Mechanizing the Search for Explanatory Hypotheses

Five types of search are considered: random, exhaustive, selective, heuristic, and opportunistic search, which generalizes to some conclusions about mechanizing hypothesis formation.

Abstract

Mechanized methods in science have attracted much attention among philosophers since Bacon. Of the many facets of scientific activity discovery has seemed most inscrutable. Until recently, Peirce and Hanson were the only ones to claim publicly that there can be rational methods for discovering hypotheses as well as for testing them. Their arguments, and more recent ones, are based on historical examples and analysis, and thus lack the convincingness of an existence proof. In this paper I take an empirical look at the question of whether there are rational methods of discovery and claim that computer programs provide a laboratory for experimentation on this question. Recent work in artificial intelligence, or AI, has produced programs capable of serious intellectual work in science. Results from AI will be used to show that there exist mechanized procedures for discovering hypotheses and that these methods often lead to plausible hypotheses (but do not guarantee always finding the correct hypothesis).