UKRI Centre of Doctoral Training in Safe and Trusted AI PhD student (the STAI CDT), Matt MacDermott, won the Best Paper Award at TARK 2023 for the paper, ‘On Imperfect Recall in Multi-Agent Influence Diagrams’, with co-authors James Fox, Lewis Hammond, Paul Harrenstein, Alessandro Abate and Michael Wooldridge.
TARK is a bi-annual conference which brings together researchers interested in issues surrounding knowledge and rationality. TARK has an interdisciplinary approach by encouraging contributions from a wide variety of fields including Artificial Intelligence, Cryptography, Distributed Computing, Economics and Game Theory, Linguistics, Logic, Philosophy, and Psychology. This year it took place in June 2023 in Oxford, UK.
Matt explains the paper here, “The paper is about multi-agent influence diagrams (MAIDs), a popular game-theoretic model which has been used as the basis of various pieces of theoretical AI safety work, for example formally defining fairness and analysing the incentives of content recommender systems. The underlying theory of MAIDs isn’t as well-developed as that of some other game-theoretic models, so in this paper we tried to mitigate that by introducing imperfect recall, mixed policies, correlated equilibria, and complexity results – all standard ideas in game theory, but ones which don’t usually come up in MAIDs”.
Matt really enjoyed TARK 2023 and hopes to attend again in 2025. He said, “TARK was a really fun multidisciplinary conference. We were pretty surprised to get the best paper award, especially since the runner-up paper (Sequential Language-based Decisions by Adam Bjorndahl and Joe Y Halpern) was so great”.
Congratulations to Matt and his co-authors James Fox, Lewis Hammond, Paul Harrenstein, Alessandro Abate and Michael Wooldridge. You can read their paper here: paper.cgi (unsw.edu.au)