Mental World Modeling

World models enable a predictive substrate for planning and action, yet existing formulations merely answer a physical question: what/where it is, and how will it evolve. Human behavior, however, is driven by hidden mental state (what a person believes, wants, intends, feels, and considers socially permissible), so a model that tracks the physical scene but not what each agent knows and believes about it predicts the wrong action for the right-looking scene. We formulate Mental World Modeling (MWM), a generic theoretical framework that makes mental variables core components of a world model rather than posthoc rationales: MWM maintains a coupled physical-mental world state, renders a target-specific partial observation, and simulates how candidate actions jointly update both components. We instantiate the framework in MENTIS, a training-free and fully inspectable baseline that decomposes the process into state parsing, target-observation generation, action decomposition, coupled physical and mental transition, and branch-level value evaluation.

Introduction. “If the organism carries a ‘small-scale model’ of external reality and of its own possible actions within its head, it is able to try out various alternatives.” Kenneth Craik, 1943 [1] Large language models (LLMs) [2, 3] now write programs, use tools, compose plans, and serve as the cognitive core of autonomous agents [4–9], yet they do not by themselves constitute agents that live in, predict, and intervene on an evolving world. A capable agent must know what is true now, what would happen to itself if an action were taken, and how the world would change before the action is executed. This pressure has returned world models to the center of AI research: a world model gives an agent a predictive substrate for counterfactual simulation, planning, and decision making [1, 10, 11]. By now, the term world model is still used in several ways, but three technological families are especially relevant here.

Discussion / Conclusion. This paper introduces Mental World Modeling as a framework for world models that reason about people rather than only objects. The central move is to treat physical and mental states as one coupled world state, render a target-specific partial observation from that state, and simulate how a candidate action changes both the material scene and the mental-social configuration. We formalize this idea, specify a practical state taxonomy, implement MENTIS as an inspectable baseline, and evaluate the necessity and limits of MWM on a process-annotated testbed. The evidence is consistent across 8 modern LLM-based world models from two families: explicit mental world modeling is necessary for predicting human decisions, the mental channel, the physical channel, and their coupled transition are all required, and the gains are largest on interpersonal scenes, where hidden mental variables govern the decision, while holding across text, image, and sounding-video inputs. Oracle interventions localize most of the remaining gap to the human reference in transition simulation, which gives future improvement a clear priority order.