REFERENCE

AI Timeline

Updated 2 May 2025
timelinehistorymilestones

AI Timeline

A chronological record of the key moments in artificial intelligence — from the theoretical foundations to the present day.


The Foundations (1940s–1960s)

YearEvent
1943McCulloch & Pitts publish first mathematical model of a neural network
1950Alan Turing publishes “Computing Machinery and Intelligence” — proposes the Turing Test
1956Dartmouth Conference — term “Artificial Intelligence” coined
1957Frank Rosenblatt builds the Perceptron (first trainable neural network)
1966ELIZA chatbot (MIT) — first conversational AI

The Winters (1970s–1990s)

YearEvent
1969Minsky & Papert publish “Perceptrons” — shows limitations, kills neural network funding
1974-1980First AI Winter — Funding dries up after overpromising
1986Backpropagation popularised (Rumelhart, Hinton, Williams)
1987-1993Second AI Winter — Expert systems fail to deliver
1989Yann LeCun demonstrates CNN for handwritten digit recognition
1997IBM Deep Blue defeats world chess champion Garry Kasparov

The Deep Learning Era (2000s–2010s)

YearEvent
2006Hinton’s “Deep Belief Networks” — deep learning renaissance begins
2009ImageNet dataset created (Fei-Fei Li) — benchmark that drives progress
2011IBM Watson wins Jeopardy!
2012AlexNet wins ImageNet — deep learning proves its power, GPU training works
2013Word2Vec (Mikolov, Google) — word embeddings work remarkably well
2014GANs introduced (Goodfellow) — generative AI begins
2014Sequence-to-Sequence (Sutskever, Google) — neural machine translation
2015OpenAI founded (Altman, Musk, Sutskever, et al.)
2015ResNet (152 layers) — proves very deep networks are trainable
2016AlphaGo defeats Lee Sedol — AI masters Go (DeepMind)

The Transformer Era (2017–2022)

YearEvent
2017“Attention Is All You Need” — Transformer architecture introduced (Google)
2018BERT (Google) — bidirectional pre-training, NLP breakthrough
2018GPT-1 (OpenAI) — generative pre-training for language
2019GPT-2 (OpenAI) — “too dangerous to release” (1.5B params)
2020GPT-3 (OpenAI) — 175B params, few-shot learning, scaling laws validated
2020AlphaFold 2 — Solves protein folding (DeepMind)
2021DALL-E (OpenAI) — text-to-image generation
2021GitHub Copilot launches — AI coding assistant
2022Stable Diffusion (Stability AI) — open-source image generation
2022ChatGPT launches (Nov 30) — Fastest-growing consumer app ever, triggers AI boom

The Current Era (2023–Present)

YearEvent
2023 JanMicrosoft invests $10B in OpenAI
2023 MarGPT-4 released — multimodal, major capability jump
2023 MarClaude (Anthropic) public launch
2023 MayGeoffrey Hinton resigns from Google, warns about AI risks
2023 JulLLaMA 2 (Meta) — open-weight models for everyone
2023 OctBiden Executive Order on AI
2023 NovOpenAI board crisis — Altman fired and reinstated in 5 days
2024 FebGemini 1.5 — 1M token context window (Google)
2024 MarClaude 3 Opus — Anthropic reaches frontier performance
2024 MarEU AI Act formally adopted
2024 JunSSI founded by Ilya Sutskever
2024 SepOpenAI o1 — “reasoning” models (chain-of-thought at inference)
2024 OctNobel Prizes to AI researchers (Hinton — Physics, Hassabis — Chemistry)
2024 DecLLaMA 3.1 405B — largest open model
2025Claude 4, GPT-5 era, AI agents become mainstream, coding agents proliferate

Key Patterns

  1. Winters and Springs — AI has a history of hype → disappointment → actual progress
  2. Scaling — Bigger models + more data + more compute = consistently better results (so far)
  3. The GPU revolution — Modern AI was enabled by repurposing gaming hardware
  4. Open vs Closed — Tension between open-source democratisation and closed safety/business models
  5. Acceleration — The gap between milestones is shrinking (decades → years → months)

Related

enes