Exploring the Relevance of "Hanabi" in the AI Community

In this episode, Nieves Montes, a PhD student at the Artificial Intelligence Research Institute in Barcelona, Spain, delves into the game "Hanabi" and its significance within the AI community. Hanabi is a cooperative game where players collaborate to construct stunning fireworks displays. What sets this game apart is the need for players to share information through a hint system and make inferences about each other's intentions and beliefs.

Nieves Montes, with her research focus on multi-agent systems and values in AI, introduces her agent model called TomAbd. This model combines theory of mind and abduction to reason about other agents' beliefs and make optimal decisions. By utilizing logical reasoning to represent beliefs and explanations, the TomAbd model offers an easily interpretable and understandable approach.

The episode delves into the evaluation of the TomAbd model's performance in Hanabi, with a focus on the final score and communication efficiency. The model demonstrates improved performance compared to other approaches, although there is a computational cost associated with reasoning about other agents' beliefs. However, the TomAbd model proves particularly valuable in domains where theory of mind plays a crucial role, such as cooperative or competitive board games.

Nieves Montes shares her plans for further exploration of the TomAbd model's application in other domains and potential integration of learning approaches to enhance scalability. Additionally, she intends to gain industry experience and graduate before making long-term plans.

Listeners interested in following Nieves Montes' work can find more information on the website of the Artificial Intelligence Research Institute and on Google Scholar.

Nieves Montes, a Ph.D. student at the Artificial Intelligence Research Institute in Barcelona, Spain, joins us. Her PhD research revolves around value-based reasoning in relation to norms. She shares her latest study, Combining theory of mind and abductive reasoning in agent‑oriented programming.