I’m a mathematician / computer scientist working on the intersections between artificial and collective intelligence. For more details, jump to my research page.
An ordinary evening on the Brooklyn Bridge
This essay was first drafted on Nov. 20, when Anthropic mathematician Levent Alpöge posted a formal proof of the Riemann Hypothesis, perhaps the most important open problem in mathematics.
Today I walked the Brooklyn Bridge, and as I walked I felt a cool tongue of perfume and exhaust waft up from the river, from a girl who laughing passed me by. Today, an AI proved the Riemann Hypothesis.
I read this news on the Brooklyn side of the bridge, Manhattan still in the distance, and I remembered wistfully that one day long ago I dreamed of building an AI to do something similar, to write a proof that I knew I could never write myself.
That the Riemann Hypothesis is true—perhaps a relief. That an AI solved it, in code—well, if I had been a better mathematician I would have tried to work this way too, to suggest tactics and write proofs in collaboration with a machine.
I cannot read Claude’s proof, and even if I could read Lean I could hardly comment on its beauty or clarity or lack thereof. It is too big and too real. I only remember my complex analysis professor speaking of the critical line, of the music of the primes like jazz, transparencies of disorder so delicately constrained within order. I wonder where is the beauty of a proof I cannot read.
And as Manhattan rises before me, the Verizons and the Merrills and the many banks along the Hudson, I think to myself what’s the matter, that in San Francisco a brilliant man and his yet more brilliant helper have struck the bell of a new era?
The evening light ebbs down the waterfront. I walk and walk and walk. Another mathematician, I read, compares Claude to a telescope. More and more will be uncovered, he says. It is not the end of mathematics, or of mathematicians.
What’s the matter then? A man perched on the steel fence of the walkway looks up from his phone and almost meets my eye. What’s the matter with this industrialization of mathematics? What are we afraid of? That math might become like big physics, with a few ensconced in rich labs with their tools and helpers while the remainder retreat to theory?—well, there’s nothing wrong per se with big physics, and it is not as if math does not have its elites. That even ‘theory’ is no recourse for pure math, and those without resources will not be mathematicians but philosophers?—well, there’s also nothing wrong with philosophy per se, though it is not math.
What are we afraid of? Of being scooped?—we were always afraid of being scooped. Perhaps it is that even those mathematicians with resources will be reduced to a bitter remainder, solving mainly for how many tokens it will take to generate a proof.
My old supervisor Gromov once said to me, a mathematician is defined by two things: freedom, because there are so many problems to work on and you can choose to work on any one of them, and desperation, because there are so many problems to work on and so you should look everywhere you can for help.
God, I am tired, and I need help. The day grows long, the ferries are heading home, and just how many problems do I need to solve? And how much money will I need to solve them? I do not like this grubby, bitter calculation. This is not what I want to work on, these hammered divisions of divisions of value, rummaging through a thin rump of problems left over by the trillion-dollar companies. I would rather dream of an experimental mathematics, a new Galilean age where we take like tall-sailed ships across the amber sea, spyglasses out in search of continents.
What are we afraid of? Perhaps it is that mathematics, which once felt so large, suddenly feels so small. The sun floods the horizon, merging into the far piers out west in Jersey. As I walk I wonder, is the beauty of a proof that somewhere within it is another question? Then where is the beauty of a proof I cannot question, I wonder, as I walk off the bridge and rejoin the city, as I walk into the people, the gorgeous people of this city, young and wrinkled white and brown veiled and bare, children red hearted for New York.

What’s a co-mathematician?
Note: LinkedIn wouldn’t let me write more than 2000 characters in my old “co-mathematician” job description, so I’m posting it here instead.
Now, it is commonly known (from Erdős) that a mathematician is a device for turning coffee into theorems, i.e. Mathematician : Coffee → Theorems. But in category theory, every morphism f : X → Y has a dual, co(f), which is a morphism co(f) : co(Y) → co(X) in the opposite category (more commonly written f^op : Y → X, letting co(Y) = Y). Therefore, a co(Mathematician) must be a map from co(Theorems) → co(Coffee). Since coffee is involutive—co(co(X)) = X—we have Co-mathematician : Cotheorems → Ffee.
