World Models for AI Agents

Introduction: The Map and the Territory

1. Why This Book Exists
2. What We Mean by “World Model”
3. The Architecture of This Book
4. Who This Book Is For
5. What Makes This Book Different
6. A Note on Existing Work
7. Let’s Begin

Chapter 1: Agentic Memory

1. The Fundamental Problem: Amnesia by Design
2. Memory Is Not Retrieval
3. Laying the Architectural Foundation
4. Memory as Core Architecture
5. The Neuro-Symbolic Synthesis
6. Building World Models on Memory
7. Toward Cognitive Agents
8. Bridge to Chapter 2
9. References and Further Reading

Chapter 2: Types of Memory

1. Human Memory: The Foundation
2. Episodic Memory for Agents
3. Semantic Memory for Agents
4. Procedural Memory for Agents
5. Working Memory: The Active Workspace
6. Cognitive Architectures: SOAR and ACT-R
7. Mapping Memory Types to Agent Architecture
8. Connecting to the Horizon

Chapter 3: What to Remember

1. We Don’t Remember Facts — We Maintain a World View
2. The World Model Is What Matters
3. The Paradox of Total Recall
4. From Memory to Model: How Humans Do It
5. What Agents Should Actually Remember
6. The Art of Forgetting Serves the Model
7. Memory Consolidation: Building the Model
8. Memory Compression: Shaping the Model
9. The Dimensions of Memory Value
10. This Is What the Book Is About

Chapter 4: World Model

1. What Is a World Model?
2. The Historical Arc: From Cybernetics to Deep Learning
3. LeCun’s Vision: Joint Embedding Predictive Architecture
4. World Models for Planning: The Dreamer Series
5. Do LLMs Have World Models? A Contested Question
6. The World Model Gap: A Missing Layer in Current Agents
7. Foundation World Models vs. Agent World Models
8. Toward Agent World Models: The Path Forward
9. Bridging to Human Cognition
10. Summary

Chapter 5a: Causal World Models

1. Introduction: From Correlation to Causation
2. Pearl’s Ladder of Causation and the Levels of Learning
3. The Necessity of Causal Models: Recent Theoretical Results
4. Structural Causal Models and the Foundation of Causal Reasoning
5. Interventionist Causality and Agent Architecture
6. How Agents Learn Causal Structure: Exploration as Experimentation
7. Causal Representation Learning and Neural Implementation
8. Language Models and Causal World Models: A New Integration
9. Causal World Models and the Broader Architecture
10. Conclusion: Causality as a Core Competence
11. References

Chapter 5: Humans as a World Model

1. The Brain as a Prediction Machine
2. Active Inference and the Action-Perception Loop
3. Embodied Cognition and the Extended Mind
4. Mental Models and Simulation-Based Reasoning
5. Theory of Mind: Modeling Other Agents
6. Simulation Theory and Understanding Action
7. Synthesis: Architectural Principles from Human Cognition
8. Looking Ahead

Active Inference and the Free Energy Principle

1. The Free Energy Principle: From Physics to Biology to AI
2. Generative Models as World Models
3. Expected Free Energy and Planning
4. Generative Models in Hierarchies and Across Time
5. Active Inference in Robotics and Multi-Agent Scenarios
6. Bridging to Large Language Models
7. Connections Throughout This Book
8. Challenges and Open Questions
9. Conclusion
10. References

Chapter 6: Perception of Time by Agents, Metagraphs and Tensor Clocks

1. The Wrong Way to Remember When
2. The Temporal Challenge
3. From Physical Clocks to Logical Clocks
4. Vector Clocks and the Strength of Causality
5. Tensor Clocks and the Limits of Dimensionality
6. Hierarchical Time Models for Multi-Agent Systems
7. Metagraphs as a Temporal Substrate
8. Semantic Spacetime: Time as Structure
9. Temporal Knowledge for Agents: From Theory to Practice
10. Bridging to Social Understanding

Chapter 7: Perception of Social Interaction and Trust by the Agent

1. Agents as Social Beings
2. The Social Graph: Mapping Relational Context
3. Computational Trust Models: Beyond Binary Belief
4. Theory of Mind in Multi-Agent Worlds
5. Agent Identity: The Question of “This Agent”
6. Reputation Systems for Agent Ecosystems
7. Social Memory: Temporal Dimensions of Relationships
8. Networked Agent Organizations: Emergence of Collective Structure
9. Toward a Social Architecture for Agents
10. Bridge: From Trust to Promise Theory

