The philosophical problems that have to be solved will be clearer in connection with a particular kind of proposed intelligent program, called a reasoning program or RP for short. RP interacts with the world through input and output devices some of which may be general sensory and motor organs (for example, television cameras, microphones, artificial arms) and others of which are communication devices (for example, teletypes or keyboard-display consoles). Internally, RP may represent information in a variety of ways. For example, pictures may be represented as dot arrays or a list of regions and edges with classifications and adjacency relations. Scenes may be represented as lists of bodies with positions, shapes, and rates of motion. Situations may be represented by symbolic expressions with allowed rules of transformation. Utterances may be represented by digitized functions of time, by sequences of phonemes, and parsings of sentences.
However, one representation plays a dominant role and in simpler systems may be the only representation present. This is a representation by sets of sentences in a suitable formal logical language, for example -order logic with function symbols, description operator, conditional expressions, sets, etc. Whether we must include modal operators with their referential opacity is undecided. This representation dominates in the following sense:
1. All other data structures have linguistic descriptions that give the relations between the structures and what they tell about the world.
2. The subroutines have linguistic descriptions that tell what they do, either internally manipulating data, or externally manipulating the world.
3. The rules that express RP's beliefs about how the world behaves and that give the consequences of strategies are expressed linguistically.
4. RP's goals, as given by the experimenter, its devised subgoals, its opinion on its state of progress are all linguistically expressed.
5. We shall say that RP's information is adequate to solve a problem if it is a logical consequence of all these sentences that a certain strategy of action will solve it.
6. RP is a deduction program that tries to find strategies of action that it can prove will solve a problem; on finding one, it executes it.
7. Strategies may involve subgoals which are to be solved by RP, and part or all of a strategy may be purely intellectual, that is, may involve the search for a strategy, a proof, or some other intellectual object that satisfies some criteria.
Such a program was first discussed in McCarthy (1959) and was called the Advice Taker. In McCarthy (1963) a preliminary approach to the required formalism, now superseded by this paper, was presented. This paper is in part an answer to Y. Bar-Hillel's comment, when the original paper was presented at the 1958 Symposium on the Mechanization of Thought Processes, that the paper involved some philosophical presuppositions.
Constructing RP involves both the epistemological and the heuristic parts of the artificial intelligence problem: that is, the information in memory must be adequate to determine a strategy for achieving the goal (this strategy may involve the acquisition of further information) and RP must be clever enough to find the strategy and the proof of its correctness. Of course, these problems interact, but since this paper is focused on the epistemological part, we mention the Missouri program (MP) that involves only this part.
The Missouri program (its motto is, `Show me') does not try to find strategies or proofs that the strategies achieve a goal. Instead, it allows the experimenter to present it proof steps and checks their correctness. Moreover, when it is `convinced' that it ought to perform an action or execute a strategy it does so. We may regard this paper as being concerned with the construction of a Missouri program that can be persuaded to achieve goals.