <img src="https://latex.codecogs.com/png.latex?%5Cmathbb%7BR%7D%0A%5En">, which we call convological spaces in analogy to diffeological spaces. Convex spaces have a natural notion of “tangent cone”, which models the fact that convex spaces often have boundaries, so you might only be able to go in one direction from a point and not in the reverse of that direction. Moreover, convex spaces are deeply connected to probability theory, because the space of probability distributions on a measure space has natural convex structure. However, more work is needed to understand exactly what the properties and use of these convological spaces are.</p>
<p>Another approach, which is not due to us, is to work with what are called PAP spaces, which are modeled after subsets of <img src="https://latex.codecogs.com/png.latex?%5Cmathbb%7BR%7D%5En"> that are the support of analytic functions <span class="citation" data-cites="huot23a">(Huot et al. 2023)</span>. These have been used to provide semantics for probabilistic programs which are also differentiable, which is needed to run inference algorithms like Hamiltonian Monte Carlo on them.</p>
<div class="no-row-height column-margin column-container"><div id="ref-huot23a" class="csl-entry">
Huot, Mathieu, Alexander K. Lew, Vikash K. Mansinghka, and Sam Staton. 2023. <span>“<img src="https://latex.codecogs.com/png.latex?%5Comega">PAP Spaces: Reasoning Denotationally about Higher-Order, Recursive Probabilistic and Differentiable Programs.”</span>
</div></div><p>Building nice categories of spaces via sheaves is such a general technique that even if PAP spaces or convological spaces turn out to not be useful for what we want to do, it is likely that some category of spaces defined in this way will show up at some point.</p>



</section>
</section>
</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>dynamical systems</category>
  <category>category theory</category>
  <guid>https://topos.institute/blog/2023-06-15-compositional-world-modeling/</guid>
  <pubDate>Thu, 15 Jun 2023 00:00:00 GMT</pubDate>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Elena Di Lavore</dc:creator>
  <dc:creator>Mario Román</dc:creator>
  <link>https://topos.institute/blog/2023-02-22-what-made-fra-structure-successful/</link>
  <description><![CDATA[ 




<div id="post-content">
<p>The Topos Institute recently hosted “Finding the Right Abstractions for Healthy Systems”, with 24 researchers, mainly from the applied category theory and AI safety communities. This is a post trying to understand what made it successful and some ideas for similar events.</p>
<ol type="1">
<li><p><strong>Ask the experts.</strong> <a href="https://www.lesswrong.com/users/duncan_sabien">Duncan Sabien</a> moderated and directed the workshop. Caleb Rak from <a href="https://www.canopyretreats.org/">Canopy Retreats</a> organized all the logistics. We would like to thank them both, and we believe they made the workshop much better.</p>
<p>As a more general principle, it seems that there is an art to organizing workshops that is usually neglected. Having experts taking care of this enhances the whole experience and any workshop that can afford this opportunity will probably benefit. The rest of the advice we can extract is conditional on this first point.</p></li>
<li><p><strong>No distractions.</strong> We probably underestimate how small things can tax your mind during a research retreat. Having well-managed logistics (venue, schedule, markers and whiteboards, sleep, covid testing, food…) minimizes the possibility of minor inconveniences disrupting work. Participants’ attention is a valuable resource; this just helps using it properly.</p></li>
<li><p><strong>Direct the research.</strong> Extracting the most out of participants does also include the same care at the research level. We had some meta-discussion about how the research was to be organized; and some points seem to be good general principles.</p>
<p>Research works best with clear goals, clear communication protocols, efficient use of the time, carefully curating and deciding what research directions to pursue, and support for a collaborative environment. Sitting in assembly to discuss research directions every day helps keep the group working as a whole.</p></li>
<li><p><strong>And reassess.</strong> Research is an exploration; changing direction is part of it. Math and category theory are very good at keeping claims accountable: something is either an algebra for your operad or it is not; and you may have to change your position based on this. However, doing this is not enough. It was great to share goals with others to avoid getting lost in rabbit holes and remain open to new ideas. Participants were experts in domains adjacent to the main goal of the workshop: it was invaluable to exchange ideas in well-organized group sessions and talks.</p></li>
</ol>
<p>Research seminars and workshops traditionally follow the same general standard structure; but it is actually difficult to compare what makes them work if each one faces different constraints (from funding to scope). Hopefully, starting a discussion about what makes them work may help us improve the ways in which we carry collaborative research.</p>


</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>events</category>
  <guid>https://topos.institute/blog/2023-02-22-what-made-fra-structure-successful/</guid>
  <pubDate>Wed, 22 Feb 2023 00:00:00 GMT</pubDate>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>David Spivak</dc:creator>
  <link>https://topos.institute/blog/2022-11-08-matter-and-pattern/</link>
  <description><![CDATA[ 




