mlsanityData to 14 Oct 2026

General agents contain world models

arXiv 2506.01622 ↗First posted 2 Jun 2025cs.AI cs.LG cs.ROcited by 17

On the map: World models → World models for language agents

Abstract (arXiv)

Are world models a necessary ingredient for flexible, goal-directed behaviour, or is model-free learning sufficient? We provide a formal answer to this question, showing that any agent capable of generalizing to multi-step goal-directed tasks must have learned a predictive model of its environment. We show that this model can be extracted from the agent's policy, and that increasing the agents performance or the complexity of the goals it can achieve requires learning increasingly accurate world models. This has a number of consequences: from developing safe and general agents, to bounding agent capabilities in complex environments, and providing new algorithms for eliciting world models from agents.

Key facts. Cited by 17 papers in the map (data through 14 Oct 2026). Placed in World models for language agents, one of 11 directions in World models. Builds on 51 of its 98 references that are in the map. First posted on arXiv 2 Jun 2025.

Builds on

51 of 98 references.

Closest ancestors · 4

Same direction, newest first

  1. 2024Emergent World Models and Latent Variable Estimation in Chess-Playing Language Models21 Mar
  2. 2024Robust agents learn causal world models16 Feb
  3. 2023Reasoning with Language Model is Planning with World Model24 May
  4. 2020Language Models are Few-Shot Learners28 May

Other references · 47

  1. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
  2. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 3,080
  3. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 2,407
  4. 2023Voyager: An Open-Ended Embodied Agent with Large Language Modelscited by 2,191
  5. 2023PaLM-E: An Embodied Multimodal Language Modelcited by 2,071
  6. 2016Concrete Problems in AI Safetycited by 1,642
  7. 2018World Modelscited by 1,513
  8. 2016Understanding intermediate layers using linear classifier probescited by 1,353
  9. 2019Mastering Atari, Go, Chess and Shogi by Planning with a Learned Modelcited by 1,353
  10. 2018Learning Latent Dynamics for Planning from Pixelscited by 1,264
  11. 2023Mastering Diverse Domains through World Modelscited by 1,253
  12. 2022A Generalist Agentcited by 808
35 more
  1. 2018Deep Reinforcement Learning in a Handful of Trials using Probabilistic Dynamics Modelscited by 774
  2. 2018Scalable agent alignment via reward modeling: a research directioncited by 447
  3. 2022Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Taskcited by 431
  4. 2021Safe Learning in Robotics: From Learning-Based Control to Safe Reinforcement Learningcited by 371
  5. 2023Language Models Represent Space and Timecited by 316
  6. 2019Challenges of Real-World Reinforcement Learningcited by 291
  7. 2020Transfer Learning in Deep Reinforcement Learning: A Surveycited by 265
  8. 2018Machine Theory of Mindcited by 226
  9. 2022Goal-Conditioned Reinforcement Learning: Problems and Solutionscited by 158
  10. 2017Imagination-Augmented Agents for Deep Reinforcement Learningcited by 140
  11. 2021Can Language Models Encode Perceptual Structure Without Grounding? A Case Study in Colorcited by 119
  12. 2016Reinforcement Learning With Temporal Logic Rewardscited by 117
  13. 2024Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systemscited by 96
  14. 2019Reinforcement Learning for Temporal Logic Control Synthesis with Probabilistic Satisfaction Guaranteescited by 77
  15. 2025Superintelligent Agents Pose Catastrophic Risks: Can Scientist AI Offer a Safer Path?cited by 75
  16. 2021LTL2Action: Generalizing LTL Instructions for Multi-Task RLcited by 74
  17. 2021A Survey on Interpretable Reinforcement Learningcited by 73
  18. 2024Scaling Instructable Agents Across Many Simulated Worldscited by 71
  19. 2017Environment-Independent Task Specifications via GLTLcited by 67
  20. 2018Theoretical Impediments to Machine Learning With Seven Sparks from the Causal Revolutioncited by 66
  21. 2023Towards a Mechanistic Interpretation of Multi-Step Reasoning Capabilities of Language Modelscited by 50
  22. 2022A Review of Uncertainty for Deep Reinforcement Learningcited by 40
  23. 2023Honesty Is the Best Policy: Defining and Mitigating AI Deceptioncited by 40
  24. 2020Encoding formulas as deep networks: Reinforcement learning for zero-shot execution of LTL formulascited by 36
  25. 2022Counterfactual harmcited by 29
  26. 2024STEER: Assessing the Economic Rationality of Large Language Modelscited by 27
  27. 2024The Reasons that Agents Act: Intention and Instrumental Goalscited by 23
  28. 2024DeepLTL: Learning to Efficiently Satisfy Complex LTL Specifications for Multi-Task RLcited by 21
  29. 2022Path-Specific Objectives for Safer Agent Incentivescited by 20
  30. 2025MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hackingcited by 19
  31. 2016Towards Resolving Unidentifiability in Inverse Reinforcement Learningcited by 18
  32. 2025Interpreting Emergent Planning in Model-Free Reinforcement Learningcited by 15
  33. 2017Good and safe uses of AI Oraclescited by 14
  34. 2024Can a Bayesian Oracle Prevent Harm from an Agent?cited by 13
  35. 2023Towards Bounding Causal Effects under Markov Equivalencecited by 8

