Subject: Cog Lunch Tuesday

Date: Monday, February 27, 2012 2:21 AM

From: Sam Norman-Haignere <>

To: <>, <>

Conversation: Cog Lunch Tuesday


Hi All,


Peter Battaglia from the Tenenbaum lab will be speaking at Cog Lunch this Tuesday at noon. The talk will be in room 3310 (on the Picower side of the building), and there will be food from Cafe Luna.


Title: Intuitive Mechanics in Physical Reasoning


Abstract: People rely on their "physical intelligence" -- the ability to infer physical properties of objects and predict future states in complex, dynamic scenes -- to interpret their surroundings and plan safe and effective actions. For instance, you can choose where to place your coffee to prevent it from spilling, arrange books in a stable stack, and shoot billiard balls to cause desired sequences of collisions. These behaviors suggest the brain performs sophisticated reasoning using rich physical knowledge, but the specific content of this knowledge, and how it is applied, remain unclear. Previous work has focused on identifying biases and errors, or testing simple models on highly-specific judgments; here, we seek a unifying account that can quantitatively predict a broader spectrum of human abilities. 


This talk explores the idea that the brain has an "intuitive mechanics", a realistic model of physics that can estimate physical properties and predict probable futures. This intuitive mechanics is surprisingly faithful to the laws of classical mechanics, it captures statics, dynamics, forces, collisions, and friction. It is fundamentally probabilistic, it supports Bayesian inferences that robustly handle uncertainty, and, like people, its predictions sometimes deviate from objective reality. And, it is resource-bounded, supporting only judgments that can be made based on a few low-precision, short-lived simulation runs. We conducted a series of psychophysical experiments in which participants made physical judgments about various complex, 3D scenes, and found that a formal model of intuitive mechanics well-predicts people's responses by accounting for their accuracy and several systematic biases. These results suggest that an approximate, probabilistic model of physics forms the basis of human physical reasoning. More generally, this principled computational approach provides a unifying framework for analyzing and understanding this crucial part of human cognition.







Bcs-talks mailing list