The History of Artificial Intelligence
From Alan Turing's thought experiment to billion-parameter language models — an interactive journey through 76 years of AI breakthroughs.
The Foundations (1950–1979)
Alan Turing publishes "Computing Machinery and Intelligence"
Proposes the Imitation Game (Turing Test) and asks "Can machines think?" — launching the philosophical foundation of AI.
MilestoneDartmouth Conference coins "Artificial Intelligence"
John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon organize the workshop that names the field and sets its ambitious goals.
MilestoneFrank Rosenblatt invents the Perceptron
The first neural network model capable of learning — a single-layer network that could classify simple patterns. The New York Times reported it could learn to "walk, talk, see, write, reproduce itself, and be conscious."
ResearchELIZA chatbot created at MIT
Joseph Weizenbaum builds ELIZA, a pattern-matching chatbot that simulates a psychotherapist. Users formed emotional bonds with it, surprising even its creator.
ProductMinsky & Papert publish "Perceptrons"
Proves single-layer perceptrons can't solve XOR and other non-linear problems, effectively killing neural network funding for over a decade.
SetbackAI Winters & Expert Systems (1980–1999)
Expert systems boom begins
Rule-based AI systems like XCON (Digital Equipment Corp) save companies millions. The AI industry grows to $1 billion. Japan launches the Fifth Generation Computer project.
ProductBackpropagation popularized by Rumelhart, Hinton & Williams
Demonstrates efficient training of multi-layer neural networks, reviving connectionism. This algorithm remains the foundation of all modern deep learning.
ResearchSecond AI Winter begins
Expert systems fail to scale. The Lisp machine market collapses. DARPA cuts AI funding. Japan's Fifth Generation project is abandoned. AI researchers avoid using the term "AI."
SetbackIBM Deep Blue defeats world chess champion Garry Kasparov
Deep Blue wins a six-game match, marking the first time a computer defeats a reigning world champion under tournament conditions. Uses brute-force search, not learning.
MilestoneLSTM networks invented
Hochreiter and Schmidhuber introduce Long Short-Term Memory networks, solving the vanishing gradient problem and enabling learning of long-range sequence dependencies.
ResearchYann LeCun's LeNet-5 for handwriting recognition
CNNs achieve practical success recognizing handwritten checks for banks. Proves deep learning can solve real-world problems at scale.
ProductThe Machine Learning Era (2000–2011)
Geoffrey Hinton's deep belief networks
Demonstrates that deep networks can be effectively pre-trained layer by layer, reigniting interest in deep learning after years of skepticism.
ResearchImageNet dataset created
Fei-Fei Li leads creation of ImageNet — 14 million labeled images across 20,000 categories. The associated competition (ILSVRC) will drive the deep learning revolution.
ResearchIBM Watson wins Jeopardy!
Watson defeats champions Ken Jennings and Brad Rutter, demonstrating natural language understanding and knowledge retrieval at superhuman speed.
MilestoneApple launches Siri
The first mainstream AI voice assistant brings AI into millions of pockets. Google Now (2012) and Amazon Alexa (2014) follow.
ProductThe Deep Learning Revolution (2012–2017)
AlexNet wins ImageNet by a massive margin
A deep CNN trained on GPUs reduces image classification error by 10.8%, shattering the previous best. This moment marks the beginning of the deep learning era.
MilestoneResearchWord2Vec creates word embeddings
Google researchers show words can be represented as vectors where king − man + woman ≈ queen. Embeddings become the foundation of modern NLP.
ResearchGANs, Seq2Seq, and Attention invented
A landmark year: Goodfellow introduces GANs, Sutskever proposes sequence-to-sequence learning, and Bahdanau introduces the attention mechanism. Three papers that reshape AI.
ResearchDeepMind acquired by Google for $500M
Google acquires DeepMind, signaling Big Tech's serious investment in AI research. The acquisition validates AI as a strategic priority.
MilestoneResNet enables very deep networks
Residual connections allow training 152-layer networks, surpassing human accuracy on ImageNet. Skip connections become a standard component in all deep architectures.
