Behaviorism, finite automata, and stimulus response theory (Q1222627)

The paper under review is prompted by an apparent contradiction between a result previously obtained by the author [``Behaviorism is false''. J. Philos. 66, 417--452 (1969)], and one of \textit{P. Suppes} [J. Math. Psychol. 6, 327--355 (1969; Zbl 0184.45701)]. Suppes had argued that any ``connected finite automaton'' can be asymptotically represented by a purely stimulus-response learning model (i.e. by a ``behavioristic'' model, with no reference to internal states). On the other hand, the author had argued that there exist automaton behaviors which cannot be described without explicit reference to internal states, and concluded that, to the extent that animal behavior can be modelled by automata, behaviorism is false. In the present paper, the discrepancy between the two results is removed. The author shows that the automata considered by Suppes are of a special type, which he calls ``finite-state language acceptors''. These are state-output automata with a prescribed initial state and a set of distinguished final states; such automata can ``recognize'' particular sets of input strings, which are in fact precisely those generated by a finite-state grammar. On the other hand, a more general finite automaton is called by the author a ``finite state transducer''; such automata need not be in a distinguished initial state, and possess a separate output function. The paper concludes with a very interesting discussion of the relation of psychology to biology and physics, and what is to be expected of a theory of psychological processes.