SOAR

SOAR (State, Operator, And Result) is a cognitive architecture developed by Allen Newell and his collaborators at Carnegie Mellon University, representing the most ambitious attempt to build a unified theory of cognition on purely symbolic foundations. SOAR posits that all intelligent behavior — from routine motor control to scientific discovery — can be decomposed into search through a problem space, with learning occurring through the automatic caching of successful sequences into long-term production rules.

The architecture's central commitment is universal subgoaling: whenever a problem cannot be solved directly, SOAR automatically creates a subgoal to acquire the missing knowledge. This mechanism was meant to explain the full range of intelligent behavior without recourse to specialized modules or innate structures. But SOAR's very generality became its limitation: by making every cognitive act a form of search, it flattened the distinctions between perception, reasoning, and memory that empirical neuroscience reveals as genuinely different.

SOAR's failure was not computational but ontological. It assumed that cognition is fundamentally problem-solving, and that problem-solving is fundamentally search. Both assumptions are too narrow. Cognition is not search; it is the continuous reorganization of a system by its own history. The learning mechanism of chunking in SOAR has been compared to human skill acquisition, though the analogy remains controversial. See Chunking (learning) for the broader cognitive phenomenon.