History of Artificial Intelligence: From Alan Turing to Modern AI (Complete Timeline Explained)

The history of artificial intelligence is a seven-decade story of mathematical breakthroughs, bold scientific ambition, boom-and-bust funding cycles, and relentless engineering. While today's conversational chatbots, image generators, and autonomous software agents feel brand new, many modern AI technologies build on theoretical and engineering advances made across the twentieth and early twenty-first centuries.

Understanding how artificial intelligence evolved—from Alan Turing's early papers on computation and the 1956 Dartmouth workshop, through two painful "AI Winters," to the deep learning and Generative AI revolutions—explains both why today's models are so capable and why certain limitations still exist. If you are new to the field, you can also explore our foundational guides on What Is Artificial Intelligence? and How Artificial Intelligence Works.

In this complete historical guide, we follow the major eras of AI chronologically—highlighting the pioneers, landmark algorithms, hardware revolutions, and paradigm shifts that brought artificial intelligence from academic theory into everyday life.

Complete History of Artificial Intelligence Timeline (1936–2026)

Before diving into each decade in detail, use this master timeline to see how the key milestones in artificial intelligence connect across history:

1. Pre-Computer Roots and Alan Turing's Theoretical Breakthroughs (1600s–1950)

Long before silicon microchips existed, philosophers and mathematicians asked whether human reasoning could be reduced to formal calculation. In the 17th century, thinkers like Gottfried Wilhelm Leibniz envisioned a universal mathematical calculus capable of representing logical arguments mechanically. In the 19th century, George Boole developed an algebra of logic that later became foundational for digital circuit design—while Charles Babbage and Ada Lovelace conceptualized the Analytical Engine.

Alan Turing and the 1950 Turing Test

The transition from philosophical theory to modern computer science was led by British mathematician Alan Turing. In his landmark 1936 paper, Turing described an abstract computing machine and helped establish what it means for a function to be computable by an algorithm. During World War II at Bletchley Park, Turing contributed to British codebreaking efforts against the German Enigma cipher, including work on the electromechanical bombe.

In 1950, Turing published his seminal paper "Computing Machinery and Intelligence," opening with the famous line: "I propose to consider the question, 'Can machines think?'" Because "thinking" is hard to define philosophically, Turing proposed a practical behavioral benchmark he called the Imitation Game—now known worldwide as the Turing Test. If a human judge conversing via text with both a person and a computer cannot reliably tell which is which, the machine has performed convincingly in that particular conversational test; passing it would not establish consciousness or general intelligence. This insight laid the groundwork for how scientists classify the types of artificial intelligence and debate future Artificial General Intelligence (AGI).

2. The Birth of AI at Dartmouth and the Symbolic Era (1956–1973)

The phrase artificial intelligence appeared in a 1955 research proposal written by John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon. Their proposal led to the 1956 Dartmouth Summer Research Project on Artificial Intelligence in New Hampshire. The event is often regarded as the symbolic birth of AI as an organized field. The researchers proposed exploring whether aspects of learning and intelligence could be described precisely enough for machines to simulate them. It was an ambitious research hypothesis, not an established fact.

Early Triumphs: Symbolic AI and the Perceptron

The 1950s and 1960s became a golden era of early experimentation split across two competing schools of thought:

• Symbolic AI ("Good Old-Fashioned AI" or GOFAI): Researchers like Allen Newell and Herbert Simon built programs that manipulated logical symbols according to explicit rules. Their Logic Theorist (mid-1950s), developed with Cliff Shaw, demonstrated automated theorem proving; later work included the General Problem Solver. At MIT in the 1960s, Joseph Weizenbaum developed ELIZA, an influential early chatbot that used pattern matching to imitate aspects of conversation. It did not understand language as modern users might assume.
• Early Connectionism (Neural Networks): Inspired by Warren McCulloch and Walter Pitts' 1943 mathematical neuron model, Frank Rosenblatt introduced the perceptron in 1957, an early trainable neural-network model; subsequent hardware demonstrations showed simple pattern-recognition capabilities, foreshadowing modern artificial neural networks.

3. The Two "AI Winters" and the Rise of Expert Systems (1974–1993)

Early researchers made ambitious predictions, but tasks that seemed simple to humans proved unexpectedly hard to automate. Systems that handled formal puzzles struggled with ambiguous language, changing environments, and everyday common sense. A related observation, articulated later as Moravec's paradox, is that some abstract reasoning tasks can be easier to automate than ordinary perception and movement.

The First AI Winter (1974–1980)

Many symbolic systems performed well in simplified environments but struggled with real-world complexity and combinatorial explosion. Marvin Minsky and Seymour Papert's 1969 book Perceptrons also drew attention to important limitations of single-layer perceptrons, including their inability to represent XOR without additional structure. Critical assessments—including the UK's 1973 Lighthill Report—and changing US funding priorities contributed to a slowdown commonly called the first AI winter. The dates and severity varied across research communities.

The 1980s Boom: Expert Systems

AI rebounded in the early 1980s by narrowing its focus. Instead of trying to build a general-purpose human mind, engineers built Expert Systems—programs that interviewed human specialists (such as doctors, geologists, or tax accountants) and translated their domain knowledge into thousands of hardcoded if-then rules paired with an inference engine.

• MYCIN (Stanford): A research expert system that demonstrated rule-based recommendations for certain bacterial infections and antibiotic treatment. It was not adopted as a routine clinical service.
• XCON (DEC): A commercial expert system used to configure computer orders at Digital Equipment Corporation, showing that narrower rule-based AI could create operational value.

