Primary supervisor

Julian Garcia Gallego

AI can help an individual find information, generate alternatives and make decisions. Yet a group may lose something when everyone relies on the same tools. Independent exploration can decline, errors can become correlated, and a population can settle on a common answer that is worse than the alternatives it no longer considers. Individual improvement does not guarantee collective intelligence.

This project studies how patterns of AI use spread through a population and what they do to diversity, competence and welfare. Possible questions include:

  • When is AI used as a complement to independent reasoning, and when does it replace it?
  • Can individually beneficial reliance on AI lead to an inefficient social norm?
  • What happens when rewards for immediate accuracy reduce the diversity of ideas available to the group?
  • Can preferences for independent thought, truth-seeking or prosocial behaviour survive when they are initially costly?
  • Which interventions preserve useful diversity without giving up the individual benefits of AI?

The project will develop a game-theoretic or cultural-evolution model and study its population dynamics through analysis or simulation. An Honours project would normally begin with a small set of AI-use strategies and ask whether the resulting equilibria differ in collective performance. Cooperation and collective risk may enter as mechanisms, but the central outcome is the quality of the group's knowledge or decisions.

Aim/outline

References

  • de Arruda, H. F. and Moreno, Y. “The social consequences of AI delegation.” arXiv (2026). https://arxiv.org/abs/2606.11058
  • Wang, G., Su, Q., Wang, L. and Plotkin, J. B. “Individual incentives that promote collective intelligence.” PNAS 122(51) (2025). https://doi.org/10.1073/pnas.2516535122
  • Solé, R. et al. “Large-Language Models as a Cognitive Virus.” arXiv (2026). https://arxiv.org/abs/2609.03344

Required knowledge

An interest in AI, social learning and the gap between individual and collective outcomes.

  • Good programming skills and a solid mathematical background.
  • Prior experience with evolutionary game theory or dynamical systems is useful but not required.
  • Having taken FIT3139 is useful but not required.