GOFAI, or Good Old-Fashioned AI, treats intelligence as reasoning over discrete, compositional symbols with explicit rules. The physical symbol system hypothesis, associated with Allen Newell and Herbert Simon, places symbols at the root of intelligent action. section 7.6
Its rationalist commitment is that reasoning should be specified in logic. This makes structure explicit, but can produce brittle behavior when real-world details or connections between symbols and the world are underspecified. section 7.6
GOFAI sits within a broader dispute about how intelligence is structured:
- Connectionism learns statistical regularities and distributed representations through gradient-based methods rather than prespecifying symbolic rules. section 7.6
- Fodor and Pylyshyn argued that compositional thought and language require an architecture that explains their structure. section 7.6
- Smolensky countered that symbolic-like rules might emerge as higher-level descriptions of lower-level soft constraints. section 7.6
Benchmarks over philosophical wagers
The Malik–Hinton wager illustrates a pragmatic test: Malik demanded results on PASCAL, while Hinton pointed to ImageNet, where AlexNet supplied a measured benchmark gain for neural networks. section 7.6
