6.7 Discovery as heuristic search

A long AI tradition treats discovery as search through a structured space of concepts, hypotheses, or programs. Lenat’s AM explored elementary mathematics using a growing collection of concepts and heuristic rules [ Lenat , 1976 ] . The BACON family sought empirical laws by applying heuristics to numerical regularities, and the broader computational account of scientific discovery linked representation, search, and theory formation [ Langley et al. , 1987 ] . Genetic programming and later symbolic regression continued the program by searching executable expression spaces; Schmidt and Lipson showed that compact free-form laws could be recovered from experimental data without supplying the target equations [ Schmidt and Lipson , 2009 ] .