<div id="post-content">
<p>Etymologically, the word <em>matter</em> comes from mother and the word <em>pattern</em> comes from father. Like two parents, matter and pattern represent a fundamental dichotomy: matter is the pure material, unconcerned with our ideas about it; pattern is pure structure, unconcerned with what substantiates it. Considering this dichotomy brings up thoughts of Cartesian dualism. Like Descartes, we need to think about where matter and pattern meet, and hopefully we can do better than “the pineal gland” as an answer. In this post, I’ll discuss how all this relates to language and compositionality, and hence to category theory, as well as explain my hope that we may someday find a dynamic operad or other categorical framework that can account for why the meeting place of matter &amp; pattern seems to condense over time.</p>
<section id="introduction" class="level1">
<h1>Introduction</h1>
<p>At a Computational Theology workshop, held a few months ago in Austin TX, a number of us—Scott Garrabrant, Sophie Libkind (who also helped me refine the ideas for this post), Eliana Lorch, Anna Salamon, and I—spent several hours trying to work out how math and matter relate. How does the math fact that 4+3=7 relate to the paper-and-ink display “4+3=7”, or to some person’s vocal chords vibrating to create the sound “four plus three equals seven”? The math is pure pattern, and the sound, as vibrating air, is pure matter. What’s going on with the link between them?</p>
<p>Eventually, we came upon the transistor as an excellent example of where math and matter meet.</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="transistor.png" class="lightbox" data-gallery="quarto-lightbox-gallery-1" title="A picture of an NPN transistor and its circuit diagram"><img src="https://topos.institute/blog/2022-11-08-matter-and-pattern/transistor.png" class="img-fluid figure-img" style="width:50.0%" alt="A picture of an NPN transistor and its circuit diagram"></a></p>
<figcaption>A picture of an NPN transistor and its circuit diagram</figcaption>
</figure>
</div>
<p>On the left we see the transistor as matter, a thing in the material world. On the right we see the transistor as pattern, a logical idea. <a href="https://youtu.be/rpivf1SoEdc?t=195">John Vervaeke says</a> that the term <em>symbol</em> is appropriate here—that the symbol is not just a signifier of some transformation but also an active participant in materially achieving that transformation, i.e.&nbsp;that it’s both pattern and matter—though I don’t know how widespread that terminology is. It seems a little more on the “pattern” side, but I need a word, so I’ll go with it throughout this post, saying that the transistor is a <em>symbol</em> in this sense. If you think of a better word than “symbol” for referring to the fulcrum or janus-point which, like a transistor, is both matter and pattern, please put it in the comment section below!</p>
<p>It is now believed that all of math can be recorded and developed within a proof assistant like Agda, Coq, HOL, or Lean. These programs run on computers, computers are made of logic gates, logic gates are made of NAND gates, and NAND gates are formed by attaching two transistors like so:</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="nand.png" class="lightbox" data-gallery="quarto-lightbox-gallery-2" title="A circuit diagram for a NAND gate"><img src="https://topos.institute/blog/2022-11-08-matter-and-pattern/nand.png" class="img-fluid figure-img" style="width:50.0%" alt="A circuit diagram for a NAND gate"></a></p>
<figcaption>A circuit diagram for a NAND gate</figcaption>
</figure>
</div>
<p>One can imagine the connection between math and matter, via proof assistants that run on transistors, as taking place in a kind of hour-glass shape.</p>
<p><a href="hourglass.png" class="lightbox" data-gallery="quarto-lightbox-gallery-3"><img src="https://topos.institute/blog/2022-11-08-matter-and-pattern/hourglass.png" class="img-fluid" style="width:70.0%"></a></p>
<p>We do math purely conceptually, and yet these concepts need to turn into material action in order to affect the world. The transistor is a symbol that serves in a dual role, as both logic and matter, and happens to be close to the atomic scale in both: it is half of a NAND gate (upon which all the logic necessary for a computer is built), and as of late 2022 it is about 2 nanometers (the size of 10 silicon atoms) in length.</p>
<p>In the next section I’ll discuss other examples of symbols that serve in a dual role of matter and pattern, and that like the transistor are as condensed as possible. Then I’ll say what this has to do with category theory and its role in the world.</p>
</section>
<section id="examples" class="level1">
<h1>Examples</h1>
<p>Besides transistors, what other examples can we find of “symbols”: i.e. condensed instantiations of both matter and pattern? And why are they important?</p>
<ol type="1">
<li><p><strong>DNA.</strong> As an acidic molecule in 3D space that acts according to physical laws, DNA is matter; but as an ordered sequence of four letters whose three-letter words code for amino acids, it is pattern. Reading the DNA and elaborating its meaning involves many other parts of the cell, and we could consider that whole complex to be both matter and pattern, but DNA is more condensed, more symbolic. Is DNA matter or pattern?</p></li>
<li><p><strong>A signature</strong>. As the process of moving a pen across paper to form an ink stain, a signature is matter; but as a token of people’s agreement to regulate their behavior, it is pattern. As we sign a contract in good faith, we’ve set up our internal state in such a way that we think it’s likely that our actions in the material world will follow the pattern dictated by the contract. Is signature matter or pattern?</p></li>
<li><p><strong>Grandma neuron</strong>. Neuroscientists say that a single neuron can code for a single concept, e.g.&nbsp;neuron X fires if and only if one recognizes their own grandma. As a carbon-based object, a neuron is matter; but as representing one’s grandma, it is pattern. If we look at just the grandma neuron, would you call it matter or pattern?</p></li>
</ol>
<p>There are many examples, and some are clearer than others. But one thing that strikes me in all this is that there is some process by which examples keep being created, symbols keep getting formed and condensed in our world. Probably early life’s control mechanism was far less condensed than “modern” DNA. Whatever happened in that prebiotic chemistry on the early earth, the robust and effective language for building custom proteins must have been repeatedly refined over eons. Similarly, coordination of animal activity is ancient, but the condensation of this coordination into a binding contract, or the symbolic signature itself, is quite new. And a similar thing could be said about the grandma neuron or the transistor. In each case there seems to be some sort of natural push or process urging the formation of smaller, more concentrated symbols to instantiate pattern as matter. More condensed symbols seem to work better.</p>
<p>What is this natural process? How does it work in so many domains and on so many scales at once? And why is condensing and compressing these symbols somehow “preferred” by evolution? I would love to know.</p>
</section>
<section id="compositionality-and-language-formation" class="level1">
<h1>Compositionality and language formation</h1>
<p>Transistors wouldn’t be nearly as important in our world today if they didn’t form NAND gates, and hence logic gates, adder circuits, CPUs, etc. But all that is purely conceptual, i.e.&nbsp;pattern. The thing that makes this grammar work is that it simultaneously fits the material embodiment as well. The material transistors can be arranged <em>according</em> to the conceptual pattern and there is a kind of “functoriality” there: just as the matter instantiates the pattern at the lowest level, so does the matter instantiate the pattern at the higher levels too. Logic gates are made of NAND gates wired together, both in pattern and in matter, and this analogy keeps holding all the way up. Your computer has been programmed to run the programs that let you read this, but it is operating on physics in the material world, thanks to the robustness and compositionality of the transistor as symbol.</p>
<p>Similarly, a single contract is important, but an organization works with a whole slew of contracts. When each is being decided upon, the composition of the whole—how all those agreements are going to be carried out in the one organization—requires a conceptual understanding of roles and activities, all of which will be instantiated by human bodies. The composition of agreements within the company reflects the composition of movements by its employees.</p>
<p>The same can be said for DNA or neurons. A single nucleotide does not define DNA, nor does a single neuron define the brain. The organization of nucleotides in DNA or of neurons in the brain is extremely important for determining what the DNA or brain function will actually be. The point is that in each case the function is carried out materially, and this works because the analogy is <em>compositional</em> with respect to the patterns of connection in these organized systems.</p>
<p>So we’re not only interested in the symbol at the center of matter and pattern, but in the compositionality—the grammar—by which the composition of conceptual patterns and the composition of material flows maintain their alignment.</p>
<p>We can see that language occurs materially and affects the material world, and it is also patterned in that it follows conceptual grammatical rules. When I say “pass the salt”, <img src="https://latex.codecogs.com/png.latex?10%5E%7B20%7D"> atoms including your arm and a salt shaker move through space, resulting in some sodium chloride crystals arriving in my bowl of soup. Language controls so much of our world, especially if you see things like DNA as language. And just as important as the fundamental symbols at the base of language is the fact that language is compositional.</p>
<p>Thus the drive toward finding symbols, which bridge matter &amp; pattern, and do so compositionally, seems to be the same as, or at least tightly linked with, language formation. So it’s interesting to ask: by what process is language—including the language of thought, the language of computation, the language of life, etc—formed? This formation appears to be a natural multi-scale process, beginning at least as far back as DNA, and continuing to this day.</p>
</section>
<section id="an-applied-category-theory-question" class="level1">
<h1>An applied category theory question</h1>
<p>Though I’ve used different terms for it over the years, I’ve been interested in using category theory to study collective intelligence and sense-making for just over 15 years now. A current instantiation of this question is represented in this blog post:</p>
<blockquote class="blockquote">
<p><em>Can we use CT to formalize a process by which language formation, i.e.</em> <em>by which the condensation of compositional matter-pattern symbols,</em> <em>would naturally occur?</em></p>
</blockquote>
<p>I would love to help answer this question.</p>
<p>Recently, <a href="../../people/brandon-shapiro/">Brandon Shapiro</a> and I developed something that might be relevant: a category theoretic framework called <em>Dynamic Organizational Structures</em>, e.g. <a href="https://arxiv.org/abs/2205.03906">dynamic operads, dynamic categories</a>, etc. We have four examples: deep learning, prediction markets, non-cooperative strategic games, and (Sophie and I developed the example of) Hebbian learning. In each case, there is a multi-scale system for updating the ways parts interact to form wholes.</p>
<p>Perhaps the sort of natural process by which matter-pattern symbols are condensed is scale-invariant enough that it would fit into this framework. In other words, I’m wondering: is there a dynamic operad for the language-formation process? This is of course a very loose question, but I think that’s ok; it’s only meant to inspire people to consider and work on it. Indeed, the <a href="../../blog/2022-10-28-hebbian-learning/">Hebbian Learning example</a> above was only loosely inspired by <a href="https://en.wikipedia.org/wiki/Spike-timing-dependent_plasticity">spike-time dependent plasticity</a>, but it still serves to give us some insight into the grammar of Hebbian learners. Applied category theory is very broad, and there’s no reason one needs to use dynamic operads, though they may be useful or at least inspire a more useful framework. I’m not really hoping for a definitive category-theoretic answer to the question of how language or symbols are repeatedly formed, though it’d be interesting to hear if someone thought they had such a thing; instead, I’m looking to consider it and make progress on it.</p>
<p>Category theory itself is language, and it’s really condensed and powerful. And with tools like <a href="https://github.com/AlgebraicJulia/Catlab.jl">Catlab</a>, it’s moving closer to the matter side of the divide: the categorical patterns are being instantiated in silicon and metal. Is it just a coincidence that we see this condensation happening yet again? Imagine how powerful it would be to instantiate a compositional language-formation process within computational category theory software. If it’s true that nature keeps making this sort of thing happen, as though by some “invisible hand”, then such an implementation may well be in our future. The more well-rounded our understanding of it is, the more elegantly the system can be designed, and the more frictionless I imagine the result will be.</p>


