Universal Algorithmic Intelligence

Yegon Kim will present tomorrow, Monday March 23rd at 3 pm EDT on a model-free universal artificial intelligence: https://arxiv.org/abs/2602.23242

We will use the regular zoom link.

Bio:

Yegon Kim is a PhD student researching AI at KAIST. Yegon also has an interesting blog/personal website here: https://yegonkim.github.io/

Abstract:

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 ε-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 ε-optimal and asymptotically ε-Bayes-optimal. Our results significantly expand the diversity of known universal agents.

This is a (difficult) classical topic in AIXI research, e.g. Optimal Direct Policy Search and Extreme State Aggregation, and I look forward to learning about some progress!