Breakthroughs
The ideas and systems that changed what AI can do.
Neural networks and backpropagation
Networks of simple connected units that learn from examples. Backpropagation, popularised in the 1980s, made it practical to train networks with several layers.
Deep learning and ImageNet
With large datasets and graphics processors (GPUs), deep networks such as AlexNet in 2012 beat older approaches at recognising images.
Game-playing systems
Deep Blue showed that search at scale could beat a chess champion. AlphaGo combined deep networks with reinforcement learning to win at Go.
The Transformer
Introduced in 2017, it uses attention to relate every part of a text to every other part and trains efficiently on large amounts of data.
Large language models
Transformers trained on huge amounts of text can write, summarise, translate and answer questions. ChatGPT in 2022 made them widely known.
AlphaFold
DeepMind's system predicts protein structures with high accuracy. Demis Hassabis and John Jumper shared the 2024 Nobel Prize in Chemistry for it, with David Baker for protein design.