</section>
</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>operads</category>
  <category>applied category theory</category>
  <category>language</category>
  <guid>https://topos.institute/blog/2022-11-08-matter-and-pattern/</guid>
  <pubDate>Mon, 07 Nov 2022 00:00:00 GMT</pubDate>
  <media:content url="https://topos.institute/blog/2022-11-08-matter-and-pattern/hourglass.png" medium="image" type="image/png" height="74" width="144"/>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Brendan Fong</dc:creator>
  <link>https://topos.institute/blog/2022-11-04-youtube-and-gender/</link>
  <description><![CDATA[ 




<div id="post-content">
<p>In late September, we published what I think is a <a href="https://www.youtube.com/watch?v=eXBwU9ieLL0">wonderful introduction to category theory by educational designer Paul Dancstep</a>. Viewers agree:</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="youtubecomment.png" class="lightbox" data-gallery="quarto-lightbox-gallery-1" title="YouTube comment: ‘outstanding’"><img src="https://topos.institute/blog/2022-11-04-youtube-and-gender/youtubecomment.png" class="img-fluid figure-img" alt="YouTube comment: ‘outstanding’"></a></p>
<figcaption>YouTube comment: ‘outstanding’</figcaption>
</figure>
</div>
<p>And YouTube does too, with an unusually high proportion of our views coming via YouTube recommendations:</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="trafficsources.png" class="lightbox" data-gallery="quarto-lightbox-gallery-2" title="Traffic sources for Paul’s video, as of October 7: 82% of traffic comes from YouTube recommendations"><img src="https://topos.institute/blog/2022-11-04-youtube-and-gender/trafficsources.png" class="img-fluid figure-img" alt="Traffic sources for Paul’s video, as of October 7: 82% of traffic comes from YouTube recommendations"></a></p>
<figcaption>Traffic sources for Paul’s video, as of October 7: 82% of traffic comes from YouTube recommendations</figcaption>
</figure>
</div>
<p>This is wonderful: we love that so many people are learning about category theory! Category theory is a beautiful part of mathematics, and we believe it will be an important ingredient in deciphering the complexity of the systems around us, and shaping them for the better.</p>
<p>And yet, there’s a frustrating turn. Just <em>who</em> is learning about category theory? Let’s take a closer look:</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="demographics_paul.png" class="lightbox" data-gallery="quarto-lightbox-gallery-3" title="Audience by age and gender for Paul’s video, as of October 7: 100% of viewers are identified as male"><img src="https://topos.institute/blog/2022-11-04-youtube-and-gender/demographics_paul.png" class="img-fluid figure-img" alt="Audience by age and gender for Paul’s video, as of October 7: 100% of viewers are identified as male"></a></p>
<figcaption>Audience by age and gender for Paul’s video, as of October 7: 100% of viewers are identified as male</figcaption>
</figure>
</div>
<p>The numbers are stark: 0% female, 100% male.</p>
<p>To understand what’s happening better, we must also look back at our typical viewership. Despite our firm belief that category theory and mathematics is for all, our YouTube viewership in general strongly skews male. Nonetheless, in August, the month before this video was published, YouTube identified around one in five of our viewers as female.</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="demographics_august.png" class="lightbox" data-gallery="quarto-lightbox-gallery-4" title="Audience by age and gender for all Topos videos, August 2022: 80% of viewers are identified as male"><img src="https://topos.institute/blog/2022-11-04-youtube-and-gender/demographics_august.png" class="img-fluid figure-img" alt="Audience by age and gender for all Topos videos, August 2022: 80% of viewers are identified as male"></a></p>
<figcaption>Audience by age and gender for all Topos videos, August 2022: 80% of viewers are identified as male</figcaption>
</figure>
</div>
<p>Through initiatives such as our <a href="../../em-cats/">Em-Cats seminar</a> and our support of the <a href="http://adjointschool.com">Adjoint School</a> and <a href="https://www.womeninlogic.org">Women in Logic</a>, we’re working hard to make our community more inclusive, diverse, and equitable, and in doing so make the gender balance in all parts of the Topos community more representative of the wider world. What we’re observing here, however, which is an established pattern with all our videos, puts us in a bind.</p>
<p>While these statistics must be taken with a grain of salt (the gender information is largely inferred by YouTube, and may not be accurate), the implication remains: when YouTube promotes our videos, it promotes them almost exclusively to viewers that it identifies as male.</p>
<p>More pithily: YouTube thinks category theory is for men.</p>
<p>YouTube is our most powerful outreach platform. It’s revolutionary that videos about category theory, from introductions to the field to the cutting edge of research, are now available to anyone with an internet connection, wherever they may be. Tens of thousands of people watch our videos each month. And in terms of absolute number, we’re confident YouTube makes our work readily available to more women than possible without using the platform.</p>
<p>But YouTube’s algorithmic bias means that using YouTube pits our efforts to create a growing category theory community directly against our efforts to create an inclusive category theory community.</p>
<p>Why do we care about this? Mathematics and science is our common heritage, and everyone deserves to hear about and understand it. More than this, our society is increasingly built upon technologies that are enabled by mathematics, and this trend is only intensifying. For technologies to serve all people, the ability to understand them, shape them, and reshape them must be in all of our hands. So it is essential to the Topos mission that we give everyone a voice in the future of technology. Among many other things, this starts with who sees ‘What is Category Theory?’ on their YouTube feed.</p>
<p>More broadly, we see in this dilemma the urgency of our work. As computer scientists, we understand that these recommendation engines are based on deep learning and modern statistical artificial intelligence techniques, and that in their incredible ability to detect patterns and to optimise, such biases are difficult to combat. We must work towards more accountable structures for artificial intelligence, and a critical step towards accountability will be machine learning methods that are more transparent. New mathematics, such as <a href="https://sinews.siam.org/Details-Page/deeply-interactive-learning-systems">the work of our connected intelligence team</a>, is essential for this.</p>
<p>At the same time, this problem cannot just be solved with more mathematics. We must examine the ways we teach, raise awareness of how technologies can conflict with our values, and push for change. Technology can disadvantage certain communities in hidden ways. While this example is only a small instance in <a href="https://www.penguinrandomhouse.com/books/241363/weapons-of-math-destruction-by-cathy-oneil/">the many ways our algorithms can be biased</a>, the first step to change will be to bring these hidden aspects to light. So here we are.</p>
<p>Anyway, if you haven’t already, please do check out Paul’s video! And please share it with someone who might not have the opportunity to see it otherwise.</p>
<p><lite-youtube videoid="eXBwU9ieLL0"></lite-youtube></p>


