A short, honest history of AI

In 1956, a small summer workshop at Dartmouth College gave “artificial intelligence” its name and a bold prediction: significant progress within a generation. The researchers were right about the destination and very wrong about the schedule — a pattern AI would repeat for seventy years.

Wave one: rules (1950s–1980s)

The first approach was to write intelligence down: if the patient has a fever and a stiff neck, then consider meningitis. These “expert systems” genuinely worked in narrow domains — but the real world kept producing situations nobody had written a rule for. Funding and enthusiasm collapsed twice into what the field calls AI winters.

Wave two: learning from data (1990s–2000s)

The next idea flipped the approach: stop writing rules, start showing examples. This is machine learning — the spam-filter idea from lesson 1. It quietly powered credit scoring, search ranking, and product recommendations for two decades without most people calling it “AI” at all.

Wave three: deep learning (2012–)

Around 2012, three things matured together: much bigger datasets, much faster chips, and better ways to train neural networks (we’ll build a friendly picture of those in lesson 4). Suddenly computers could recognise images, transcribe speech, and translate language at useful accuracy.

Wave four: generative AI (2020s)

The current wave doesn’t just recognise content — it generates it. Models trained on enormous amounts of text and images learned to produce new text, images, code, and audio. That’s the technology behind the chat assistants this course will teach you to use well.

One honest note: every previous wave brought real progress and inflated promises. Expect the same now. This course will keep showing you what today’s AI genuinely does well, and where it still fails.

Check your understanding

1. What caused the 'AI winters'? Computers were banned in the 1970s Expectations ran far ahead of what the technology could deliver Researchers lost interest in intelligence

2. What separates generative AI from earlier machine learning? It creates new content instead of only labelling or scoring existing content It follows more rules It never makes mistakes

Recap

Rules, then learning, then deep learning, then generation — each wave solved the last wave’s bottleneck. Next: what “machine learning” actually means, with no maths required.