Why AI exploded now, not in 1990
Alan Turing asked 'Can machines think?' in 1950 and proposed the imitation game we now call the Turing Test. Early symbolic AI could play checkers and prove theorems, but brittle rules collapsed in the messy real world. Funding collapsed twice - the famous 'AI winters'.
Deep Blue beat Kasparov in 1997 using brute search, not understanding. The real revolution was statistical: machine learning that learns rules from data instead of being handed them. By 2012, a deep neural network called AlexNet crushed an image competition, powered by GPUs and millions of labelled pictures.
A 2017 paper, 'Attention Is All You Need', introduced the transformer: an architecture that reads a whole sentence at once and weighs which words matter for each other. Transformers scale beautifully with data and compute, giving us GPT-style large language models. Three ingredients made the boom: abundant data, cheap parallel compute, and the transformer.
A machine passes if a human cannot reliably tell it from a human in conversation.
Periods when AI funding and interest collapsed after over-promising.
2017 architecture using attention; foundation of modern LLMs.
Performance improving as data, parameters and compute grow.
Tools: Paper and pen
