Page 198

		These results may be viewed as a fundamental limitation on our ability to build general purpose devices for empirical inquiry and motivate the use of heuristic methods in Artificial Intelligence. However, they also suggest a weaker criterion of identification in which a team of scientists is employed and success of the team is equated with the success of any member in the team. Let us illustrate this idea in the context of functions.

		Consider the collections of functions and in Proposition 9.1. Let M1 Ex-identify and M2 Ex-identify . Now, if we employed the team of M1 and M2 to identify and weakened the criterion of success to the requirement that success is achieved just in case at least one member in the team is successful, then under this changed criterion, the collection is identifiable by the team of M1 and M2. This idea can be extended to teams of n scientists, at least m of which are required to be successful.

		We now formalize function and language identification by teams. Anomalies in the hypothesized programs are also incorporated in the definitions.

		9.3 Definition (Smith [178]) Let m, and .

		(a) A team of n scientists M1, M2, . . ., Mn is said to (written: ) just in case f is Exa-identified by at least m of the Mi's.

		(b) is the collection of such that some team of n scientists each function in .

		9.4 Definition (Osherson, Stob, and Weinstein [140]) Let m, and .

		(a) A team of n scientists M1, M2, . . ., Mn, is said to (written: ) just in case L is TxtExa-identified by at least m of the Mi's.

		(b) is the collection of such that some team of n scientists each language in .

		For both and the fraction is referred to as the success ratio of the paradigm. As has been the practice, the cases a = 0 are simply referred to as and . Most of the results presented in the present chapter are for the case a = 0; other cases are considered in the exercises. Let us note that Definitions 9.3 and 9.4 fail to formalize one aspect of scientific practice that is central to