Presenter
Jules Hedges
Jules is a mathematician and computer scientist who was a pioneer of the recently developed field of applied category theory. His main scientific interests are in microeconomics and machine learning. He is a co-founder of the Institute for Categorical Cybernetics (https://cybercat.institute), a nonprofit organization for research and open source software development, and 20squares (https://20squares.xyz/), a startup specializing in applied mechanism design and economic modelling in the blockchain space. By day he is a lecturer in computer science at the University of Strathclyde in Glasgow, Scotland.
Summary:
Compositional game theory is an approach to modelling interactions of strategic agents which is designed to be inherently scalable, in contrast to traditional normal-form and extensive-form games that are unsuitable for anything larger than toy problems. It supports a model-is-code paradigm and reuse of model components, inspired by software engineering. I won’t have time to talk much about how compositional game theory works, so instead this will be a taster of what it looks like and what it can do: we have used it in settings such as applied mechanism design, smart contract analysis and the theory of institution design. At the end I’ll hint at a much bigger picture: the mathematical foundations of compositional game theory also appear in several different parts of machine learning and other fields, something we call “categorical cybernetics”, and I’ll discuss what the implications of this could be.
Challenge:
What challenge would you like to see solved for progress in your field?
This is a bit tricky because our actual limiting factor is lack of resources (people and funding), but here’s my best attempt:
“Our biggest issue is one of programming ergonomics. We want to reach a point where domain experts can build and use strategic models as part of their day-to-day work with only moderate programming experience, but despite lots of progress our tools are still too intimidating for our target users. We also have some purely technical problems which are interesting but comparatively less important: for example compositional games that branch on a discrete but non-finite random variable (such as a binomial distribution) involves some extremely difficult mathematics that we only partially understand.”
