General Agents Contain World Models, even under Partial Observability and Stochasticity

2026-02-03

AI Summary

This work addresses whether agents operating in partially observable and stochastic environments necessarily possess an implicit world model capable of reconstructing their surroundings. By relaxing the notion of universality and dispensing with assumptions of determinism and full observability, the paper establishes—for the first time under more realistic conditions—that any agent exhibiting a certain degree of generality must implicitly encode a model of its environment. Drawing on formal tools from theoretical computer science and reinforcement learning, the argument is rigorously developed through black-box querying and approximate reconstruction techniques. Furthermore, the analysis demonstrates that randomized policies cannot circumvent the need to learn structural properties of the environment, thereby providing a strong theoretical foundation for the claim that competent agents inherently contain world models.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Multiagent Systems: Modeling other Agents
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

Deciding whether an agent possesses a model of its surrounding world is a fundamental step toward understanding its capabilities and limitations. In [10], it was shown that, within a particular framework, every almost optimal and general agent necessarily contains sufficient knowledge of its environment to allow an approximate reconstruction of it by querying the agent as a black box. This result relied on the assumptions that the agent is deterministic and that the environment is fully observable. In this work, we remove both assumptions by extending the theorem to stochastic agents operating in partially observable environments. Fundamentally, this shows that stochastic agents cannot avoid learning their environment through the usage of randomization. We also strengthen the result by weakening the notion of generality, proving that less powerful agents already contain a model of the world in which they operate.

Problem

• world models
• partial observability
• stochasticity
• general agents
• environment modeling

Innovation

• world models
• stochastic agents
• partial observability
• general agents
• black-box reconstruction