Now, by the Curry-Howard-Lambek correspondence, a proposition is just an object in a category where the morphisms are proofs. Such categories often have a terminal object, True, such that for all objects X there is a unique morphism X → True, and an initial object, False, such that for all objects X there is a unique morphism False → X. Since co(True) is isomorphic to False in the opposite category, and a theorem A is an object X with morphism A : True → X, a cotheorem co(A) should just be an object co(X) with a morphism co(A) : co(X) → False—in other words, a false proposition. Supposing that f is idempotent, i.e. f = f∘f, we thus have Co-mathematician : Falsehoods → Fee.
In other words, a co-mathematician turns lies into money.
Trust me, this is a very funny joke in category theory.
P.S. It’s not necessary for the job description, but arguably co-mathematicians should live inside a *-autonomous category, i.e. the category for linear logic, where resources are constrained. 🤔 This means that a co-mathematician *spends* lies (as a mathematician consumes coffee; once consumed the coffee is gone) in order to make money. Perhaps that’s reasonable; a falsehood, once used, cannot be used again.
P.P.S. If we think of co(-) as the typical involution operator * in a *-autonomous category, then co(Y) = Y* is usually interpreted as the set of obstructions to Y, i.e. the refutations or counter-examples to Y, the demonstrations that Y is false. This feels like a more natural interpretation of cotheorem, and it suggests that a co-mathematician is a device that removes obstructions in order to make money. So give me the pesky counter-example to that theorem you want to prove, and I’ll make it disappear… for a price 😉
Why we don’t need an IPCC for AI
N.b. I wrote an early version of this post in November 2023; I substantially revised it in March 2024.
The Intergovernmental Panel on Climate Change is a body of scientists charged by the UN to provide authoritative scientific reports on the status of climate change, especially to governments. A number of writers have recently called for an IPCC for AI (Suleyman et al., 2023, Mulgan et al. 2023, Maihe 2018). Full disclosure: I contributed to one of these calls. These calls tend to focus on AI safety, and their general premise is that:
- AI poses a safety risk,
- governments need to act,
- to act, governments need a neutral, scientific assessment of the state of AI, and
- an IPCC-like solution can best offer that assessment.
What follows are some notes on why the IPCC is not the right institutional solution for assessing the state of AI. Effectively, IPCC : climate :: IPAIS : AI is an incomplete analogy and thus suggests an incomplete institutional response.
Continue readingThe constitution of AI
The reason for putting artificial intelligence (AI) and governance together in the first place is due to an intuition I had very early on in 2011 as a student of robotics: that there was no mathematical eureka that would “solve” intelligence, that the problem was simply too big, and that this meant that AI had to be assembled piece-by-painstaking-piece [1] But neither was AI a monolithic engineering project, directed by a technical pharaoh. There were too many people with different ideas, too many scientific and technical unknowns, and too many simultaneous research projects and paradigms. It was not even clear to me that a centralized research program was preferable even if one were possible. I believed that building and training an AI was a coordination problem involving millions of people, and I wanted an architecture to solve—or even to define—that coordination problem [2].
To build that architecture, I needed to go beyond AI (at least, as it is practiced today). In the field of AI, the usual meaning of “architecture” refers to technical architectures—software architectures and languages like Prolog, SOAR, subsumption, and ROS, but also hardware architectures like PCs, mobile phones, GPUs, the Raspberry Pi, and the iRobot Baxter—all of which reduce the cost of building and running interesting programs and robots. These mechanisms provide convenient design and programming abstractions to engineers, and they organize the AI’s task by enacting certain knowledge representations (KR). Neural networks are an architecture in this sense, one with particularly nice properties (e.g. modularity, scalability). Technical architectures also often have the effect of making it easier for one programmer to build on the work of another programmer, e.g. by reducing communication and transaction costs, though that was rarely their explicit purpose. In practice, technical architectures (even so-called hybrid architectures) often siloed researchers within competing languages, platforms, and KRs, making it more difficult for people to work across their separate domains.
Continue readingThe failures of category theory
There were many, many amazing discussions at the just-finished Applied Category Theory workshop in Leiden, but my favorites were two that discussed the need for an “approximate” category theory, especially in applications to the sciences.
