Primary supervisor
Julian Garcia GallegoResearch area
Agents and Decision-MakingCooperation is difficult whenever individual incentives conflict with a shared goal. This problem appears in human societies, biological systems and groups of artificial agents. It also lies behind practical questions about collective action: whether communities manage shared resources, whether autonomous systems coordinate safely, and whether agents trained separately can work together when they meet.
This project studies how cooperation emerges, persists and fails among intelligent agents. Possible questions include:
- How can an agent learn to cooperate when it does not know who it will encounter?
- Which forms of reputation, punishment or social norms sustain cooperation?
- Do artificial agents respond to strategic incentives in the same way as people?
- How do conformity, communication and social learning affect collective decisions?
- What happens when agents differ in their information, objectives or ability to learn?
The precise question will be developed with the student. A project may combine evolutionary and behavioural game theory, multi-agent reinforcement learning, agent-based simulation, or experiments with people and AI systems. Some students may develop new theory or algorithms; others may use existing methods to explain behaviour in a particular setting. In either case, the aim is to connect formal models with observable behaviour.
Our group studies cooperation across multi-agent systems and AI, social systems, and biology. Recent work includes evolutionary models of cooperation and methods for learning to cooperate against diverse opponents. Publications: http://garciajulian.com
References
- García, J. and Traulsen, A. “Evolution of coordinated punishment to enforce cooperation from an unbiased strategy space.” Journal of the Royal Society Interface 16 (2019): 20190127. https://doi.org/10.1098/rsif.2019.0127
- Perera, I., de Nijs, F. and García, J. “Learning to cooperate against ensembles of diverse opponents.” Neural Computing and Applications (2025). https://doi.org/10.1007/s00521-024-10511-9
- de Arruda, H. F., Gracia-Lázaro, C., Aleta, A. and Moreno, Y. “Collective cooperation without individual fidelity in LLM agents.” arXiv (2026). https://arxiv.org/abs/2606.30454
Required knowledge
- An excellent academic record in computer science or a cognate field. An Honours degree with H1 or equivalent is required for Graduate Research admission.
- An interest in game theory, mathematical modelling and the study of cooperation.
- Good programming skills. Experience with machine learning is useful but not required.
- Excellent written and verbal communication skills.
