A Short History: From Expert Systems to Transformers

AI research began in the 1950s. The field has gone through booms and winters — periods of huge funding followed by collapse. Understanding this rhythm helps you calibrate today's excitement.

Act 1: Symbolic AI (1950s–1980s)

Early researchers believed intelligence was logic. They wrote programs that manipulated symbols according to formal rules. Expert systems of the 1980s, like MYCIN for medical diagnosis, encoded human expert knowledge as if-then rules. They worked for narrow problems but crumbled outside them.

• 1956: Dartmouth Conference coins the term artificial intelligence
• 1966: ELIZA simulates a therapist with simple text patterns
• 1980s: expert systems boom, then bust due to brittleness

Act 2: Statistical ML (1990s–2010)

Researchers pivoted to data-driven methods. Support vector machines, decision trees, and shallow neural networks dominated. IBM's Deep Blue beat Kasparov at chess in 1997, but it was hand-tuned search, not general intelligence.

Act 3: Deep Learning (2012–2017)

In 2012, a neural network called AlexNet won the ImageNet competition by a huge margin, kicking off the deep learning revolution. GPUs, big datasets, and backpropagation combined to finally make deep networks trainable. By 2016, AlphaGo beat the world champion at Go, a feat nobody thought was close.

Act 4: The Transformer Era (2017–today)

The 2017 paper Attention Is All You Need introduced the transformer architecture. It replaced the recurrent networks used for language with a simpler, more parallel structure. Every modern LLM — GPT, Claude, Gemini, Llama — is a transformer at heart.

“AI winters end not with a new theory, but with enough compute.”

The big idea: today's AI is the fourth major wave, built on GPUs, internet-scale data, and the transformer. Knowing the cycle helps you see that hype is not new, but neither is real progress.