Dartmouth and the AI Winters
Level 4 · AI history
How AI began at a summer meeting in 1956. What the first learning machine could do, and why the money for AI dried up twice.
Course contentsLesson 8 of 10
Level 1 · What AI is
Level 2 · AI built from rules
Level 3 · Kinds of AI
Level 4 · AI history
Level 5 · Big questions
After this lesson you can
- Say when and where AI began as a field
- Explain what the perceptron could and could not do
- Describe what causes an AI winter
12 mincovers 3 glossary entries
Worth reading first: Turing Test and Chinese Room
How AI began, and why it froze twice
AI became a field of study at a summer meeting at Dartmouth College in 1956. The plan, written in 1955, asked for ten people and two months. It said every part of learning could, in principle, be described so exactly that a machine could copy it. The plan also gave the field its name.
The people behind it, among them John McCarthy, Marvin Minsky and Claude Shannon, led AI for decades. They did not build a working AI that summer. They mostly swapped ideas. But hopes ran high, and some said a machine would match a person within a generation.
In 1958, Frank Rosenblatt showed the perceptron, one of the first machines that learned from examples. It added up its inputs, each with its own weight, and said yes if the total was big enough. When it got an answer wrong, it moved its weights a little. The machine he built soon after saw through 400 light sensors.
A single perceptron has a hard limit. It can only split things with a straight line. In 1969, Minsky and Seymour Papert wrote a book showing simple tasks it could never learn. Interest in learning machines dropped for years, though later networks with many layers got past that limit.
An AI winter is a time when money and interest in AI drop sharply. It comes when promises run far ahead of results. The first began in the mid 1970s. In 1973, a report for the UK government said AI had not met its grand claims, and funding was cut there and in the US.
The second winter began in the late 1980s. Firms had spent heavily on expert systems and the special computers that ran them. The systems cost a lot to keep up and broke on new cases. In 1987 the market for those computers fell apart, and AI became a word to avoid.
People often think AI research stopped in a winter. It did not. Work went on with far less money, and some of the ideas that later won, like networks with many layers, grew in those cold years. The lesson of the winters is about hype: promise less than you can show.
Worked example
Jack's bank buys an expert system
In 1985, Jack's bank buys an expert system that judges loans with 2,000 rules. Two staff are hired just to keep the rules up to date.
Each new kind of loan needs dozens of new rules, and some clash with old ones. By 1990 the bank turns it off. Thousands of firms had the same story, and that is what the second winter looked like.
The words for this lesson3
Check yourself
What causes an AI winter?
Finished this lesson?
Marking it complete also marks its 3 glossary entries learned, which updates your roadmap and your saved page. 0 are already marked.