Description
The question of whether computers and robots will ever be able to think as well as we do has occupied researchers working in Artificial Intelligence for more than 50 years, ever since Alan Turing (1950) proposed his behavioral test. In this test, a computer tries to imitate a human being’s use of natural language in a conversation carried on via a teletype machine. The human participant in this exchange is required to decide whether he is conversing with another human being or with a computer. This test has been criticized by Searle (1980), who pointed out that similar behaviors, in themselves, need not imply that similar underlying mechanisms are governing these behaviors. This criticism is quite general and applies to computational models of all cognitive phenomena including visual perception. This lecture describes a new way of explaining percepts, a way that will actually tell us what a human observer or a computer is actually seeing. This will be explained with examples taken from the perception of 3D shapes. The essence of my new approach to a valid Turing-type test for perception is that it must be based on fundamental principles used to explain phenomena in the “hard” natural sciences, such as physics, rather than being based on, and confined to, observable behavior as has been done till now. These fundamental principles are: symmetry, least-action principle and conservation laws.
Location
STEW206
Start Date
9-24-2015 4:30 PM
DOI
10.5703/1288284315986
Symmetry Provides a Turing-Type Test for 3D Vision
STEW206
The question of whether computers and robots will ever be able to think as well as we do has occupied researchers working in Artificial Intelligence for more than 50 years, ever since Alan Turing (1950) proposed his behavioral test. In this test, a computer tries to imitate a human being’s use of natural language in a conversation carried on via a teletype machine. The human participant in this exchange is required to decide whether he is conversing with another human being or with a computer. This test has been criticized by Searle (1980), who pointed out that similar behaviors, in themselves, need not imply that similar underlying mechanisms are governing these behaviors. This criticism is quite general and applies to computational models of all cognitive phenomena including visual perception. This lecture describes a new way of explaining percepts, a way that will actually tell us what a human observer or a computer is actually seeing. This will be explained with examples taken from the perception of 3D shapes. The essence of my new approach to a valid Turing-type test for perception is that it must be based on fundamental principles used to explain phenomena in the “hard” natural sciences, such as physics, rather than being based on, and confined to, observable behavior as has been done till now. These fundamental principles are: symmetry, least-action principle and conservation laws.
Infographic