Skip to main content
Open Access Publications from the University of California


UCLA Electronic Theses and Dissertations bannerUCLA

On the Human Capacity for Physical and Analogical Inference


The field of intuitive physics has been reinvigorated in recent years, providing converging computational, behavioral, and neural evidence demonstrating that humans possess a wealth of knowledge about the attributes and dynamics of physical entities in the world. However, progress in the field has led to many questions regarding (i) the human capacity for perception, physical inference, and higher-level reasoning; and (ii) how to design a machine to emulate newfound capabilities. My dissertation focuses on a range of research topics, from the characteristics of human perception and representation in dynamic scenes, to mental simulation of novel situations and transfer of higher-level knowledge to seemingly unrelated problem domains. In sum, my dissertation makes five major contributions:

1. There is a signicant division between early research in intuitive physics and more recent investigations. This is largely due to a discrepancy between the reasoning systems utilized in each era's tasks; namely, explicit reasoning in early (pencil-and-paper)

tasks, and implicit reasoning in recent (prediction and recognition) ones. While many earlier results have been explained by modern computational approaches, a number of findings remain unexplained. My dissertation provides a review of early intuitive physics research and places it in the context of modern computational approaches utilizing Bayesian statistics and machine learning. Critically, my dissertation outlines the shortcomings of each approach and the hurdles that must be overcome before an intelligent and generalizable intuitive physics architecture can be developed.

2. We examine how humans perceive and represent observable information in dynamic, two-body object collision situations. The hypothesis that humans perceive motion in relation to meaningful|sometimes moving|landmarks in space is tested under the

noisy Newton framework for mass and causal inference. We term this the relative noisy Newton model, with the key distinction that the slow motion prior for motion perception is evaluated with respect to stationary points in the local environment.

3. We report human experiments demonstrating the viability of virtual reality (VR) technology in intuitive physics research. VR technology allows for the construction of novel environments whose characteristics differ from the real world in ways that preclude laboratory experimentation. One manipulation explored in my dissertation is the variation of gravitational acceleration in projectile motion situations. We show that the influence of unfamiliar gravity fields on human reasoning varies based on task demands. Shortcomings of current VR technologies are further discussed.

4. We introduce the intuitive substance engine (ISE) for explaining human reasoning about non-solid substance dynamics through mental simulation. This model is formed under the noisy Newton framework and provides converging evidence that people reason about complex physical situations by propagating noisy representations forwards in time using approximate, Newtonian principles. We explore whether people utilize coarse perceptual approximations of non-solid substance volumes when making their predictions; specifically, by representing substances as collections of rigid balls. We further propose specific task and stimulus characteristics which facilitate scene representation and subsequent mental simulation.

5. We explore the benefits of animated instructional materials in demonstrating higher-level concepts based on low-level physical interactions. We report a moderated mediation (path analysis) model exploring how individual differences in fluid intelligence

impact comprehension of source information and subsequent transfer to novel target problems. The reported findings are the first to explore the interaction between fluid intelligence and spontaneous analogical transfer.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View