GOFAI, short for Good Old-Fashioned Artificial Intelligence, is the name philosopher John Haugeland gave to classical symbolic AI in his 1985 book “Artificial Intelligence: The Very Idea” (MIT Press). The term is verified here through John McCarthy’s own review of the book, hosted on his Stanford site, in which McCarthy discusses Haugeland’s use of “GOFAI” and his characterization of the field. The label has since become the standard shorthand for the symbol-processing tradition that dominated AI from the 1950s through the 1980s.
The defining commitment of GOFAI is that intelligent behavior can be produced by manipulating symbols according to formal rules. Knowledge is represented as explicit structures, such as logical sentences, frames, or production rules, and reasoning is the systematic transformation of those structures by search and inference. This is the world of the Logic Theorist, the General Problem Solver, expert systems, planners, and logic programming, all resting on the physical symbol system hypothesis that Newell and Simon had articulated. Haugeland summarized the underlying bet as the claim that human thinking and machine computing are, at bottom, radically the same.
The term carries an edge, because it was coined just as the approach was being challenged. Connectionists argued that intelligence emerges from sub-symbolic networks rather than explicit rules; researchers in situated and embodied robotics argued that intelligence comes from interaction with the world rather than internal symbol crunching. Calling the older paradigm “good old-fashioned” marked it as one option among several, not the whole of AI. Notably, McCarthy disputes Haugeland’s framing, arguing the real claim is that intelligent behavior can be realized computationally, not that intelligence simply equals computation.
Why a business reader should care: today’s most visible AI is statistical and connectionist, but a great deal of deployed, dependable software, rules engines, planners, configurators, knowledge graphs, is squarely GOFAI, and knowing the distinction helps separate what symbolic methods do well from what learning-based methods do well.