香蕉视频

Prof. Roger Brockett

Harvard University

听

One of the weak points of the long debated Turing definition of machine intelligence is that it is based completely on language. Even though animals that appear not to use language exhibit behavior thought to be intelligent, Turing's definition leaves them out of the picture.听 Of even more importance, Turing's test occurs in an entirely sterile environment, providing no acknowledgment of the skills required to make use of the environment to solve problems.听听 In this talk it will be argued that a definition of machine intelligence based on the degree to which machines are able to solve problems by partnering with their environment serves us better in our efforts to听 understand "intelligence" in a robotic context.听 This partnership may include using the environment to supplement motor skills, to calibrate capabilities, to test the accuracy of memory, and to experiment with possible courses of action. We intend to give arguments that support the idea that a machine can be said to be intelligent if, when it is placed in a partially unknown environment, it is able to extract sufficient tools and information from the environment to enable it to achieve its goals.听 The practical implications of this point of view for the design of robots will be emphasized.