新研究：忽略人类信念的世界模型会预测错误行为

Researchers extend AI world models with the "Mental World Modeling" framework to include mental states like human beliefs and intentions. The reference implementation MENTIS simulates action options and scores them on physical plausibility, mental consistency, and social appropriateness. In experiments, the approach significantly improves how accurately language models predict human behavior. World models are supposed to be the missing foundation for autonomous AI agents, predicting how a scene changes when an action is taken. A new paper argues that the current generation of these systems ignores a critical building block: what's going on inside the heads of the people involved. Existing world models like Sora , Genie 3 , JEPA , and Marble only model the physical layer of the world, according to the paper: objects, positions, motion, and occlusion. What people in that world believe, want, or consider socially appropriate never shows up in their state space. For service robots, medical assistants, or collaborative agents, that's not enough, because hidden mental states largely drive human behavior. The authors illustrate the gap with a simple example. If someone's cup gets moved into a cabinet while they're not looking, the scene looks correct to a purely physical world model. But it still predicts the wrong next action. Only a model that also tracks the person's belief about where the cup is can explain what they'll actually do. Ad Their framework, called "Mental World Modeling" (MWM) and published on GitHub , extends classic world models with mental variables like beliefs, attention, goals, intentions, emotions, norms, and social relationships. The target agent only sees an egocentric partial view, while the world model holds the complete state. Ad Every action splits into a physical carrier, such as speaking, pointing, or grasping, and a mental payload, such as comforting, deceiving, or rejecting. The same gesture of sliding a cup across the table can be an apology, a deception, or an act of care. Only the world model holds the variables that tell them apart. The authors explicitly don't claim to simulate consciousness. Mental states are hypotheses drawn from behavior and context, not measurements. Systems built on the framework should represent uncertainty and keep their assumptions transparent. Ad To test the theory, the researchers built MENTIS, a modular pipeline that requires no additional training. It breaks the process into six steps. First, the scene is parsed and the ego perspective is rendered. Then the action options are split into physical and mental components, and the resulting states are simulated in parallel. Each branch is then scored on three criteria: physical plausibility, mental consistency, and social appropriateness. After that, the pipeline makes a deterministic decision. Every stage writes a machine-readable intermediate result, so errors can be traced back to a specific step. Ad For evaluation, the authors built Menti-Bench, a dataset of 448 decision scenes: 320 text descriptions, 100 picture stories, and 28 sound-video clips. Each scene contains six response options and a human-created reference solution that documents not just the correct action but also the underlying mental and physical states. 78 percent of the scenes involve at least two characters. Ad