ResearchOpenAI and TensorFlow launched
Elon Musk, Sam Altman, and others found OpenAI as a non-profit AI safety lab. Google open-sources TensorFlow, democratizing deep learning tools.
MilestoneAlphaGo defeats world Go champion Lee Sedol
DeepMind's AlphaGo wins 4-1 against the world's top Go player. Go was considered decades away from being solved. Move 37 in game 2 was a creative play no human had considered.
Milestone"Attention Is All You Need" — The Transformer is born
Google researchers introduce the Transformer architecture, replacing recurrence with self-attention. This single paper becomes the foundation for GPT, BERT, Claude, Gemini, and virtually all modern AI.
MilestoneResearchThe LLM Era (2018–2026)
BERT and GPT-1 launch the pre-training revolution
Google's BERT (bidirectional) and OpenAI's GPT-1 (autoregressive) demonstrate that large-scale pre-training on text creates powerful, versatile language models. NLP will never be the same.
ResearchGPT-2: "Too dangerous to release"
OpenAI initially withholds GPT-2 (1.5B parameters) citing misuse concerns about its text generation quality. This sparks debate about responsible AI release practices.
SafetyResearchGPT-3 demonstrates in-context learning
At 175B parameters, GPT-3 can perform tasks from just a few examples in the prompt — no fine-tuning needed. Launches the prompt engineering era and the API economy for AI.
MilestoneAlphaFold solves protein folding
DeepMind's AlphaFold2 achieves atomic-level accuracy in predicting protein structures, solving a 50-year grand challenge in biology. Later wins the 2024 Nobel Prize in Chemistry.
MilestoneResearchDALL-E and CLIP connect language to vision
OpenAI's DALL-E generates images from text descriptions. CLIP learns to match images with text. Together they launch the multimodal AI revolution.
ProductAnthropic founded with focus on AI safety
Dario and Daniela Amodei leave OpenAI to found Anthropic, prioritizing AI safety research. They develop Constitutional AI — training AI systems using principles rather than just human feedback.
MilestoneStable Diffusion democratizes AI image generation
Stability AI releases Stable Diffusion as open source. For the first time, anyone with a GPU can generate photorealistic images from text. Sparks massive creative AI adoption and debate about AI art.
ProductChatGPT launches and reaches 100M users in 2 months
OpenAI releases ChatGPT (based on GPT-3.5 with RLHF). It becomes the fastest-growing consumer application in history, bringing AI to the mainstream. Every industry begins exploring LLM integration.
MilestoneProductGPT-4, Claude, and the multimodal frontier
GPT-4 passes the bar exam and handles images. Anthropic launches Claude with strong safety properties. Google releases Gemini. Meta open-sources Llama 2. The AI race intensifies across every major tech company.
ProductMilestoneOpen-source AI explodes
Meta's Llama 2, Mistral 7B, and dozens of community models prove that competitive AI can be open-source. Fine-tuning techniques like LoRA and QLoRA make customization accessible to individuals.
ResearchAI agents and autonomous coding emerge
Claude and GPT-4 gain tool use and agentic capabilities. GitHub Copilot, Cursor, and Claude Code transform software development. AI systems begin executing multi-step tasks autonomously.
ProductEU AI Act becomes law
The world's first comprehensive AI regulation classifies AI systems by risk level and sets requirements for transparency, safety, and human oversight. Other nations develop their own frameworks.
SafetyMilestoneVideo generation (Sora) and reasoning models (o1)
OpenAI demonstrates Sora for photorealistic video generation from text. Reasoning models like o1 use chain-of-thought to solve complex problems. Google's Gemini 2 pushes multimodal boundaries.
ProductResearchAI becomes infrastructure
AI integration becomes expected across all software. Claude, GPT, and Gemini handle million-token contexts. AI agents manage complex workflows. The question shifts from "should we use AI?" to "how do we use AI responsibly?"
MilestoneThe present: AI as everyday tool
AI assistants write code, analyze data, create content, and manage tasks. Multi-agent systems collaborate on complex projects. The focus increasingly shifts to alignment, safety, and ensuring AI benefits everyone.
MilestoneSafety