Consistent physics underlying ballistic motion prediction

Consistent physics underlying ballistic motion prediction Kevin A Smith (k2smith@ucsd.edu), 1 Peter Battaglia (pbatt@mit.edu), 2 Edward Vul (evul@ucsd.edu) 1 1. University of California, San Diego, Department of Psychology, La Jolla, CA 92093 2. MIT, Department of Brain and Cognitive Sciences, Cambridge, MA 02139 Abstract Research into human models of intuitive physics typically falls into one of two camps, either claiming that intuitive physics is biased and not representative of real physics, or claiming that it consists of a collection of veridical physical laws. Here we investigate the causes of this tension, suggesting that prediction is based on real physics, but explanation is susceptible to biases. We gave participants three tasks based on the same physical principles: two prediction tasks and one task that required drawing the future path of motion. We found distinct biases in all three tasks; however, the two prediction tasks could be explained by consistent application of real physical principles under uncertainty, while the drawing task produced many more idiosyncratic biases. This suggests that different tests of intuitive physics are capturing different types of knowledge about the world. Keywords: intuitive physics; uncertainty; ballistic motion prediction Introduction Classic studies have suggested that many people base their physical intuitions on incorrect and inconsistent physical theories (Anzai & Yokoyama, 1984; McCloskey, Caramazza, & Green, 1980). Others have reported that people are biased by surface-level differences between tasks (Kaiser, Jonides, & Alexander, 1986), and that their inferences about simple physical situations rely on shallow heuristics and are frequently mistaken (Proffitt & Gilden, 1989; Todd & Warren, 1982). However over the past few years, a number of researchers have explained human physical predictions using quantitative cognitive models that assume people have an accurate and consistent understanding of the laws of physics that they apply flexibly across tasks (Hamrick, Battaglia, & Tenenbaum, 2011; Sanborn, Mansinghka, & Griffiths, 2013; Smith & Vul, 2013; Teglas et al., 2011). We suggest that a core difference between the above studies is the task given to participants. Some have asked participants to make a single judgment about the future state of the world, for instance, the direction a tower of blocks will fall (Hamrick, et al., 2011) or where a ball will cross a line (Smith & Vul, 2013). In contrast, classic studies tap into explicit explanations of physics, through verbal problems (Anzai & Yokoyama, 1984) or line drawings of motion (McCloskey, et al., 1980). Here we argue that people can apply correct physical principles consistently to simulate the world forward; however, explicit explanations of how the world will unfold draw upon an idiosyncratic set of background knowledge. We assessed participants’ understanding of the movement of balls after they had fallen off of pendulums in three separate tasks: predicting where a ball would land if cut from a pendulum, deciding when to cut a pendulum string such that the ball would fall into a fixed bucket, and drawing the path of the ball after the string is cut. We picked these tasks because there is evidence that people understand the motion of pendulums (Pittenger, 1985, 1990) and can predict the motion of projectiles under gravity (Saxberg, 1987), both of which must be combined to determine the ultimate trajectory of the balls. But there is also evidence that people show systematic errors when asked to explicitly draw the path of the ball (Caramazza, McCloskey, & Green, 1981), and that these errors are attenuated with kinematic information (Kaiser, Proffitt, Whelan, & Hecht, 1992). The same physical principles apply to each of these tasks, and so in the present experiment we investigated whether the tasks that require implicit prediction (catching the ball and cutting the string) can be explained by veridical physical principles. We find that subjects’ performance on the catching and cutting tasks differs between the tasks, but in the tasks that involved perceptually guided movements the differences can be reconciled by considering a single, valid model of physics that incorporates the different sources of perceptual and motor uncertainty from each task. Conversely, the sketches based on explicit conceptualization were inconsistent and idiosyncratic. Experiment Methods Fifty-seven UC San Diego undergraduates (with normal or corrected vision) participated in this experiment for course credit. All were treated in accordance with UCSD's IRB protocols. Procedure Participants viewed a computer monitor from a distance of approximately 60cm, which initially depicted a ball swinging from a string, consistent with pendulum motion. At some point in time the string would be cut and the ball would be released, thus entering ballistic motion. Beneath the pendulum there was always a bucket, and in every trial the participant's goal was to cause the ball to drop into the bucket after being released. How they were allowed to interact with the scene differed between two tasks, which were organized into blocks that were randomized across participants. With the exception of one initial practice trial per task that familiarized participants with the task, the path of the falling ball was occluded in order to prevent