1950 to 2026
AI Timeline
Seventy-five years of AI as a night sky. Click a star to read its story, or click anywhere for a shooting star.
Best on desktop. On smaller screens the sky stays still: browse the milestone list below.
Seventy-five years of AI as a night sky. Click a star to read its story, or click anywhere for a shooting star.
Best on desktop. On smaller screens the sky stays still: browse the milestone list below.
Twenty-nine moments from Turing to agents, the same stories drawn in the sky. Each links out to a public source.
Alan Turing’s paper in the journal Mind asks “Can machines think?” and proposes the imitation game (now called the Turing test), judging a machine by whether its conversation can pass for a person’s.
John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon gather researchers at Dartmouth College for a summer project on “artificial intelligence”. It is the event usually counted as the founding of the field.
Frank Rosenblatt introduces the perceptron, a simple artificial neuron that adjusts its own weights from examples. The Mark I Perceptron learned to recognise basic shapes, an ancestor of today’s neural networks.
Joseph Weizenbaum’s program at MIT mimics a psychotherapist by reflecting users’ words back as questions. Many people confided in it, an early sign of how readily we read minds into machines.
IBM’s Deep Blue defeats world chess champion Garry Kasparov 3½–2½ in a six-game rematch in New York. It was the first time a computer beat a reigning champion in a match under standard conditions.
Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton win the ImageNet challenge by a wide margin with a deep neural network trained on GPUs. The result kick-starts the deep learning era.
Ian Goodfellow and colleagues pit two networks against each other: one generates images, the other tries to spot fakes. GANs become the leading way to make synthetic images for years.
Google engineer Alexander Mordvintsev runs an image classifier in reverse, amplifying the patterns it detects. The hallucinatory results bring neural-network imagery to a wide audience.
DeepMind’s AlphaGo beats Go champion Lee Sedol 4–1 in Seoul. Its move 37 in game two, which experts first took for a mistake, becomes a symbol of machine creativity.
Google researchers publish “Attention Is All You Need”, introducing the transformer architecture. It becomes the foundation of almost every large language model that follows.
OpenAI releases GPT-3, a 175-billion-parameter language model that can write, summarise and code from a few examples in a prompt, showing how far scale alone could go.
OpenAI shows DALL·E generating images from text, alongside CLIP, a model that links images and language. Together they open the door to prompt-driven image making.
OpenAI’s second DALL·E uses diffusion to produce far more realistic, higher-resolution images from a sentence, and can edit existing pictures.
David Holz’s independent lab opens Midjourney to the public through Discord. Its painterly style quickly makes it a favourite among artists and designers.
Stability AI, with researchers from CompVis (LMU Munich) and Runway, releases an image model with open weights that runs on consumer GPUs, sparking a wave of community tools.
OpenAI releases ChatGPT as a free research preview. It becomes one of the fastest-growing consumer apps ever and brings conversational AI to the mainstream.
On the same day, OpenAI announces GPT-4 and Anthropic opens access to Claude, a moment that marks the start of a new race between frontier assistants.
Runway opens Gen-2 to the public, letting anyone generate short video clips from a text prompt or image. It was one of the first widely available text-to-video tools.
OpenAI previews Sora, a video model that generates clips up to a minute long with consistent characters and camera motion, resetting expectations for AI film.
Kuaishou’s Kling and Luma AI’s Dream Machine arrive within weeks of each other, putting realistic AI video directly into creators’ hands.
Google DeepMind announces Veo 2, with improved physics, cinematic camera control and output up to 4K, raising the bar for realism in generated video.
Chinese lab DeepSeek releases R1, an open-weight reasoning model that rivals leading closed systems at a fraction of the reported training cost, rattling markets and opening reasoning to everyone.
Google’s Veo 3 generates video with native sound: dialogue, effects and ambience in a single pass. AI clips start to feel like scenes rather than silent shots.
OpenAI releases GPT-5, folding quick answers and deeper reasoning into one system and making it the default model in ChatGPT.
Google DeepMind’s Genie 3 generates explorable worlds in real time from a prompt. In 2026 Google opens it to subscribers as Project Genie.
OpenAI launches Sora 2, with synchronised audio, alongside a TikTok-style social app built entirely around AI-generated clips.
ByteDance’s Seedance 2.0 goes viral for startlingly realistic clips, and draws cease-and-desist letters from Disney and Paramount. Kuaishou ships Kling 3.0 the same month.
Six months after launch, OpenAI announces it is shutting the Sora app; reports link the move to compute costs and a shift toward enterprise tools. A reality check for AI video.
AI agents move from demos to infrastructure, running multi-step coding, research and browsing tasks in the background inside everyday tools and workflows.