The AI Winter: when everyone gave up on artificial intelligence

1974

In the 1960s, AI researchers made extraordinary promises. Herbert Simon predicted that within 20 years, machines would be capable of doing any work a man can do. Marvin Minsky told Life magazine that within 3 to 8 years, we'd have a machine with the general intelligence of an average human being. Funding poured in from DARPA and governments worldwide.

Then reality hit. The problems turned out to be exponentially harder than anyone imagined. Machine translation produced gibberish. Computer vision couldn't distinguish a dog from a cat. Expert systems were brittle and couldn't handle situations outside their narrow rules.

In 1973, the British government commissioned the Lighthill Report, which concluded that AI had failed to deliver on its promises. Funding was slashed across the UK. DARPA cut AI funding in the US. The first AI Winter had begun.

A brief thaw came in the 1980s with expert systems: rule-based programs that could diagnose diseases or configure computers. Companies spent billions. Japan launched the Fifth Generation Computer project to build thinking machines. AI was hot again.

Then the expert systems crashed too. They were expensive to maintain, couldn't learn, and broke when encountering situations their programmers hadn't anticipated. By 1987, the specialized AI hardware market collapsed. Lisp machine companies went bankrupt. The second AI Winter set in, lasting roughly from 1987 to 1993.

During the winters, researchers learned to avoid the term "AI" entirely. They rebranded their work as "machine learning," "data mining," "knowledge systems," or "computational intelligence." Funding proposals that mentioned AI were rejected. Grad students were warned that working on AI was a career dead end.

What eventually ended the winter wasn't a breakthrough in theory; it was data and compute. The internet generated massive datasets. GPUs provided parallel processing power. In 2012, a deep learning system called AlexNet crushed the ImageNet competition, and suddenly neural networks worked. The current AI boom is built on top of ideas from the 1980s that simply needed more data and faster hardware.

The AI winters teach an important lesson: hype cycles are dangerous. When technology is oversold, the inevitable disappointment can set an entire field back by decades. Today's AI researchers quietly worry about a third winter if current AI systems fail to deliver on the enormous expectations being set.