</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>community</category>
  <category>video</category>
  <category>outreach</category>
  <guid>https://topos.institute/blog/2022-11-04-youtube-and-gender/</guid>
  <pubDate>Fri, 04 Nov 2022 00:00:00 GMT</pubDate>
  <media:content url="https://topos.institute/blog/2022-11-04-youtube-and-gender/category_theory_youtube.png" medium="image" type="image/png" height="103" width="144"/>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Tish Tanski</dc:creator>
  <dc:creator>Paul Dancstep</dc:creator>
  <link>https://topos.institute/blog/2022-10-04-what-is-category-theory-anyway/</link>
  <description><![CDATA[ 




<div id="post-content">
<p>Ask most category theory experts what it is, and they’ll launch into a session at a white board that can leave your head spinning. That is, if you are not a card carrying mathematician, as many of us are not. Even at Topos Institute, we have staff, donors, and friends who are not mathematicians. We have struggled to find a simple but clear and compelling way to answer the question: So, what is category theory anyway?</p>
<p>Until now. We have partnered with Paul Dancstep, gifted science educator and communicator, to come up with a quick video to answer the “what” question. We are attempting to cruise at the 30,000 foot level so we can enthrall many with the beauty and elegance of category theory without getting into the detail that might bog them down.</p>
<p>We don’t attempt to answer the “how” does Topos Institute advance and apply category theory or “why does it matter” questions — yet. In this video we are striving for a simple description that meets the rigor of mathematics in an engaging way that is easy to grasp.</p>
<p>So here it is, with Paul Dancstep narrating. Take a look and let us know how we did. Share it with your category theory friends, your 8 year old, or your grandmother. Let us know if it works!</p>
<p><lite-youtube videoid="eXBwU9ieLL0"></lite-youtube></p>


</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>category theory</category>
  <category>video</category>
  <guid>https://topos.institute/blog/2022-10-04-what-is-category-theory-anyway/</guid>
  <pubDate>Tue, 04 Oct 2022 00:00:00 GMT</pubDate>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Joshua Meyers</dc:creator>
  <link>https://topos.institute/blog/2022-08-10-computation-category-theory/</link>
  <description><![CDATA[ 




