Strategy Selection for Physical Reasoning

A longstanding debate in intuitive physics has been whether our reasoning is based on either simulation or rules and logic. In my work, I demonstrate that both of these capabilities are used for physical reasoning, and that we choose between systems of reasoning based on meta-reasoning about the expected costs and benefits of each.

In addition to using the outcome of simulation to reason about physics, we can also use logical reasoning constructed from pieces of (potentially erroneous) knowledge about the world. This use of multiple systems can explain a dichotomy in the literature on intuitive physics: we can interact with the world flexibly and accurately, but show characteristic errors in many judgments involving even simple physical principles. I have found that when we have perceptual-motor interactions with the world, we rely on calibrated simulation for their judgments; however, when we are asked more explicit queries, our judgments are instead based on the logical reasoning cognitive system.

But how do we choose which system to use? This choice can be studied using the framework of meta-reasoning: choosing a strategy that maximizes the expected benefits of using one strategy, minus the cognitive or opportunity costs of using that strategy. By studying scenarios where people use both simulation and logical rules to make their judgments – deciding how a balance beam will fall – I find that the relative use of different strategies is consistent with this meta-reasoning choice.

Related papers:

Avatar
Kevin A Smith
Research Scientist in Brain and Cognitive Sciences