The first AI winter came when machines couldn't learn

After the optimism of the 1950s and 1960s — when researchers promised intelligent machines within a decade — AI funding collapsed in the 1970s and again in the 1980s, periods known as 'AI winters'. The core problem was that rule-based systems were brittle: they couldn't generalise to new situations or learn from data. The winters ended only when neural networks, large datasets, and GPU computing converged in the 2010s to make machine learning practical. The history serves as a warning: AI hype cycles are older than most people think.

Source: Russell & Norvig, Artificial Intelligence: A Modern Approach (4th ed., 2020)