X hits on this document





214 / 247



it scientifically useless. Some also reject the validity of the Turing test for judging intelligence at all. But the test clearly has meaning, and enormous historical,intuitive, and emotional appeal.It is hard to avoid the conclusion that if the experts could have created a worthy con- versational partner, they would have done so, and happily announced itif not through the Loebner event, then in some other venue.

So the linguistic intelligence required to pass the Turing test re- mains elusive. Ray Kurzweil thinks the test will be passed by the year 2029, perhaps by one of his $1,000 equivalents to the human brain, but gives few specifics.Two avenues, however, are natural to pursue. One possibility depends on the fact that verbal communication car- ries more information than written forms: it is the idea of reducing the ambiguity of human speech by prosodic analysis, whichas we have seenis already under development.

The second possibility, which has implications for artificial intel- ligence in general,not only its linguistic component, is to enormously expand the databases an artificial being needs for intelligent conversa- tion. One necessary database is the speech corpus, which determines how many words the being recognizes and can say; the other is a database of general knowledge, essential to converse with humanlike diversity. Both can now be established at huge sizes, terabyte upon terabyte, without storing them within every artificial being, because they could be accessed from the Internet by any being with a high- speed wireless connection.

According to some researchers, a database of general knowledge is an absolute prerequisite for artificial intelligence in its broadest sense. As Roger Schank and Lawrence Birnbaum of Northwestern Univer- sity have put it,

The truth is that size is at the core of human intelligence. . . . In order to get machines to be intelligent they must be able to access and modify a tremendously large knowledge base.There is no intelligence without real, and changeable, knowledge.

Establishing sufficiently large databases, however, is still only the be- ginning:We do not yet know how to make a synthetic being hear any human comment and find among its databases a response that is rel-

Document info
Document views836
Page views836
Page last viewedSat Jan 21 13:44:08 UTC 2017