Backpropagation 1986: the algorithm that reopened deep networks
Rumelhart, Hinton, and Williams popularized gradient descent via error backpropagation in multilayer networks.
1956–1986
Symbolic AI
Expert systems, logic, and hand-coded knowledge — the era of symbolic reasoning.
Hopfield networks (1982): associative memory and statistical physics
John Hopfield linked recurrent nets to spin systems, inspiring a wave of energy-based models in AI.
Japan’s Fifth Generation project (1982): a bet on Prolog and logic
MITI launched FGCS to lead “intelligent” computers based on logic programming; the real outcome was mixed and costly.
Edward Feigenbaum and knowledge engineering
Feigenbaum argued that expert AI power lies in encoded domain knowledge, not only generic search mechanisms.
DENDRAL and MYCIN: the rise of expert systems
1970s–80s expert systems captured domain knowledge in rules; DENDRAL and MYCIN were emblematic cases.
The first AI winter (1970s): promises vs. reality
British government cuts (Lighthill report) and DARPA pullbacks chilled AI after oversized 1960s expectations.
SHRDLU (1970): natural language in a blocks world
Terry Winograd’s SHRDLU combined parsing, planning, and a micro-world to converse about manipulable blocks.
Minsky and Papert 1969: Perceptrons and the brake on connectionism
The book Perceptrons argued severe limits of single-layer networks and helped eclipse neural nets for a time.
Rosenblatt’s Perceptron (1958): learning in hardware
Frank Rosenblatt introduced the Perceptron, a learning machine that adjusts weights from examples—an emblem of the first connectionist boom.
Dartmouth 1956: the term Artificial Intelligence is born
The 1956 Dartmouth workshop, proposed by John McCarthy and colleagues, coined “artificial intelligence” and launched the field.