James Goodman

James Goodman

Communicating to co-operate and compete in multi-agent adversarial games

Multi-player strategy games often see the relationship between players shift between co-operation and competition over time as alliances form and break up. AI agents need to plan in an environment containing other agents with unpredictable behaviour, and with whom they must communicate to do well. The aim of this research is to create an ecology of interacting agents that learn to communicate with each other and human players in a consistent fashion. This supports the creation of more immersive environments for players, while simultaneously permitting game designers to configure in-game agents with less time-consuming scripting for increased scalability. Another application is to enable long-running online games such as Diplomacy to smoothly cope with player drop-out, parachuting in a replacement AI that can continue conversations.

This work will initially focus on effective ways to integrate strategic opponent modelling and communication into statistical forward planning algorithms such as Monte Carlo Tree Search (MCTS) and Rolling Horizon Evolutionary Algorithms (RHEA).

 

At various points James has picked up degrees in Chemistry, History, Mathematics, Business Administration and Machine Learning. After a career in Consultancy and IT Project Management of global financial systems he is now finally doing the research he always wanted to. Otherwise found playing tabletop games and writing LARPs.

Home institution: Queen Mary

Supervisor: Professor Simon Lucas

Ready to apply?

Once you have identified your potential supervisor, we would encourage you to contact them to discuss your research proposal.

Learn More