Research map / Large language models / AI and society
World Models Should Prioritize the Unification of Physical and Social Dynamics
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Position paper arguing that world models must bidirectionally unify physical and social dynamics, proposing ACE principles, a WM_P-S formulation, a research roadmap, and a hierarchical evaluation protocol.
Problem
Current world models model physical dynamics and social behavior in separate silos, failing to capture the bidirectional interplay between physical environments and social constructs needed for real-world prediction and planning.
Approach
Frames world model prediction along physical and social dimensions and argues for their deep bidirectional unification; proposes ACE principles (Abstraction of social complexity, Contingent causality of social dynamics, Entangled co-evolution and emergence) and a joint physical-social state transition formulation WM_P-S.
Key results
- Position paper with no new experiments; proposes a three-tier hierarchical evaluation protocol (perceptual fidelity, disentangled dynamics, entangled dynamics).
- Surveys and tabulates physical world model methods (intuitive physics, MBRL, video generators, LLM for physical, 3D) and social world model methods (ToM, MBMARL, LLM for social) and benchmarks.
Limitations
- No empirical validation of the proposed WM_P-S framework or ACE principles
- Social world model literature is acknowledged as underdeveloped, so the survey relies on sparse examples
Experimental setup
Builds on: model-based reinforcement learning Dreamer theory of mind LLM social simulation
Closest papers by meaning
- Social World Model-Augmented Mechanism Design Policy Learning (NeurIPS 2025 · similarity 0.85)
- Socialized Coevolution: Advancing a Better World through Cross-Task Collaboration (ICML 2025 · similarity 0.84)
- MF-LLM: Simulating Population Decision Dynamics via a Mean-Field Large Language Model Framework (NeurIPS 2025 · similarity 0.84)
- Cooperate or Collapse: Emergence of Sustainable Cooperation in a Society of LLM Agents (NeurIPS 2024 · similarity 0.84)
- General agents need world models (ICML 2025 · similarity 0.84)
- Simplifying Latent Dynamics with Softly State-Invariant World Models (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.