<div id="post-content">
<p>In a recent talk, <a href="../../people/david-spivak/">David Spivak</a>, my advisor at Topos Institute, described Poly as “the language of computation”, due to its facility in describing concepts in computer science such as data migration, dependent types, and Turing machines. But is Poly really the language of computation?</p>
<p>To address this question, I decided first to take a step back and ask, “what is computation?”</p>
<p>This question has been considered for a long time. It has been answered in many different ways, variously in terms of Turing machines, lambda calculus, and general recursive functions. The famous Church-Turing thesis states the remarkable fact that all these descriptions turn out to be equivalent, pointing towards an underlying concept. In Kurt Gödel’s Remarks before the Princeton bicentennial conference on problems in mathematics (1946, Collected Works v. II p.&nbsp;150), he highlights this convergence:</p>
<blockquote class="blockquote">
<p><em>Tarski has stressed in his lecture (and I think justly) the great</em> <em>importance of the concept of general recursiveness (or Turing’s</em> <em>computability). It seems to me that this importance is largely due to</em> <em>the fact that with this concept one has for the first time succeeded</em> <em>in giving an absolute definition of an interesting epistemological</em> <em>notion, i.e., one not depending on the formalism chosen.</em></p>
</blockquote>
<p>Indeed, the functions which can be computed by a Turing machine are exactly those which are definable in the lambda calculus, which are in turn exactly general recursive functions. And no system of computation has been devised which can compute any more functions than this class. This situation is troubling, as we apparently have a singular notion, computability, which does not depend on the formalism we use, but we have no definition of it except by means of particular formalisms.</p>
<p>What is the essence of computation then, which lurks behind these diverse definitions? This problem, hitherto unsolved, seems apt for category theory, with its spirit of abstraction across multiple frameworks. It is frustrating to me as a mathematician to so often have to make recourse to a concept which is known to be absolute but can only be grasped by means of highly idiosyncratic formalisms. (Think of how many arbitrary choices are made when defining a Turing machine, which nonetheless have no effect whatsoever on the class of functions which can be computed!) If only I could grasp the concept itself!</p>
<p>If category theory can be utilized to understand computation itself, fundamental computer science will be rendered as a field of mathematics, broadening the line of communication between category theorists and computer scientists and facilitating the deployment of category theory in hardware and software, as per Topos Institute’s <a href="../../strategic-plan/">strategic plan</a>. Moreover, if applied category theorists truly understand computation, we will be able to use its power more intentionally, empowering us to create systems that benefit us all.</p>
<p>Others have attempted this program before. Robin Gandy proposed a generalization of a Turing machine, called a “Gandy machine”, in his 1980 paper “Church’s Thesis and Principles for Mechanisms”. Gandy describes 4 principles which purportedly capture the essence of a “computing machine”. Then he shows that any function which can be computed by a machine satisfying these 4 principles is Turing-computable. Furthermore, he constructs counterexamples “which show that if the principles be weakened in almost any way, then there will be devices which satisfy the weakened principles and which can calculate any number-theoretic function.”</p>
<p>Gandy’s paper is very technically involved, and his analysis was subsequently streamlined by Wilfried Sieg and John Byrnes in their 1999 paper “An Abstract Model for Parallel Computations: Gandy’s Thesis”. In <a href="https://arxiv.org/abs/1904.10109v1">A Category Theoretic Interpretation of Gandy’s Principles for Mechanisms</a>, Joseph Razawi and Andrea Schalk give a categorical description of what Gandy is doing.</p>
<p>All three of these papers are difficult, and I have not worked through all of the details, but I doubt that the essence of computation is really something so technical, as it seems a quite natural notion, albeit hard to pin down. Moreover, in my skimming of the papers, I have not found any sections which claim to justify that the formalisms they choose are the right ones to capture the concepts at hand. If anyone thinks that Gandy machines do in fact get to the essence of computation and can explain to me how, I would be very interested.</p>
<p>In 2008, Nachum Dershowitz and Yuri Gurevich published “A Natural Axiomatization of Computability and Proof of Church’s Thesis”. This paper is very long and full of quotes from famous logicians talking about their concept of computability. Then principles for computability are stated, seemingly justified by appealing to the word of these great masters. Perhaps the paper succeeds at a theory of how these logicians conceive of computability, but that is not the same as getting to the essence of computability itself.</p>
<p>The fact that <em>we still do not understand computability</em> is poignantly illustrated by the existence of a 2020 paper by Matteo De Benedetto, “Explication as a Three-Step Procedure: the case of the Church-Turing Thesis”. In this paper, De Benedetto compares the two recent approaches to explicating computability, that of Sieg (he went on to do a lot more work on computability after his streamlining of Gandy’s work) and that of Dershowitz and Gurevich. To do this, he interrogates the concept of explication itself, arguing that Carnap’s “two-step” picture of explication is not sufficient to understand the differences between these two conceptions of computability and that really we need a “three-step” picture. That such an involved meta-analysis is needed casts doubt that these authors are on the right track whatsoever.</p>
<p>Let us begin our own investigation.</p>
<p>Firstly, we center the process of computation over the property of computability. Looking at “the class of computable functions” and attempting to analyze its properties is tantamount to a project to understand 3-dimensional figures solely by looking at their 2-dimensional projections. The essential fact that makes computable functions computable is the process of computation, so this process is what we must understand, not merely its result.</p>
<p>Historically, this centering of the process of computation was what made Turing’s 1936 paper “On Computable Numbers”, in which he introduces the Turing machine, such a breakthrough. As Turing wrote in 1937, “The identification of ‘effectively calculable’ functions with computable functions is possibly more convincing than an identification with the <img src="https://latex.codecogs.com/png.latex?%5Clambda">-definable or general recursive functions.” Church agreed, writing in his review of Turing’s paper,</p>
<blockquote class="blockquote">
<p><em>As a matter of fact, there is … equivalence of three</em> <em>different notions: computability by a Turing machine,</em> <em>general recursiveness in the sense of</em> <em>Herbrand–Gödel–Kleene, and <img src="https://latex.codecogs.com/png.latex?%5Clambda">-definability in</em> <em>the sense of Kleene and [myself]. Of these, the first has</em> <em>the advantage of making the identification with</em> <em>effectiveness in the ordinary (not explicitly defined)</em> <em>sense evident immediately … The second and third have</em> <em>the advantage of suitability for embodiment in a system of</em> <em>symbolic logic.</em></p>
</blockquote>
<p>Gödel too was convinced by Turing’s formulation. Kleene wrote:</p>
<blockquote class="blockquote">
<p><em>According to a November 29, 1935, letter from Church to me,</em> <em>Gödel ‘regarded as thoroughly unsatisfactory’ Church’s</em> <em>proposal to use <img src="https://latex.codecogs.com/png.latex?%5Clambda">-definability as a definition of</em> <em>effective calculability … It seems that only after</em> <em>Turing’s formulation appeared did Gödel accept Church’s</em> <em>thesis.</em></p>
</blockquote>
<p>[all three quotes from The Essential Turing ed.&nbsp;B. Jack Copeland, pp.&nbsp;44-45]</p>
<p>Hence we deny that <a href="https://golem.ph.utexas.edu/category/2019/08/turing_categories.html">“Turing categories”</a> get to the heart of computation, since they lack an account of the process of computation, but merely describe general recursiveness. We must instead describe the process of computation.</p>
<p>What made Turing machines in particular so compelling to Turing, Church, and Gödel as an explication of “effective calculation”? I find Martin Davis’ 1978 paper “What is a computation?” very elucidating on this point (Davis actually describes a Post-Turing machine, which is equivalent to a Turing machine). He writes (p.&nbsp;243):</p>
<blockquote class="blockquote">
<p><em>Turing based his precise definition of computation on an</em> <em>analysis of what a human being actually does when he computes.</em> <em>Such a person is following a set of rules which must be carried</em> <em>out in a completely mechanical manner. Ingenuity may well be</em> <em>involved in setting up these rules so that a computation may be</em> <em>carried out efficiently, but once the rules are laid down, they</em> <em>must be carried out in a mercilessly exact way. If we watch a</em> <em>human being calculating something (whether he is carrying out a</em> <em>long division, performing an algebraic manipulation, or doing a</em> <em>calculus problem), we observe symbols being written, say on a</em> <em>piece of paper, and the behavior of the person doing the</em> <em>calculating changing as he notes various specific symbols</em> <em>appearing as results of computation steps.</em></p>
</blockquote>
<blockquote class="blockquote">
<p><em>The problem which Turing faced and solved was this: how can one</em> <em>extract from this process what is essential and eliminate what</em> <em>is irrelevant? Of course some things are clearly irrelevant;</em> <em>obviously it does not matter whether our calculator is or is</em> <em>not drinking coffee as he works, whether he is using pencil or</em> <em>pen, or whether his paper is lined, unlined, or quadruled.</em> <em>Turing’s method was to introduce a series of restrictions on</em> <em>the calculator’s behavior, each of which could clearly be seen</em> <em>to be inessential. However, when he was done all that was left</em> <em>were a few very simple basic steps performed over and over</em> <em>again many times.</em></p>
</blockquote>
<p>Davis goes on to argue that without loss of computational power, the calculator could be restricted to write on a 1-dimensional tape rather than a 2-dimensional sheet of paper; they could be restricted to only use the symbols 0 and 1; and they could be restricted to only look at one square of the tape at a time. Furthermore, their possible actions could be restricted to changing the symbol on the tape and shifting their attention to an adjoining square. With these restrictions, the calculator then follows a finite list of instructions telling them what action they must take, given the symbol on the square they are looking at. To see this whole argument, I recommend reading Davis’ paper.</p>
<p>So Turing machines (or, Post-Turing machines) seem to essentialize the process of human computation, that is, a human following a “mercilessly exact” procedure for calculation. Each successive restriction Davis puts on this human does not reduce their computational power, although they may make the computation a great deal more arduous (imagine trying to multiply two numbers in binary on a 1-dimensional tape, where you can only look at one square at a time! You could do it but it wouldn’t be fun). The insight of Turing which so resonated with the logicians of his day was to put the human into the analysis. It is not enough to just consider which functions are effectively calculable (i.e.&nbsp;computable), says Turing, we must consider the process by which a human calculates them.</p>
<p>From Turing machines all computations can be executed, just as from set theory all mathematics can be constructed. But as category theorists, we know that a particular set-theoretic implementation of a concept does not qualify as the essence of the concept, since it could just as easily be implemented infinitely many other ways. The true essence of a concept is its internal structure, and how it relates to other concepts (its “external structure”, perhaps). Thus from the categorical viewpoint, Turing machines do not succeed in essentializing computation any more than von Neumann ordinals essentialize the naturals, the Sheffer stroke essentializes Boolean algebra (you can get all operations from it! :P), or a particular basis of a vector space essentializes that vector space. Category theory agrees with Davis’ question: “how can one extract from this process what is essential and eliminate what is irrelevant?” But then it departs from his answer, as the particular choices he makes of how to restrict the calculator are introductions rather than eliminations of irrelevancy. What is the categorical essence of computation?</p>
<p>Let us start from Turing’s basic insight, that to understand computation we must understand it as a process involving a human. And not just any process, but one in which the human carries out a “mercilessly exact” procedure, i.e.&nbsp;an algorithm. So now we are led to the concept of “algorithm”. If we can understand this concept formally, as well as what it means to “execute” an algorithm, then we can rest easy that computation is simply the execution of an algorithm. (Needless to say, the fact that the human writes symbols on paper is inessential. They could just as well use an abacus, a clay tablet, or do everything in their head.)</p>
<p>In my next post on this topic, I will consider the question “What is an Algorithm?”</p>


</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>category theory</category>
  <guid>https://topos.institute/blog/2022-08-10-computation-category-theory/</guid>
  <pubDate>Wed, 10 Aug 2022 00:00:00 GMT</pubDate>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>Brendan Fong</dc:creator>
  <link>https://topos.institute/blog/2022-06-07-all-models-are-wrong/</link>
  <description><![CDATA[ 




