The Machine Learning Wave (1990s-2000s)

Rather than trying to encode intelligence as rules, a different approach gained ground: let computers learn patterns directly from data. Machine learning - the idea that algorithms could improve through experience rather than explicit programming - had existed since the 1950s, but it came into its own in the 1990s and 2000s as data became cheaper and computers got faster. Support vector machines, random forests, and other statistical techniques proved remarkably effective at practical tasks like spam filtering, credit scoring, recommendation engines and fraud detection. This was quieter, less glamorous AI - no one was claiming machines would think like humans. Instead, the focus was on solving specific, bounded problems with measurable accuracy. The internet was crucial here, generating enormous datasets that these algorithms needed to work well. Google's search engine, Netflix's recommendation system and Amazon's product suggestions all relied on machine learning long before most people had heard the term. This era established a pattern that still holds: AI works best when you have lots of relevant data, a clearly defined problem, and realistic expectations about what "good enough" looks like.