Computing machinery and intelligence
Computers & thought
An Efficient Algorithm for Finding Certain Minimum-Cost Procedures for Making Binary Decisions
Journal of the ACM (JACM)
Recursive functions of symbolic expressions and their computation by machine, Part I
Communications of the ACM
| Hi-index | 0.03 |
Many workers have studied the problem of getting machines to exhibit aspects of intelligent behavior. It has become clear that a major limitation of the artificial intelligence field is the cost and difficulty of programming, which remains essentially a handicraft technique. What we need in artificial intelligence research are methods for the machine to be self-programming in a much deeper sense than the compilation process in use today. Briefly, we would like for the machine to "learn" to solve a problem, rather than being programmed in the usual laborious way. This report presents a computer program, using what I will call an "experience-driven compiler," which is able to program itself to solve a restricted class of problems involving cause-and-effect relationships. Although the program is not yet able to learn to solve problems of practical significance, it is hoped that the technique will be developed to the point where this can be done. I am in agreement with Turing's suggestion (1950) that the best way to achieve artificial intelligence is to find out what an infant has in his brain that allows him to become intelligent, put this capability into a computer, then allow the machine to "grow up" in much the same way as the baby does. (Of course, this is a statement of the problem rather than an explanation of how to solve it.) An additional purpose of the work is in psychology: we would like to have computer methods for testing theories of how learning might occur in living creatures.