Early AI: Symbolic Intelligence
This text is a preview of History of the Reception of Hegel, volume II, to be published in 2026 by Reflexivity Press.
The Dartmouth Project, 1956
The term “Artificial Intelligence” was coined in 1956 at a conference in Dartmouth. John McCarthy, Marvin Minsky, and others gathered with a bold vision: within a generation, machines would be able to do everything humans can do — think, learn, be creative.
The program was ambitious: machines were to play chess, prove theorems, understand language, recognize images. The approach was symbolic: intelligence consists in manipulating symbols according to rules. The brain is a kind of computer that processes symbols.
The early successes were impressive. Programs proved mathematical theorems. Chess computers beat amateur players. Expert systems diagnosed diseases. The euphoria was great: another ten, twenty years, and we would have “thinking” machines.
The Limits of Symbolic AI
But the limits soon became apparent. Chess is a closed system with clear rules — computers excel there. But the “open world” is different. A child effortlessly recognizes a face, understands a story, moves through a room. For computers, these were intractable problems.
The “frame problem” became central: how does a system know what changes and what stays the same when it acts? If I move a cup, the room stays the same, but the position of the cup changes. This is trivial for humans. For symbolic AI it was a nightmare: one would have to explicitly specify what does not change — and that is potentially infinite.
Hubert Dreyfus, a philosopher and scholar of Heidegger, criticized symbolic AI at its foundation: it misses what intelligence actually is. Intelligence is not symbol manipulation but embodied being-in-the-world. The body, the situation, the context — all of this is lost in symbolic AI.