A Model-Free Universal AI
This work addresses the problem of creating a model-free optimal agent for general reinforcement learning, which is a foundational problem for AI researchers.
This paper introduces Universal AI with Q-Induction (AIQI), the first model-free agent proven to be asymptotically ε-optimal in general reinforcement learning. It achieves this by performing universal induction over distributional action-value functions.
In general reinforcement learning, all established optimal agents, including AIXI, are model-based, explicitly maintaining and using environment models. This paper introduces Universal AI with Q-Induction (AIQI), the first model-free agent proven to be asymptotically $\varepsilon$-optimal in general RL. AIQI performs universal induction over distributional action-value functions, instead of policies or environments like previous works. Under a grain of truth condition, we prove that AIQI is strong asymptotically $\varepsilon$-optimal and asymptotically $\varepsilon$-Bayes-optimal. Our results significantly expand the diversity of known universal agents.