Efficient determination of simplified representations for mental simulation

Determine the cognitive and algorithmic mechanisms by which humans efficiently select and construct simplified representations (construals) of complex environments for simulation-based reasoning, specifying how people decide which scene details to encode and which to abstract away without performing exhaustive precomputation.

Background

Mental simulation is central to human planning and physical prediction, but real-world scenes contain far more detail than can be feasibly maintained in working memory. Prior theories propose that people use reduced representations that omit irrelevant details, motivated by resource-rational trade-offs between utility and cognitive cost. However, most accounts focus on what should be represented rather than how these construals are formed, and evaluating an optimal construal naively can be more costly than simulating the full environment. This creates a need for a process-level account of how people efficiently construct representations during simulation.

The paper proposes a Just-in-Time (JIT) framework that interleaves simulation, visual lookahead, and incremental encoding to build construals online. While this provides an algorithmic account and empirical evidence in planning and physical reasoning tasks, the motivating question remains explicitly articulated as unclear, highlighting the broader open problem of characterizing the mechanisms that enable efficient determination of such simplifications.