Artificial Intelligence: The Very Idea
Source: https://archive.org/details/artificialintell0000haug ↗
Philosophically the most serious book of the symbolic AI era.
Haugeland coined the term "GOFAI" — Good Old-Fashioned Artificial Intelligence — and gave the clearest account of what the symbolic programme actually claimed: that intelligence is formal symbol manipulation, and that a physical symbol system is both necessary and sufficient for general intelligent action.
The book is not a dismissal but a rigorous, sympathetic exposition followed by an equally rigorous examination of where the framework breaks down.
For understanding the philosophical commitments that shaped AI's first decades — and that still echo in debates about LLMs and reasoning — Haugeland remains essential.
Central argument
Haugeland argues that the symbolic AI programme rests on a specific and falsifiable philosophical claim: that intelligence is nothing more than formal symbol manipulation, and that any physical system capable of manipulating symbols according to rules is both necessary and sufficient for general intelligence. He coins 'GOFAI' not as a pejorative but as a precise label for this research commitment, then subjects it to rigorous internal critique — showing that the framework's own logic generates the frame problem and related failures, where formal systems cannot determine which facts are relevant without an infinite regress of further rules. The book's core finding is that GOFAI does not fail due to insufficient computing power but due to a structural mismatch between formal manipulation and the kind of embedded, context-sensitive understanding that intelligence actually requires.
Critique
Written in 1985, the book's critical horizon is bounded by connectionism's infancy and has no purchase on statistical learning, large-scale neural networks, or the emergent behaviours of LLMs — systems that sidestep the symbolic programme entirely without obviously solving the problems Haugeland identifies. A thoughtful reader might press whether the frame problem and the critique of formalism apply with equal force to sub-symbolic systems, or whether Haugeland's argument inadvertently inherits the assumption that 'real' intelligence must be explicable in principled terms at all. The book is rigorous about what GOFAI claimed but cannot adjudicate whether contemporary AI has dissolved these philosophical problems or merely hidden them beneath statistical scale.
Why it matters for product
Product leaders evaluating AI feature investments often conflate two very different bets: that a system can manipulate symbols reliably within a closed domain versus that it can reason flexibly in open-ended, context-dependent situations — Haugeland's distinction gives that intuition precise conceptual teeth. Understanding why GOFAI broke down at context and relevance helps CPOs ask the right scoping questions: not 'can the model do this task?' but 'how tightly can we bound the context this feature will encounter in production?' The frame problem, transplanted to product strategy, is fundamentally a discovery and edge-case problem — and recognising it as structural rather than solvable with more data changes how you staff, test, and set expectations with stakeholders.