Several years back I read a chapter from "Brainchildren" by Daniel C. Dennett.
According to Dennett, Turing's test was meant
to be a show stopper for those endless
philosophical discussion about what intelligence is. Turing's idea works just like auditing for an orchestra. In the case of orchestra, the candidate should only be judged based on his/her musical performance rather than his/her appearance. So if you want to judge whether a machine is intelligence, we can reuse the same idea: put that "thing" into a black box and talk to it. In the case of Turing's test, the ability to perform intelligence conversation is used as a judging criteria for intelligence.
As you probably figure out now, Turing's test is not meant to be "fool-proof". Rather, it is a test that is meant to be good enough. If a machine passes the Turing's test, then it is reasonable to consider it as intelligence. But not all intelligence machines will necessary pass the Turing's test. Also, keep in mind that the test is meant to be executed a lot of times by different judges asking questions which could be simply "off-the-wall" (i.e. out of the machine's areas of expertise). Passing the "pure" Turing's test is not as easy as it seems.
Turing is no fool. At least, he already accounts for inhuman-like intelligence in his infamous test.
Slashdot Mirror
More ironic that AI people need a book to defend their own field. Go check out McCarthy's book Defending AI Research: A Collection of Essays and Reviews .
Several years back I read a chapter from "Brainchildren" by Daniel C. Dennett. According to Dennett, Turing's test was meant to be a show stopper for those endless philosophical discussion about what intelligence is. Turing's idea works just like auditing for an orchestra. In the case of orchestra, the candidate should only be judged based on his/her musical performance rather than his/her appearance. So if you want to judge whether a machine is intelligence, we can reuse the same idea: put that "thing" into a black box and talk to it. In the case of Turing's test, the ability to perform intelligence conversation is used as a judging criteria for intelligence.
As you probably figure out now, Turing's test is not meant to be "fool-proof". Rather, it is a test that is meant to be good enough. If a machine passes the Turing's test, then it is reasonable to consider it as intelligence. But not all intelligence machines will necessary pass the Turing's test. Also, keep in mind that the test is meant to be executed a lot of times by different judges asking questions which could be simply "off-the-wall" (i.e. out of the machine's areas of expertise). Passing the "pure" Turing's test is not as easy as it seems.
Turing is no fool. At least, he already accounts for inhuman-like intelligence in his infamous test.