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 4; 2026: 13.

17 papers.

Sort:

By area:

  1. 2026Do World Action Models Generalize Better than VLAs? A Robustness Studycited by 36
  2. 2026Qwen-AgentWorld: Language World Models for General Agentscited by 28
  3. 2025Assessing Adaptive World Models in Machines with Novel Gamescited by 20
  4. 2025Self-Predictive Representations for Combinatorial Generalization in Behavioral Cloningcited by 12
  5. 2025Cogito, Ergo Ludo: An Agent that Learns to Play by Reasoning and Planningcited by 12
  6. 2026Nano World Models: A Minimalist Implementation of Future Video Predictioncited by 5
  7. 2025A "good regulator theorem" for embodied agentscited by 4
  8. 2026Generalizable Dense Reward for Long-Horizon Robotic Taskscited by 3
  9. 2026Affordances Enable Partial World Modeling with LLMscited by 2
  10. 2026What Capable Agents Must Know: Selection Theorems for Robust Decision-Making under Uncertaintycited by 2
  11. 2026Epistemic Traps: Rational Misalignment Driven by Model Misspecificationcited by 0
  12. 2026Event-Driven Reinforcement Learning Enables Long-Horizon Control in Semiconductor Fabricationcited by 0
5 more
  1. 2026From Optimal Actions to World Models: Identifiability of Transition Kernels in Discounted MDPscited by 0
  2. 2026The syntax and semantics of goalscited by 0
  3. 2026DSWM: Decomposed Spatio-Temporal World Model for Demand-Driven UAV Base Station Repositioningcited by 0
  4. 2026Software World Models: From Consequence Prediction to Decision Valuecited by 0
  5. 2026The Anatomy of a Blind Spot: What an LLM Judge Must Be Shown, and How Muchcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026How do World Models and Policies Compose in LLM Agents? A Joint Spectral and Behavioral Account
  2. 2026World Models as an Intermediary between Agents and the Real World
  3. 2026Dual-Frontier: When Can an Agent Trust Its World Model?
  4. 2026General Agents Contain World Models, even under Partial Observability and Stochasticity
  5. 2026Agent-World: Scaling Real-World Environment Synthesis for Evolving General Agent Intelligence
  6. 2026Learning Task-Sufficient World Models by Synergizing Agentic Exploration and Structured Modeling
  7. 2026Foundation World Models for Agents that Learn, Verify, and Adapt Reliably Beyond Static Environments
  8. 2024WorldCoder, a Model-Based LLM Agent: Building World Models by Writing Code and Interacting with the Environment