GOFAI (Good Old-Fashioned AI)

GOFAI, short for Good Old-Fashioned AI, refers to the classical paradigm of artificial intelligence that dominated the field from the 1950s through the 1980s. The term was coined by philosopher John Haugeland in his 1985 book Artificial Intelligence: The Very Idea to distinguish symbolic, rule-based approaches from emerging connectionist alternatives. GOFAI systems represent knowledge through discrete symbols — words, predicates, logical expressions — and derive conclusions by manipulating those symbols according to formal rules of inference. Classic examples include expert systems, theorem provers, and planning algorithms, all of which encode human expertise as explicit if-then rules or logical axioms.

The core assumption underlying GOFAI is that intelligence can be fully captured through symbol manipulation: that thinking, at its heart, is a kind of formal computation over structured representations. This hypothesis, sometimes called the Physical Symbol System Hypothesis (articulated by Allen Newell and Herbert Simon), held that any system capable of general intelligent action must be able to create, modify, and interpret symbolic structures. In practice, this led to systems like MYCIN for medical diagnosis, DENDRAL for chemical analysis, and Prolog-based reasoning engines — tools that performed impressively within narrow, well-defined domains.

GOFAI's limitations became increasingly apparent as researchers attempted to scale these systems to real-world complexity. The "knowledge acquisition bottleneck" made it prohibitively expensive to hand-code rules for open-ended domains, and symbolic systems struggled with perception, ambiguity, and graceful degradation under uncertainty. These shortcomings fueled the rise of machine learning and neural networks, which learn representations from data rather than requiring explicit programming. The term GOFAI thus carries a retrospective, somewhat critical connotation — marking a paradigm that, while historically foundational, proved insufficient as a complete theory of intelligence.

Despite its decline as a dominant paradigm, GOFAI remains relevant in modern AI. Techniques like knowledge graphs, constraint satisfaction, automated planning, and formal verification all draw on symbolic traditions. Contemporary research in neurosymbolic AI actively seeks to combine the data-driven strengths of deep learning with the interpretability and structured reasoning of classical symbolic methods, suggesting that GOFAI's legacy is less a dead end than an unfinished foundation.