AI Timeline
Significant developments in artificial intelligence, ordered chronologically. Each entry links to its primary source.
Turing proposes the imitation game
Alan Turing publishes 'Computing Machinery and Intelligence', proposing a test for machine intelligence that frames the field for decades.
SourceDartmouth workshop coins 'artificial intelligence'
John McCarthy organises the Dartmouth Summer Research Project, which names the field and establishes it as a discipline.
SourceDeep Blue defeats Kasparov
IBM's Deep Blue beats world chess champion Garry Kasparov in a six-game match, the first time a computer defeats a reigning champion under standard tournament conditions.
SourceAlexNet wins ImageNet
Krizhevsky, Sutskever and Hinton's convolutional neural network wins the ImageNet Large Scale Visual Recognition Challenge by a wide margin, catalysing the deep learning era.
SourceGenerative adversarial networks published
Goodfellow et al. introduce GANs, enabling neural networks to generate realistic images by training two networks against each other.
SourceAlphaGo defeats Lee Sedol
DeepMind's AlphaGo defeats world Go champion Lee Sedol 4-1, a game previously considered too complex for brute-force search.
SourceTransformer architecture published
Vaswani et al. publish 'Attention Is All You Need', introducing the transformer architecture that becomes the foundation for GPT, BERT, and all major language models.
SourceGPT-1 released
OpenAI releases GPT-1, demonstrating that generative pretraining on a large text corpus followed by discriminative fine-tuning produces strong results across many NLP tasks.
SourceBERT released
Google releases BERT, a bidirectional transformer that sets new benchmarks on 11 NLP tasks and becomes the dominant approach for understanding tasks.
SourceGPT-3 released
OpenAI releases GPT-3 with 175 billion parameters, demonstrating few-shot learning and sparking widespread interest in large language models.
SourceDALL-E announced
OpenAI announces DALL-E, a model that generates images from text descriptions, demonstrating multimodal capability.
SourceAlphaFold 2 protein structure database released
DeepMind and EMBL release predicted structures for nearly all known proteins, an application of AI to a fundamental problem in biology.
SourceEU AI Act proposal advances
The European Parliament committees adopt their positions on the AI Act, establishing the risk-based regulatory framework that becomes law in 2024.
SourceChatGPT launched
OpenAI launches ChatGPT, reaching an estimated 100 million users within two months and bringing large language models to mainstream attention.
SourceMicrosoft integrates GPT-4 into Bing
Microsoft launches the new Bing with GPT-4 integration, the first major search engine to incorporate a large language model into core search.
SourceGPT-4 released
OpenAI releases GPT-4, a multimodal model accepting text and image input, with substantially improved reasoning over GPT-3.5.
SourceLlama 2 released as open weights
Meta releases Llama 2 with open weights, establishing the pattern of large open-weight model releases that DeepSeek, Mistral and others follow.
SourceUS Executive Order on AI Safety
President Biden signs Executive Order 14110, requiring safety testing and reporting for AI systems above a compute threshold.
SourceGoogle releases Gemini 1.5 with 1M token context
Google DeepMind releases Gemini 1.5 Pro with a 1 million token context window, an order of magnitude larger than competitors at the time.
SourceEU AI Act enters into force
The EU Artificial Intelligence Act becomes law, with a phased implementation timeline. Prohibited practices apply from February 2025.
SourceDeepSeek-V3 demonstrates cost-efficient training
DeepSeek releases V3, a 671B MoE model trained at a fraction of the cost of comparable Western models, challenging assumptions about the compute required for competitive performance.
SourceDeepSeek-R1 released with open weights
DeepSeek releases R1, an open-weight reasoning model competitive with OpenAI o1, demonstrating that reasoning capability is not exclusive to closed models.
SourceLlama 4 released
Meta releases Llama 4 Maverick and Scout, the first Llama models with mixture-of-experts architecture. Scout offers a 10M token context window.
SourceOpenAI releases o3 and o4-mini
OpenAI releases o3 and o4-mini reasoning models with tool use and agent capabilities, the first reasoning models with native function calling.
SourceAnthropic releases Claude 4
Anthropic releases Claude Opus 4 and Sonnet 4 with extended thinking and improved agentic capabilities.
Source