Baba in Wonderland: Online Self-Supervised Dynamics Discovery for Executable World Models

TL;DR

This paper introduces Alice, a system for online learning of executable world models that can induce environment dynamics without semantic priors, especially under prior misalignment.

Contribution

Alice uses structural signals from failed updates to refine hypotheses and guide exploration, improving world-model learning in misaligned settings.

Findings

Alice significantly improves learning under prior misalignment.

Class refinement and class-aware exploration both enhance performance.

Experiments on Baba in Wonderland demonstrate effectiveness.

Abstract

Executable world models can be read, edited, executed, and reused for planning, but only if the program captures the environment's transition law rather than semantic shortcuts in its surface vocabulary. We study online executable world-model learning under prior misalignment, where an agent must induce state-dependent dynamics from interaction evidence alone, without rule descriptions, reward signals, or trustworthy lexical priors. We introduce Alice, a closed-loop system that treats failed candidate updates as structural signal: when a candidate explains a new transition but loses previously explained ones, the preservation conflict reveals dynamics that the current program had conflated. Alice refines these conflicts into hypothesis classes that both provide compact, class-stratified preservation counterexamples for update and guide frontier exploration toward transitions that are novel and underrepresented with respect to the current program. We evaluate Alice on Baba in Wonderland, a prior-misaligned variant of Baba Is You that preserves simulator dynamics while replacing semantically meaningful rule-property labels with unrelated words. Experiments show that Alice substantially improves executable world-model learning under prior misalignment, and ablations show that both class refinement and class-aware exploration contribute.