Essay.02-Questioning.the.Intuitive.Physics.Engine.An.Occam’s.Razor.View.with.Cognitive.Generalization-ZhiyuanZhang-2020011132
Keywords: Intuitive Physics Engine, model building
Abstract: Humans possess the ability to make fast, approximate estimations of physics, from predicting the falling direction of a tower to inferring the mass ratio of two colliding balls given the trajectories. Yet the mechanism under the hood is not known. A popular explanation is the Intuitive Physics Engine (IPE) hypothesis, which posits that there is a simulation engine of physics in our brain that is responsible for the predictions. However, in this paper we argue that this assumption is not necessary, and should be replaced by a more general model building process that is key to human intelligence. We first argue that the model building process is compatible with the IPE hypothesis but more general and simpler, thus we should use Occam's razor to remove the IPE hypothesis. Then we demonstrate that the model building process is consistent with many other findings, including evidence from language acquisition. Finally, we briefly discuss the problems of addressing the intuitive physics using pure deep-learning based methods and propose future directions of bringing an explicit model building process into deep learning.
Submission Number: 53
Loading