Chapter 7b: Multi-Agent World Models and Shared Cognition

1. Introduction: The Limits of Isolated Understanding
2. The Fundamental Challenge: Why Shared World Understanding Matters
3. Communicating Plans, Not Percepts
4. Theory of Mind as Hierarchical World Modeling
5. Emergent Coordination Through Persona and Reasoning Design
6. Shared Mental Models in Teams
7. Multi-Agent World Models: Technical Approaches
8. Collective Intelligence: When the Group Knows More Than the Individual
9. Building Compatible World Models: Design Principles
10. Challenges and Limitations
11. Toward Integrated Understanding: Preview of Promise Theory
12. Conclusion: Toward Truly Cooperative Intelligence
13. References

Chapter 8: Promise Theory for Agent Social Interactions

1. Introduction: The Coordination Crisis
2. The Command-and-Control Fiction
3. Mark Burgess and the Origins of Promise Theory
4. Core Principles of Promise Theory
5. Promise Graphs as Architecture
6. Promise-Based Coordination in Practice
7. Trust Through Promises
8. Semantic Spacetime and Promise Theory
9. Promise Graphs Versus Action Logs
10. Practical Implementation
11. Conclusion: The Foundation for Multi-Agent Architecture

Chapter 9: Metagraphs as a Backbone of World Models

1. The Representation Problem: Why Existing Structures Fall Short
2. The Evolution of Graph Structures: A Journey Toward Expressiveness
3. Metagraph Theory: The Formal Foundation
4. Holonic Systems: Wholes That Are Also Parts
5. Bigraphs: Separating Place From Link
6. Metagraphs for World Models: Putting It Together
7. Category Theory Connections: Mathematical Foundations
8. Topological Deep Learning: Bridging Symbolic and Neural
9. Practical Approximations: From Theory to Implementation
10. From Metagraphs to Grounding
11. Summary

Chapter 10: Physical World Foundation Models

1. The Physical World Problem
2. Video Generation as World Simulation
3. Interactive World Models: Genie and Beyond
4. The Dreamer Architecture
5. Physics-Informed Neural Networks
6. Foundation Models for Embodied AI
7. Autonomous Driving as the Ultimate Test
8. Two Kinds of World Models
9. Integration with Agent Architecture
10. Learning from Imagination
11. The Road Ahead
12. Conclusion

Chapter 10b: Embodied World Models and the Grounding Problem

1. Introduction: The Limitation of Disembodied Understanding
2. The Grounding Gap: From Passive Observation to Active Understanding
3. Vision-Language-Action Models: Grounding Language in Physical Capability
4. Three-Dimensional Grounding: From 2D Prediction to 3D Understanding
5. Mental World Models: Modeling Minds, Not Just Matter
6. Can Video Generation Models Learn Physics? The Reality Check
7. The Multimodal Large Language Model - World Model Synthesis
8. Connections: From Embodiment to Episodic Memory to Abstraction
9. Practical Challenges and Remaining Gaps
10. Conclusion: Grounding as a Process, Not a Property
11. References

Chapter 10c: JEPA and Latent World Models

1. The Space of Prediction
2. LeCun’s Blueprint for Autonomous Intelligence
3. How JEPA Works: Learning to Ignore
4. The JEPA Lineage: From Images to Video to Stable World Models
5. Why Latent Prediction Matters: Comparison and Context
6. JEPA as an Agent Architecture
7. Learning Representations, Capturing Dynamics
8. Latent World Models and Graph Structures
9. Practical Implications and Open Questions
10. Conclusion
11. References

Chapter 11: Graphs as World Models

1. The Obvious Idea No One Pursued
2. Worldformer: Predicting the Difference
3. AriGraph: Where Memory Meets World Models
4. The Graph World Model: Unifying Everything
5. The Convergence
6. Connecting to Metagraphs
7. The Promise of Structured World Models
8. Bridge to Chapter 12
9. References and Further Reading

Chapter 12: Toward a Unified Agent Architecture

1. The Integration Problem
2. The Layers of a Cognitive Agent
3. The Reasoning Engine
4. Two Kinds of World Models, Reunited
5. What Autonomy Requires
6. The Road Ahead
7. References and Further Reading

About This Book

1. Memory Was Only the Beginning
2. Why World Models Are Foundational
3. What This Book Argues
4. Who This Book Is For
5. From Three Books to One

About the Author

About the Cover

1. The Three Realms
2. The Tryzub
3. The Tree as Graph
4. The Eternal Contest