An approach to computers that perceive, learn, and reason

The purpose of this paper is to mention some of the problems which I believe will be central ones in the effort to design machines that can in some sense be said to perceive, think, learn, make reasonable inferences, or learn to perform acts of skill. First I wish to discuss a general conceptual outlook which has important consequences for our approach to the design of such machines. I shall show how this outlook calls for modifying or supplementing some current approaches; however, specific techniques for dealing adequately with the problems raised remain to be found. The second part of the paper will point to specific techniques which now exist for dealing with the problem of reasonable inference and inductive judgment, but which to my knowledge have not been considered in connection with computers and control mechanisms.