Research · Track 01
AI Reasoning with Game Theory
Rigorous foundations for how multiple agents decide, cooperate and compete.
8–12 contributors
Open — seeking a lead
5–15 hrs/week contributing · 20+ hrs/week core
Why now
Context
Reasoning is the next frontier in AI, and game theory offers rigorous foundations for multi-agent decisions, cooperation and adversarial dynamics. Most reasoning benchmarks still measure a single model answering a static question. Almost none measure what happens when two systems want different things.
The idea for this track came out of an ABAIA discussion between members and an MIT professor. It stayed on the list because the question kept getting sharper rather than vaguer: as soon as systems act on each other’s outputs, the useful question stops being is the answer correct and becomes is the behaviour stable, and stable under what assumptions.
That is a question game theory has spent seventy years building tools for, and almost none of those tools have been pointed at language models.
Sub-themes
Scope
- Game-theoretic frameworks for multi-agent LLM reasoning
- Mechanism design for agent coordination and incentive alignment
- Adversarial robustness and equilibrium analysis in reasoning chains
- Cooperative AI — trust, commitment and shared goals
- Bounded rationality models for human-like reasoning
- Benchmarks for strategic reasoning that go beyond static question answering
Year one deliverables
Output
- One to two position papers
- An open benchmark and two prototypes
If the benchmark does not separate models that a practitioner would already rank differently, the track closes at the Q4 review rather than being carried into Year 2.