<div id="post-content" class="page-columns page-full">
<div class="page-columns page-full"><blockquote class="blockquote">
<p><em>“All models are wrong, but some are useful”</em> — George Box<sup>1</sup></p>
</blockquote><div class="no-row-height column-margin column-container"><div id="fn1"><p><sup>1</sup>&nbsp;Box, G. E. P. (1979), “Robustness in the strategy of scientific model building”, in Launer, R. L.; Wilkinson, G. N. (eds.), Robustness in Statistics, Academic Press, pp.&nbsp;201–236, doi:<a href="https://doi.org/10.1016/B978-0-12-438150-6.50018-2">10.101/B978-0-12-438150-6.50018-2</a>, ISBN 9781483263366.</p></div></div></div>
<p>What is the purpose of a mathematical model? The above, well-known quote comes from a paper of statistician George Box, who makes the point that models never completely predict reality. As an example, he takes the ideal gas law, relating the pressure, volume, and temperature of a gaseous body. This ‘law’, he argues, is only a statistical approximation, and in fact reality diverges in a variety of subtle ways. Nonetheless, it is useful — it’s a close enough approximation that we can use it for precision engineering. It helps us make espresso machines and surgical equipment and space shuttles! In this view, while no model is perfect, a good model is one that extrapolates well to new data.</p>
<p>In collaboration with teams led by epidemiologist and public health expert Professor Nate Osgood and computer scientist James Fairbanks, Topos scientists Evan Patterson, Sophie Libkind, and John Baez recently published two papers on rapid epidemic modelling. They’re really lovely papers, backed by software and simulations, which use compositional ideas from category theory to show how models can be rapidly constructed from high level descriptions. Assisted by additional discussions with Nate, Evan, and Sophie, this paper has highlighted what is for me a new perspective on mathematical modelling, and one I believe very much reflects the spirit of our mission at Topos. In their view, a good mathematical model is one that helps <em>diverse parties agree on the likely future consequences of their actions</em>.</p>
<p>To dive straight into the papers, check them out here:</p>
<ul>
<li><em>Compositional Modeling with Stock and Flow Diagrams</em>. John Baez, Xiaoyan Li, Sophie Libkind, Nathaniel Osgood, and Evan Patterson. <a href="https://arxiv.org/abs/2205.08373">arXiv:2205.08373</a></li>
<li><em>An Algebraic Framework for Structured Epidemic Modeling</em>. Sophie Libkind, Andrew Baas, Micah Halter, Evan Patterson, and James Fairbanks. <a href="https://arxiv.org/abs/2203.16345">arXiv:2203.16345</a></li>
</ul>
<p>Mathematical modelling has played a central, public role in the COVID-19 pandemic. Especially in the uncertainty of the early pandemic, mathematical models were one way to access a bit more information about the possible future. How many deaths will COVID cause? When will this wave end? When can life get back to ‘normal’? Yet these models are unlike our ideal gas law in a key sense: these questions were (and continue to be) dependent on our behaviour as individuals and as a society. When a pandemic model predicts an unacceptable burden of death and disease, we have the power to make it ‘wrong’, to avert this potential future becoming our reality. In this case, a model might be excellent precisely because its predictions completely diverge from future reality — the excellence lies in the fact that the model galvanizes changes in policy and behaviour to save lives.</p>
<p>That is: a model can be excellent <em>because</em> it becomes wrong!</p>
<p>This different view of “what is a good model” implies different goals for good modelling methodology. For example, models that predict complex phenomena precisely are often necessarily themselves complex. On the other hand, complexity makes models harder to understand, and therefore less accessible to those without a vast amount of technical experience. Nonetheless, it’s important that such people, which can include influential decision makers, let alone the wider public, feel comfortable trusting models that they use to make decisions. Without this trust, it’s harder to find agreement, and cooperate towards a future everyone believes in. Thus a good modelling methodology must consider the balance between precision and transparency.</p>
<p>In these papers, the authors develop a number of techniques to increase transparency of models. First, they use high level models, in terms of descriptions of concepts people can relate to, such as susceptible and infective populations and encounters between them, rather than using anonymous variables like ‘p’ and ‘q’ that change according to a differential equation.</p>
<p>Second, they use visual models. This increases accessibility of the models, giving all parties a sense of what they mean without requiring the ability to, for example, read and interpret differential equations. This is especially important when various stakeholders in the decision-making process, like public policy makers, or even the general public, have not necessarily had years of formal training in mathematical modelling. Note that in this case the diagrams have formal meaning: the diagrams are not merely depictions of a mathematical model, they are themselves the model!</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="disease-transition-model.png" class="lightbox" data-gallery="quarto-lightbox-gallery-1" title="A visual model of disease transmission, clearly depicting the relevant factors in the model and their interactions"><img src="https://topos.institute/blog/2022-06-07-all-models-are-wrong/disease-transition-model.png" class="img-fluid figure-img" alt="A visual model of disease transmission, clearly depicting the relevant factors in the model and their interactions"></a></p>
<figcaption>A visual model of disease transmission, clearly depicting the relevant factors in the model and their interactions</figcaption>
</figure>
</div>
<p>Third, they emphasise the <em>compositionality</em> of models: building up models from simpler parts. Complex phenomena like pandemics involve many distinct parts. The expertise to understand each part is often held by a different person: epidemiologists, geneticists, economists, behavioural psychologists, and politicians may each hold a distinct piece of the puzzle. Compositional modelling allows each expert to create and verify a model representing their insights into the situation, while having that model feed directly and robustly into the large scale vision. This further increases the chance that the model reliably reflects what each expert thinks, and increases buy-in and trust from each party.</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="composing-models.png" class="lightbox" data-gallery="quarto-lightbox-gallery-2" title="Composing models A, B, and C to produce a global view"><img src="https://topos.institute/blog/2022-06-07-all-models-are-wrong/composing-models.png" class="img-fluid figure-img" alt="Composing models A, B, and C to produce a global view"></a></p>
<figcaption>Composing models A, B, and C to produce a global view</figcaption>
</figure>
</div>
<p>All in all, these features of transparent model construction mean that when the models ultimately print out predictions about the future, it is clearer what assumptions led to these predictions, and hence what options we have, as a society, if we want to secure or avoid such a future.</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><a href="graph.png" class="lightbox" data-gallery="quarto-lightbox-gallery-3" title="A graph predicting various populations, such as those of susceptible, infected, and hospitalised people, over time. The transparency of the modelling techniques used here is critical for building trust in these predictions."><img src="https://topos.institute/blog/2022-06-07-all-models-are-wrong/graph.png" class="img-fluid figure-img" alt="A graph predicting various populations, such as those of susceptible, infected, and hospitalised people, over time. The transparency of the modelling techniques used here is critical for building trust in these predictions."></a></p>
<figcaption>A graph predicting various populations, such as those of susceptible, infected, and hospitalised people, over time. The transparency of the modelling techniques used here is critical for building trust in these predictions.</figcaption>
</figure>
</div>
<p>While a definition of excellence in modelling that just makes reference to data is technically neat and does provide an important measure of quality, at Topos the end goal of our work is always to improve the lives of people. This means that ultimately we believe we must tie the quality of our mathematics and our programming tools to how people interact with it. In this case, our work is successful when it helps people agree, so they can more fully cooperate, and make collective decisions with outcomes that better reflect their hopes and values.</p>



</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>modeling</category>
  <guid>https://topos.institute/blog/2022-06-07-all-models-are-wrong/</guid>
  <pubDate>Tue, 07 Jun 2022 00:00:00 GMT</pubDate>
  <media:content url="https://topos.institute/blog/2022-06-07-all-models-are-wrong/graph.png" medium="image" type="image/png" height="93" width="144"/>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>David Spivak</dc:creator>
  <dc:creator>James Dama</dc:creator>
  <link>https://topos.institute/blog/2022-06-03-account-of-sense-making/</link>
  <description><![CDATA[ 




