Allen Newell , Unified of Cognition * Theories

Among the classics of AI is GPS, the General Problem Solver (Newell, Shaw and Simon [17] in 1959). Its key notion was that solving a problem consisted in finding a sequence of operators for transforming the present state into some goal state. A given problem area would be characterized by a finite set of differences and an operator-difference table. Given two states, one would compute the differences to be reduced between them, and then, for each difference, consult the table to find an operator that had proven fairly reliable for reducing that difference in a variety of situations. However, the given operator was not guaranteed to reduce the difference in all circumstances. GPS then proceeds with search by a process called means-end analysis: pick a difference, apply an operator, look at the new goal and the new state, look at the new difference and so on until the goal is attained. The challenge is to keep track of these states, operators, and differences to extract a "good" path, i.e., a relatively economical sequence of operators which will carry out the desired transformation. More than a decade after developing GPS, Newell and Simon (1972) published a huge book called Human Problem Solving [18] in which they