From Logic Gates to ChatGPT: A Brief History of AI

If AI feels like it appeared overnight in 2022, it's worth knowing the idea is older than your grandparents. Researchers were building "thinking machines" in the 1950s. So why did it take 70 years to get from a chalkboard idea to a chatbot in your pocket? The short answer: the early approach was fighting the wrong battle.

The first wave, symbolic AI, tried to hand-code intelligence directly — programmers wrote out explicit rules and logical facts ("if the patient has a fever AND a rash, THEN consider measles") to build expert systems that mimicked specialists. This worked surprisingly well in narrow, rule-heavy domains, but it hit a wall: the real world is full of fuzzy, exception-riddled situations that nobody can fully write down as if/then rules. Funding dried up — twice — in periods historians now call AI winters, when the hype outran the results and investment collapsed.

The breakthrough wasn't a smarter rule-writer. It was a shift to machine learning: instead of programming the rules by hand, you show the system thousands (or millions) of examples and let it find the patterns itself. That idea existed for decades too, but it needed two things the 1980s simply didn't have enough of: data and compute. The internet generated an ocean of text, images, and clicks to learn from, and GPUs (originally built for video game graphics) turned out to be extremely good at the math behind learning patterns at scale.

Put those two ingredients together with deep learning — machine learning using many-layered neural networks, which we'll unpack tomorrow — and you get the generative AI boom: systems that don't just classify or predict a number, but generate new text, images, and audio that didn't exist before. ChatGPT wasn't a sudden spark of genius; it was the payoff of 70 years of theory finally meeting enough data and enough computing power to actually run it.

Keep that pattern in mind for the rest of this course: almost every "sudden" AI breakthrough you'll read about is really an old idea meeting new data or new hardware, not a brand-new idea out of nowhere.