AI: The Tumultuous History of the Search for Artificial Intelligence

Crevier, who worked in machine vision, wrote a history of artificial intelligence research from the 1950s to the early 1990s, based on interviews with most of the field's leading figures.

The narrative covers the founding period at the 1956 Dartmouth workshop, the early symbolic programs including the Logic Theorist and the General Problem Solver, the optimism of the 1960s and the forecasts that turned out to be badly wrong, the Lighthill report in Britain and the funding cuts that followed, the first winter, the expert systems boom of the 1980s in which commercial applications encoded domain knowledge as rules, the Japanese Fifth Generation programme and the American response, the collapse of the expert systems industry, and the second winter.

The intellectual history covers the recurring dispute between symbolic approaches, which represent knowledge as explicit structures and reason over them, and connectionist approaches using neural networks, which learn from data. Crevier documents Minsky and Papert's Perceptrons and its effect on neural network funding, the backpropagation work that revived the approach in the 1980s, and the state of the argument as it stood when he wrote.

He also covers the practical lesson the field learned repeatedly: tasks that appear difficult to humans, such as chess and theorem proving, turned out to be tractable, while tasks any child performs, such as vision, language understanding and movement, proved extremely hard.

The book was published in 1993 and stops before the statistical machine learning turn, before the data and compute scale that made deep learning work, and before everything that followed. Its value now is as a record of the cycles of expectation and disappointment, which the field has repeated, and of how the researchers of that period understood their own problem.