<div id="post-content">
<p><strong>Summary</strong>: When you say “oh, that makes sense” or “no, that doesn’t make sense”, you’re talking about whether or not the story fits together or adds up right. But you may also say “I have a sense of when people are feeling awkward” or “I seem to use my sense of smell more than other people do”. Could it be that all these notions of <em>sense</em> are actually the same in some deeper way? Could it be that when we “make sense” of things, we’re actually doing the very activity that has “made” all of our senses to date? In this post, we’ll give you a sense of why that might be right!</p>
<p><strong>Background</strong>: UCLA’s Institute for Pure and Applied Mathematics (IPAM) ran a workshop called <a href="http://www.ipam.ucla.edu/programs/workshops/mathematics-of-intelligences/">Mathematics of Collective Intelligence</a> this winter, where I (David) gave a talk called <a href="https://www.youtube.com/watch?v=1amiYp72GFE">Sense-making: Accounting for Intelligibility</a>. Some of the ideas in the talk were mathematical, but the story arc was philosophical, and basically speculative. Recently, I was talking with my colleague <a href="https://scholar.google.com/citations?user=KtQzmZoAAAAJ&amp;hl=en">James Dama</a>, and he was able to give a much firmer scientific story, one which goes a long way toward justifying the speculations from the talk. So here we jointly write this post—David on the story arc, and James on the science—attempting to make the idea more widely known, so it can be refined, disputed, affirmed, etc.</p>
<p><strong>Motivating story</strong>: Imagine two kids in math class, both of whom are trying hard to get an A, but in markedly different ways. The first one copies whatever the teacher writes on the board, making sure to have a faithful copy in their notebook. The second one does what we’ll call <em>sense-making</em>; let’s suppose we listen in to the second kid talking; we hear the following exchange:</p>
<blockquote class="blockquote">
<p>“Wait, what?! Why would that be <img src="https://latex.codecogs.com/png.latex?2xy">!? Shouldn’t it be <img src="https://latex.codecogs.com/png.latex?x+y">? I mean, why would… I mean, if <img src="https://latex.codecogs.com/png.latex?x"> was <img src="https://latex.codecogs.com/png.latex?4"> and <img src="https://latex.codecogs.com/png.latex?y"> was, I don’t know, <img src="https://latex.codecogs.com/png.latex?2">, or wait… if <img src="https://latex.codecogs.com/png.latex?x"> was <img src="https://latex.codecogs.com/png.latex?2"> and <img src="https://latex.codecogs.com/png.latex?y"> was <img src="https://latex.codecogs.com/png.latex?-2">… Oh, but then….. Oh!!!!! I seeeee… right, because of the distrib-whatever rule you just said! Ok that makes sense now; phew.”</p>
</blockquote>
<p>How do we account for all that activity?? What is the student doing, and why does it have that frustrated, searching flavor? And what does it mean when the student says that it “makes sense now”?</p>
<p>Imagine that the next day, when the teacher asks a question, the first student looks through their notes to find a similar example, but before they finish, the second student raises their hand excitedly and knows the answer when called on. Later, the teacher is explaining something and makes a typo; the second student shouts out “that should be <img src="https://latex.codecogs.com/png.latex?3x">, right?” and the teacher says “right”. The first student must erase the “<img src="https://latex.codecogs.com/png.latex?2x">” in their notes, while the second student is thinking about what this whole thing is about; why are we learning this? When it comes time to take an exam, the second student does much better.</p>
<p><strong>Point of the story:</strong> The second student—the one who did the thing we call “making sense”—actually developed a new ability to track the relevant aspects of the mathematical situation, as judged by their newfound ease in vibing with the math, their success on a test, etc; in other words they developed a <em>sense</em> of the math. Moreover, the sense-making activity had the character of trying to <em>account</em> for what they saw. The frustration at the beginning was coming from the fact that things weren’t fitting together, the accounts weren’t settling. The relief arrived when the “accounts settled”, when everything fit into a simpler synthesis.</p>
<p><strong>Question 1:</strong> The work we call “sense-making” in math class seemed to <em>make</em> a new <em>sense</em> for how to work successfully with the math. Is it plausible that all of our senses—our sense of danger, our sense of eyesight—were created through a process that is structurally analogous to what the math student did to get their sense of the math?</p>
<p><strong>Question 2</strong>: The way the good math student <em>made sense</em> was by trying to <em>account for</em> what was present, namely the symbolic statements the teacher was making. Is it plausible that all sense-making—the way all of our senses were made—took place using a similar sort of accounting process, i.e.&nbsp;by working to structure the present situation according to a more broadly-applicable system?</p>
<p>I’ll return to narrating this blog post in the last section, but up next James will show, using scientifically-grounded stories, that both of the above questions can be answered affirmatively, i.e.&nbsp;“yes: it’s plausible that all of our senses were made by a process analogous to the good math student’s sense-making work, and yes: it’s plausible that this work involves accounting for present variety within a uniform system. To explain what it is that is turning into sense, James adds a notion that he calls <em>sensitivity</em>, which comes before sense is formed. Anyway, here’s James.</p>
<hr>
<p>David had asked about our sense of sight and the development of the eye as a case of sense-making, and so I (James) told a plausible scientific just-so story about how the development of visual senses in biology might look like the same sense-making the students were doing. Here is how it went:</p>
<p>First, David’s stories both start with a sensitivity without sense. Both the students cared about their grades though they didn’t know how exactly to earn the grades. Analogously, life is made out of biomolecules, especially DNA and RNA, that are sensitive to light from the very beginning. Light can damage those molecules and cause mutations, but this fact is originally not well dealt-with by early life.</p>
<p>Second, the stories describe a phase of adapting to the sensitivity, but still without real sense. The students learn particular answers to particular questions, but one is not attempting to understand their connections and the other is trying but not yet understanding. In pre-visual biology, there are many examples of local adaptations to react to light exposure without coordinating the reactions together into a single sense. Organisms developed with simple light detectors in their cells—allowing the cells to react to light—before they developed the coordinating intercellular communication to organize all the separate detections into organism-level responses.</p>
<p>Third, the stories describe a phase of synthesizing the adaptations into a sense that accounts for the sensitivity. The student who learns the principles of the math problems finds a way to coordinate all the mathematical ideas into a coherent whole, and also better understands mathematical meaning itself as the sense behind the grades. In biology, visual sense seems to arise from the coordination of local cellular responses to light into coherent organism-level patterns. Each of the local detectors evolves to couple to intercellular communication networks, such as the nerves and their precursors, and the nerves evolve to coordinate whole-organism responses to the light fields falling on the whole organism.</p>
<p>Finally, the sense separates from and supersedes the original sensitivity. Once a student has learned the principles of mathematics, it’s normal for them to stop trying to memorize each of the problems that they’ve seen before (and to stop prioritizing grades when deciding which math to learn next). Once the organism has evolved to respond to light-like stimuli as a whole organism, specialized new light-sensitive organs like eyes can evolve to take over the responsibility for triggering those behaviors. At that point it’s unambiguous that the organisms have evolved to make sense of light. The old distributed light detectors might become vestigial and disappear, and the behaviors that the new eyes govern, like collision avoidance or target seeking, often have very little to do with the early DNA and RNA sensitivity to light.</p>
<p>All put together, the story goes like this:</p>
<ol type="1">
<li>An agent has passive sensitivities.</li>
<li>The agent adapts to act sensitively but incoherently.</li>
<li>The agent adapts to coordinate its sensitive activity into coherent actions.</li>
<li>The agent’s coordination of its sensitive actions constitutes a sense.</li>
</ol>
<p>The new sense begins as a system for coordinating coherent actions to respond to the original sensitivities, but it can then become much more.</p>
<p>That answers Question 1 directly, but where is Question 2? Well, “accounting” is exactly the name for one of the key communication processes that economic actors use to coordinate their economic behaviors coherently! Without accounting, complex economic activity becomes incoherent and senseless.</p>
<p>Accounting is therefore one of the rare cases of sense-making where we do the relevant coordination consciously and deliberately at a nuts-and-bolts level, and that makes it a great analogy for thinking about the nuts and bolts of the other cases that are more unconscious or automatic. We don’t yet know all the nuts and bolts of neural signaling in the biological vision case, and we don’t yet know the nuts and bolts of mathematical intuition for the students’ case. Accounting makes an unusually common, firm, and clear comparison case for building up understandings of these other, less transparent cases of sense-making.</p>
<hr>
<p><strong>Conclusion</strong>: Hopefully the above gives a reasonable enough account of how “sense-making” as <em>working to understand</em> and “sense-making” as <em>the making/creation of our senses</em> have deep structures in common. If this idea clicked for you, the reader, then you can probably track a commonly-missed aspect of how important the sense-making activity is: by making sense of things together, we are in fact improving humanity’s collective ability to <em>sense</em> the world—i.e.&nbsp;to track the relevant aspects—even as things get more and more complex.</p>
<p><strong>Post script</strong>: In my (David’s) <a href="https://www.youtube.com/watch?v=1amiYp72GFE">IPAM “Sense-making” talk</a>, I had also discussed the common intuition that when someone makes sense of previously-scattered information, it feels like something “clicks”. James and I discussed this as well, and he connected the “click” to the concept of <em>hysteresis</em>—a sudden phase change, where a significant but small amount of energy can push a system into a new equilibrium—and to his academic work on <a href="https://www.proquest.com/openview/04453da831678af281a827a0e9771515/1?pq-origsite=gscholar&amp;cbl=18750">coarse-graining</a> and <a href="https://pubs.acs.org/doi/abs/10.1021/ct500441q">adaptive sampling</a>. I was fascinated to learn that his pursuit of these ideas is rooted in an attempt to understand the causal structures of societal change. This is roughly the root of my interest as well, as I explained a few years ago on <a href="https://johncarlosbaez.wordpress.com/2015/03/27/spivak-part-1/">John Baez’s blog</a>.</p>
<p>Hopefully we (either James and David, or the “we” that includes any interested reader) can one day take some of these ideas and lift them out from statistical physics and philosophy and into a category theoretic formalism. Indeed, category theory is a principled language for creating accounting systems in general—including self-referencing ones such as the category of categories—and hence it appears suitable for articulating a principled account of the life-enabling activity we call sense-making.</p>