One of the first things we realized (even if it took about an hour to figure out) was that we were really talking about three different problems:
A comparison of category theory software
The following table compares general and technical information for a number of existing and hypothetical category theory software packages. Drawn from the discussion (ft. Jamie Vicary) on May 1, 2018 at the Applied Category Theory workshop in Leiden.
| Software package | Does it exist? | Author | Goal | User interaction | Automation | How to define categories | Categorical setting | Structures it can compute | Architecture | Directed? (i.e. categories vs. groupoids) |
|---|---|---|---|---|---|---|---|---|---|---|
| HoTT libraries in Cog or Agda | Yes | Various | Research | Code | Proof assistant | Via presentations | Higher (weak) | Limits, colimits, functors | Library | No |
| Opetopic | Yes | Eric Finster | Research | Geometrically | None | Via presentations | Higher (weak) | Library | Yes | |
| Quantomatic | Yes | Aleks Kissinger et al. | Research, scale, interoperability (?) | Geometrically | Proof assistant | Via presentations | Strict, symmetric monoidal | Standalone | No | |
| Algebraic Query Language (AQL) | Yes | Ryan Wisnesky, David Spivak | Commercial, teaching | Code | Query optimizer, data migration | Explicitly, and via presentations | Monoidal | Limits, colimits, functors | Standalone + library | Yes |
| Globular | Yes | Jamie Vicary et al. | Research, publishing | Geometrically | None | Via presentations | Higher (semi-strict) | Standalone | Yes | |
| TikZit | Yes | Aleks Kissinger et al. | Publishing | LaTeX | None | Library | ||||
| Typedefs | Under development | Jelle Herold et al. | Commercial | Code | ||||||
| Proto-Quipper-M | Yes | Francisco Rios, Peter Selinger | Research | |||||||
| Rholang | Under development | Mike Stay et al. | Commercial | |||||||
| EASIK | Yes | Bob Rosebrugh et al. | Research | Geometrically | ||||||
| Cateno | Yes | Jason Morton | Research | |||||||
| PySheaf | Yes | Michael Robinson | Research | Code | ||||||
| Specware | Yes | Kestrel Institute | Commercial | |||||||
| Catlab | Under development | Evan Patterson | ||||||||
| OICOS | Under development | Viktor Winschel, Philipp Zahn | Commercial | |||||||
| Statebox | Under development | Jelle Herold et al. | Commercial | |||||||
| TikzWD | Yes | Patrick Schulz, David Spivak | Publishing | LaTeX | ||||||
| DSL for Operads | No | TBD | Simulation | Geometrically | None | Via presentations | Operads | Standalone | ||
| Common File Format for Categorical Constructions | No | TBD | Infrastructure | Code | Type-checking | Explicitly, or via presentations | All | File format |
The last two items in the table are “wish list” items identified by members of the ACT community.
How to have a conversation about AI
I tend to have a lot of “hot-topic” dinner conversations with people about AI: will robots take our jobs, is software intelligence going to take over the world, and the near-term impacts of big data on everything from science to ecology to law. And it’s not just me: consider all the recent symposia about “the end of work“, the “AI race“, “how to stay human in a robot society“, etc.
While not necessarily shallow, these conversations are invariably speculative. I can’t really respond to people’s concerns about AI because the questions they ask don’t connect with the technical concepts that I study. Talking about AI, for most non-technical people, is a proxy for talking about the place of people in society, present and future. They characterize AI by a set of external variables: how it will displace jobs, how it will make things cheaper and faster in their life/work, how it could make society more or less fair. These are observations one can make without knowing anything about AI, which is why I call them “external”. On the other hand, I study internal variables (a.k.a. technological variables) like error bounds on particular learning algorithms, logical programming, and a slew of engineering problems like motion planning, natural language, and domain generalization. Studying these things does not make me obviously qualified to address social-sciency concerns about the place of people. For the same reason, I’m quite skeptical of AI “experts” when they prognosticate about the impact of AI on society.
Still, there must be something that the internal variables in AI can say about the external variables—but how to say it? Relating the two sets of variables would clarify non-technical people’s concerns to technical people, measure technical developments in terms of non-technical outcomes, and suggest “internal” solutions to their “external” concerns (or prove the absence of such solutions). And maybe, just maybe, it would help all of us have better conversations about AI.
