An Interdisciplinary Approach
There have been some fascinating advances in the use of modern computational tools for the study of complex systems, and Shannon information theory has been exceptionally useful for empirical studies of non-linear and causal dynamics. Using information theory I've worked in three main fields: theoretical psychology, artificial intelligence, and economics. In the article "Information theory for agents in artificial intelligence, psychology, and economics" I summarise this perspective, see the graphical abstract below. Below I also mention some projects I'm working on with collaborators and PhD students.
Psychology for Artificial Intelligence
An AI does not 'think', certainly not the way humans do, but can we develop the algorithmic foundations for an AI psychology, like a 'theory of mind' or 'introspection'? How would this help an AI navigate a complex social world and improve its performance? In this work we are working towards developing psychological algorithms that can be implemented in AI's and deployed in real-time.
Non-linear Properties of Game Theory
Game theory, particularly simple games like Prisoner's Dilemma (PD), are often under-estimated in what they can teach about the complexity of making apparently simple decisions. Press and Dyson's Zero-Determinant strategies and iterated PD hint at further hidden complexity, some of which we extend in our current works. Exciting new results to come shortly!
Social Networks for Artificial Intelligence
Humans make extensive use of social networks, for cooperation, coordination, learning, division of labour, and resource sharing. We are also the species with the most complex social structures, but how do we do this and can we do the same thing with AI? Recent advances in our understanding of human social structures could provide evidence for how we can develop AIs to do the same thing for human-AI alignment.
AI, Big Data, and Information Theory for Markets
In this work, with my PhD students we are developing new approaches for the study of non-linear dynamics in financial markets. We're primarily using artificial intelligence and information theory to unpack the complex interactions between assets that produce the non-linear macroscopic dynamics.
We are in a piKL with AI's like Cicero
The recently released AI Cicero that has been able to play very complex multi-player game "Diplomacy" uses an algorithm called piKL which has some intriguing connections with information theory (MaxEnt), games against nature, strategic game theory, and an artificial theory of mind. In this work we are working to connect the dots in this remarkable cognitive landscape, a new manuscript arriving shortly!
Detecting Critical Events in Time Seris
One of the most significant challenges we face in empirical studies is justifying the term "phase transition" when describing something that isn't a physics problem, like neural dynamics, economic dynamics, or financial markets. I've been working on understanding the detectable micro-behaviours of a system before a critical event in order to avoid or mitigate the effects.