</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>informal</category>
  <category>philosophy</category>
  <guid>https://topos.institute/blog/2022-06-03-account-of-sense-making/</guid>
  <pubDate>Fri, 03 Jun 2022 00:00:00 GMT</pubDate>
</item>
<item>
  <title>Blog / Opinion essays</title>
  <dc:creator>David Spivak</dc:creator>
  <link>https://topos.institute/blog/2022-05-18-the-artificial-distinction/</link>
  <description><![CDATA[ 




<div id="post-content">
<p>I’m like, “oh man, people always talk about ‘artificial’ this, ‘artificial’ that, but what’s up with that? I mean aren’t we humans part of nature too? Are fish and monkeys natural in some way that we humans are not? Are our products somehow not natural, whereas termite mounds are natural? This ‘artificial’ thing just feels so wrong, I don’t know what’s being expressed when people use the term. I feel like it doesn’t actually refer to anything I can make sense of. I don’t know, it’s just.. so.. so…. artificial!”</p>
<p>The distinction “artificial” succeeds in referring to something: itself. It’s an artificial distinction, and that’s exactly its strength: from my perspective the <em>concept</em> of artificiality is itself the best case of an artificial thing. It’s a human-made distinction, one that—unlike “sun” or “fast”—no other organism would presumably make (“we squirrels have such <em>artificial</em> behavior!”). When we humans start to think of ourselves as separated from nature, that very thought separates us—at least conceptually—from nature.</p>
<p>But in some sense, the fact that the artificial distinction succeeds seems to cause a separation between people and nature that is even more than conceptual. That is, I’d guess that people who operate more from within the perspective that humans are separated from nature will end up doing more things that we’d call “artificial”. My guess is that people who operate more from within the perspective that humans are a part of the natural world will act in a way that fits more harmoniously within the natural world. In other words, it’s possible that the artificial distinction <em>creates</em> the artificiality we see in the world, the “fake plastic trees” and such.</p>
<p>Understanding the role of artificiality has consequences for the sort of math we do. For example, when someone models a human as an observer or actor with a utility function on “world” states, what are they doing? I ask: is the observer/actor, including the place where all its thinking happens, <em>part of the world</em>? If so, their utility function is embedded in the world and so they must have a utility function over their own set of utility functions. So I suppose their utility function includes a value on each element of the sequence of world states that arise as they carry out the steps of their own Bayesian program? The conversation breaks down if we even try to put math to the idea that this observer/actor is part of the world that’s being so-modeled. That is, for the mathematical model to work, the observer’s thinking—and hence the observer—cannot be part of the world, but for some reason the observer cares deeply about the world! This is nonsense: the model of an observer that’s not part of the world but cares about it is motivationally self-contradictory. There’s a disconnect there. And yet when one starts from the artificial distinction, it all apparently makes sense, because one doesn’t ask if the observer is part of the world: of course not; they’re modeling us humans!</p>
<p>In golf or tennis or baseball, a good shot doesn’t end with the stick hitting the ball; it turns out that you need proper follow through. My worry for the artificial distinction is that, by denying our natural status, all the resulting work will lack a sort of follow through. If we are not part of the world, but we want to affect the world, the disconnect will hamper the success of our work. When a baseball player hits a great shot, we say “whoa! You really connected!!” The sensation of connecting with the ball feels <em>great!</em> But when you flub it, the disconnect causes weird vibrations up the stick that hurt. I claim that artificiality is a disconnect, and that it’s hurting us.</p>
<p>Here’s the thing: how did we humans get our intelligence? How did we get our values? Where do our motivations come from? All of this arose from the push and pull and struggle of various sorts of evolution: biological, social, intellectual, personal. We’re embedded in a natural world that selects those actions which fit together coherently; that’s what “fitness” should mean. It’s not a number, it’s a quality. The anthill fits—makes sense within—its surroundings. When you do something people around you appreciate, and when you appreciate the people you’re with, <em>you’re</em> <em>feeling your own fitness</em> there. Fitness is a quality. It is connection with the surrounding world.</p>
<p>One can be fascinated with the artificial distinction: humans really do seem to be different somehow, and we value our unique abilities and experiences. To me, however, it seems there’s much more power in understanding how we <em>fit</em>: how human intelligence, human values, human motivations arise from the same source—properly construed—as the ant colony does. We’re motivated by dopamine and serotonin, but how did those systems get formed in us? How did we come to enjoy the ripe strawberry, the image of a mother and child looking into each others’ eyes? How did we get our ego and our willingness to kill? How did the squirrel learn to hide nuts and the poison ivy learn to export itch?</p>
<p>The ontology of our world may be such that all these abilities arose from a common structure. It seems to me that somewhere in the center of nature, at all scales, there is learning. Learning doesn’t start in kindergarten, it doesn’t start with learning to crawl or understanding object permanence. My guess is that from the moment of conception, the whole development process is a kind of learning. “Need more filtering! Need more coordination! Need more food!” our cells cry out as the liver, brain, and circulatory system form in us and as us. My guess is that from the first life, the continuously developing ability to leave old niches and create new ones arose from a certain sort of learning, a creative pursuit of harmony, which was both specific to the context and also part of a general system. Rather than, or alongside the question of how humans are different, we can be fascinated with the question: what is this learning at the core of life, out of which we humans and our learning abilities emerged?</p>
<p>I propose that our math and science will be more effective and valuable if we focus less on the artificial distinction and more on our fitness within nature. That way, our math will really <em>connect</em> and lead to a stronger, healthier, more integrated world.</p>


</div>

<script defer="" src="https://comments.topos.institute/comentario.js"></script>

 ]]></description>
  <category>informal</category>
  <guid>https://topos.institute/blog/2022-05-18-the-artificial-distinction/</guid>
  <pubDate>Wed, 18 May 2022 00:00:00 GMT</pubDate>
</item>
</channel>
</rss>