The idea of an experiment
Rob Spekkens is a theoretical physicist who works on the foundations of quantum mechanics, which means that he thinks a lot about the meaning of experiments in quantum mechanics: what experiments are for, their relationship to theory, and how to build better ones. In a recent talk (June 2017, ETH Zurich) he distinguished between the three best reasons for doing experiments in physics:
- don’t know what we’ll find (e.g. in cosmology)
- adjudicate between competing theories
- identify phenomena that resist explanation in current theoretical paradigm
and two less good (but still valid) reasons:
- doing it improves our own understanding
- helps us develop technology based on the theory
I understand Rob’s reason for distinguishing the two sets of reasons (he thinks many experiments in quantum physics are unnecessary, since all they do is confirm the standard theory), but for the purposes of this post I won’t distinguish between the two sets. A gold-standard physics experiment should satisfy all five reasons, and they all reflect some pragmatic aspect of “what an experiment is”.
Speaking of pragmatics, Rob also has a wonderful little diagram illustrating how the different aspects of physics—realist, empiricist, and pragmatist—interact in quantum mechanics. There’s the realists, dominated by theoreticians, who construct all those pretty interpretations of physics and claim, “this is how the real world works”. There’s the empiricists, who analyze the data, come up with all the probability tables and say “whatever ‘real’ means, these are the probability tables that work; when you do A, you get B back (this percent of the time)”. And finally there’s the pragmatists: whatever the probability tables or Hilbert spaces or diagrams are telling me, they should help me propose a concrete experiment or solve a concrete engineering problem.

How different traditions in physics interact to produce (the field of) physics. Courtesy of Rob Spekkens.
I love this diagram! First, let’s recognize what it is: it is a structural, pragmatic account of physics research. It’s structural because it decomposes physics into these three different traditions, and tells us about their interaction. It’s pragmatic because it talks about how to understand and build actual experiments, and because it has a giant atomic symbol to represent all the other practical stuff! To be clear, I don’t think Rob would say that only pragmatists design experiments, only that experiments in the lab are ultimately ground in complex stuff like lasers, beam splitters, and measurement devices that go beep; “axiomatization from pragmatic principles” means taking the complex stuff in the lab and operationalizing it into abstract stuff like probability measures, unitary operators, and even “observers”. In return, a list of the abstract stuff can be converted directly into laboratory procedures.
In this post (adapted from a recent talk I gave at the Rethinking Workshop), I’d like to spend some time thinking about the following question: how can we build a “higher-order” model not only of the physical theories but of the physical experiments which test those theories, so that we can ground out “physical interpretations of theories” (e.g. interpretations of quantum mechanics) in terms of their pragmatics, i.e. the experiments they suggest?
(This post is currently in progress!)
Base change and entropy
Tom Leinster recently posed an interesting question in a talk at CLAP: “how do I generalize a theorem about objects into a theorem about maps?” The general idea comes from Grothendieck’s relative point of view, and to implement this point of view, one has to overcome certain technical hurdles related to “base change.” I thought I’d spend some time trying to lay out what it means to have a change of basis in algebraic geometry, and then how that idea shows up in Tom’s project: turning entropy into a functor.
You can read about Tom’s project (joint with John Baez and Tobias Fritz) directly here: https://ncatlab.org/johnbaez/show/Entropy+as+a+functor
(Currently writing this up, so excuse the notes below!)
The 57th Venice Biennale
For this year’s German pavilion, Anne Imhof designed a 3-foot raised platform of plexiglass that spanned the entire pavilion. If contemporary art these days is all about “creating space” (though I’m not sure how seriously I should take the gallery text when it claims that the nearby pasty smear is supposed to “create space”), then Imhof has stolen the show—the stage design was visceral, impressive, and, more than that, fun.
Of course, then she had to ruin the set by staging a performance underneath it.

Anne Imhof’s Faust (2017). So many alcohol fires, so little time. Perhaps all the 20-something performance artists could have been replaced by trained acrobats.
On the day of my visit, the pavilion was overcrowded with sightseers. But that too was a function of the set design, or at least of the constraints of the original pavilion. Upon visiting all that I saw were other sightseers rolling like pinballs from one end of the room to the other, all but obscuring the (literally) underlying action. But it was pleasant, in a way, to see the pinballs rolling around the corners of the stage, and to meditate a bit on the nature of crowds.
Not all art chooses to create a space; sometimes it is sufficient to represent one. Compare the installation above with Maria Lai’s piece below.
Pictures like these make me ask: just what, exactly, does it mean to represent a space? And how is that different from creating a space?
