A simulation-heuristics dual-process model for intuitive physics

TL;DR

The paper tackles how people reason about physical events under varying cognitive costs, proposing that mental simulation alone cannot account for all judgments. It introduces the Simulation-Heuristics Model (SHM), a dual-process framework that switches from noisy mental simulation to a linear heuristic when simulation becomes costly, with a boundary calibrated from data. Through a pouring-marble task and a four-step experimental program, the authors identify a switching point near $65^ frac{°}$ (and a time-based threshold near $68.2^ frac{°}$) and demonstrate that SHM outperforms purely stochastic simulation or heuristic approaches in predicting human judgments across 54 conditions. This work advances understanding of adaptive, cost-based strategy selection in intuitive physics and provides a unified framework for integrating complementary computational methods to model human reasoning.

Abstract

The role of mental simulation in human physical reasoning is widely acknowledged, but whether it is employed across scenarios with varying simulation costs and where its boundary lies remains unclear. Using a pouring-marble task, our human study revealed two distinct error patterns when predicting pouring angles, differentiated by simulation time. While mental simulation accurately captured human judgments in simpler scenarios, a linear heuristic model better matched human predictions when simulation time exceeded a certain boundary. Motivated by these observations, we propose a dual-process framework, Simulation-Heuristics Model (SHM), where intuitive physics employs simulation for short-time simulation but switches to heuristics when simulation becomes costly. By integrating computational methods previously viewed as separate into a unified model, SHM quantitatively captures their switching mechanism. The SHM aligns more precisely with human behavior and demonstrates consistent predictive performance across diverse scenarios, advancing our understanding of the adaptive nature of intuitive physical reasoning.

Figures (4)

• Figure 1: (A) Experimental design: Trials involved 3 cup shapes (H-shape, A-shape, V-shape), 3 object shapes (circle, triangle, trapezoid), 3 sizes (large, medium, small), and 2 filling heights (full, half), totaling 54 unique conditions. Participants predicted the tilt angle for marbles to fall out when cups are tilted to the left. (B) dual hypothesis: Participants used either mental simulation, simulating the tilting process until pouring out, or a heuristic strategy, reaching judgments from physical features when the simulation exceeds a boundary. These methods could result in different outcomes. (C) Human results: Each point represents a condition, illustrating human tendencies to either overestimate or underestimate the pouring angle. The red and blue lines are the regression results of ipe and the heuristic model, respectively. The dual effectively captures human behavior with a switching boundary.
• Figure 2: Visualizations of stimuli and error analysis. (a) Example stimuli. The top (red), middle (black), and bottom (blue) rows depict two scenarios each, with pouring angles that are smaller, close to, and larger than the established simulation bound, respectively. (b) The mean absolute error between model and human results (with SEM). The ipe model exhibits a larger absolute error when the simulation time exceeds the boundary. Conversely, the heuristic model shows contrary results, indicating its effectiveness in these scenarios.
• Figure 3: Comparison between dual and other baseline models. The correlation and RMSE between model predictions and human predictions across all 54 conditions are compared. Among the four models evaluated, dual demonstrates the highest correlation and the lowest RMSE, indicating its superior predictive accuracy.
• Figure 4: Comparison of four models' RMSE on different conditions. RMSE is calculated as the root mean square error between the model's predicted pouring angle and the human judgments. The bottom right figure represents the performance across all 54 trials. A dashed line is included to indicate the RMSE of the SHM, showing a clear advantage when compared with other models.