AI has gone through “winters” before

The predictions

The founders of AI were brilliant, and very optimistic:

• 1958, Simon and Newell: within ten years a computer would be world chess champion.
• 1965, Herbert Simon: “machines will be capable, within twenty years, of doing any work a man can do.”
• 1967, Marvin Minsky: “within a generation … the problem of creating ‘artificial intelligence’ will substantially be solved.”
• 1970, Minsky in Life magazine: “in from three to eight years we will have a machine with the general intelligence of an average human being.”

Early programs that solved logic puzzles or algebra problems looked like the start of something enormous. But they worked only in small, tidy worlds. Ordinary things like understanding language, recognising objects and common sense turned out to be far harder.

The first winter (about 1974–1980)

In 1973, the British government commissioned the mathematician Sir James Lighthill to assess AI. His report criticised the field for failing to achieve its “grandiose objectives”. UK funding was cut sharply. In the US, DARPA cut back on open-ended research, including a $3 million-a-year speech-understanding programme. Researchers called this period the AI winter.

Boom and second winter (1980s–1990s)

In the 1980s, expert systems, programs encoding the rules of human specialists, became a billion-dollar industry. But they were expensive to build and brittle when facing anything unexpected. In 1987 the market for specialised Lisp machines collapsed, and a second winter followed. Through the 1990s, many researchers avoided the term “AI” and called their work machine learning, informatics or knowledge systems.

The lesson

The winters weren’t because the ideas were worthless. Many of them, neural networks included, later succeeded. They came from a gap between promises and delivery. Whether today’s boom will be followed by another winter is one of the most argued-about questions in tech.