Toward Universal and Interpretable World Models for Open-ended Learning Agents
This work addresses the need for universal and interpretable world models in AI agents, though it appears incremental as it builds on existing Bayesian and planning methods.
The authors tackled the problem of creating interpretable and scalable world models for open-ended learning agents by introducing a sparse class of Bayesian networks, which can approximate various stochastic processes and support developmental learning for more adaptive behavior.
We introduce a generic, compositional and interpretable class of generative world models that supports open-ended learning agents. This is a sparse class of Bayesian networks capable of approximating a broad range of stochastic processes, which provide agents with the ability to learn world models in a manner that may be both interpretable and computationally scalable. This approach integrating Bayesian structure learning and intrinsically motivated (model-based) planning enables agents to actively develop and refine their world models, which may lead to developmental learning and more robust, adaptive behavior.