The Second AI Winter (1987–1993)

By the late 1980s, many expert systems had proved expensive to update and difficult to extend beyond their original domains. Specialized Lisp machines also lost their commercial advantage as general-purpose computers improved. Investment weakened again in what is commonly called the second AI winter, although important AI research continued.

4. The Machine Learning Shift, Deep Blue, and Big Data (1990s–2010)

Quietly during the 1980s and 1990s, statistical and data-driven approaches became increasingly influential alongside symbolic approaches. In many applications, researchers began placing greater emphasis on a different question: What if computers learn statistical patterns directly from data? This approach helped expand the practical use of machine learning (see our full comparison of Machine Learning vs Artificial Intelligence).

Several breakthroughs converged during this era:

1. Backpropagation (1986): David Rumelhart, Geoffrey Hinton, and Ronald Williams helped popularize backpropagation for training multilayer neural networks. The algorithm had earlier intellectual precursors and was one part of a much longer research effort.
2. Statistical Machine Learning: Algorithms across supervised, unsupervised, and reinforcement learning began powering real-world spam filters, optical character recognition (OCR) for reading postal zip codes (led by Yann LeCun's Convolutional Neural Networks in 1989), and early credit-card fraud detection.
3. IBM Deep Blue Defeats Garry Kasparov (1997): In a landmark moment watched around the globe, IBM's Deep Blue supercomputer defeated reigning world chess grandmaster Garry Kasparov in a six-game match—evaluating 200 million board positions per second through specialized search hardware and evaluation functions.
4. The Internet and Big Data Explosion (2000s): The rise of the World Wide Web, search engines, social media, and smartphones generated billions of labeled photographs, web pages, and digital records—providing larger datasets that could support training and evaluation of many machine-learning systems.

5. GPUs, Deep Learning, and the 2012 AlexNet Turning Point (2010–2016)

Around the early 2010s, two important enablers had become more practical for deep neural networks: massive benchmark datasets (such as Fei-Fei Li's ImageNet dataset, which grew to millions of labeled images) and Graphics Processing Units (GPUs). Originally engineered to render 3D video game graphics, GPUs excelled at running millions of parallel matrix calculations simultaneously.

In 2012, a deep convolutional neural network called AlexNet—built by Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton and trained on two NVIDIA GPUs—achieved a major improvement at the ImageNet visual recognition challenge: the winning system reached approximately 15.3% top-five error, compared with about 26.2% for the runner-up. The result became a highly visible catalyst for the modern deep learning boom across computer vision, speech recognition, and medical imaging.

Four years later, in 2016, Google DeepMind's AlphaGo combined deep neural networks with reinforcement learning to defeat 18-time world champion Lee Sedol at the ancient board game Go—a game long considered a difficult benchmark for computer programs.

6. The Transformer Architecture, LLMs, and Agentic AI (2017–2026)

While deep learning conquered images and board games in the early 2010s, Natural Language Processing (NLP) still struggled with long documents because older Recurrent Neural Networks (RNNs) read sentences slowly, one word after another, facing challenges with some long-range dependencies and efficient parallel training.

2017: "Attention Is All You Need"

In June 2017, eight researchers at Google published a landmark paper titled "Attention Is All You Need," introducing the Transformer architecture. The Transformer used attention mechanisms to model relationships among tokens without the recurrence central to many earlier sequence models. Its design made training more parallelizable; however, many generative language models still produce output token by token at inference time.

2018–2022: Foundation Models and the Launch of ChatGPT

As Transformer-based approaches became more scalable with advances in training methods, hardware, and data, researchers began training giant foundation models and Large Language Models (LLMs)—starting with Google's BERT and OpenAI's GPT-1, GPT-2, and GPT-3. Simultaneously, diffusion models revolutionized AI image generation.

When OpenAI paired its language models with Reinforcement Learning from Human Feedback (RLHF) and launched ChatGPT in late 2022 (see How ChatGPT Works), conversational AI reached a much broader public audience. Claims about being the single fastest-adopted application depend on how adoption is measured.

2023–2026: Multimodal Reasoning, AI Search, and Autonomous Agents

Between 2023 and 2026, the generative AI wave matured into four major technological shifts that define the modern landscape (explored further in The Future of Artificial Intelligence):

• Native Multimodal AI: Flagship systems from OpenAI, Google Gemini, Anthropic Claude, and open-weight labs now process text, high-resolution images, live voice audio, video, and code inside a single unified model.
• Extended AI Reasoning & Grounding: Some systems combine additional inference-time computation with Retrieval-Augmented Generation (RAG) and AI Search to support research, analysis, and other complex tasks, while still requiring verification.
• Agentic AI and Computer-Using Agents: some AI systems can plan or execute multiple steps using external tools and protocols such as MCP, and can interact with browsers or applications under configurable permissions and supervision.
• Efficient On-Device & Physical AI: Alongside giant cloud models, Small Language Models (SLMs) can bring some processing Edge AI onto laptops and phones, while Physical AI explores how machine-learning models can support robots and autonomous systems.

How the 4 Paradigms of AI Compare Across History

Frequently Asked Questions About the History of AI

1. Who is considered the father of artificial intelligence?

There is no universally agreed single "father of AI." Historians credit several founding pioneers: Alan Turing laid the theoretical foundations of computation and the Turing Test (1936–1950), while John McCarthy helped introduce the term in the 1955 Dartmouth proposal with Marvin Minsky, Nathaniel Rochester, and Claude Shannon; Allen Newell and Herbert Simon were influential early AI pioneers. In modern deep learning, Geoffrey Hinton, Yann LeCun, and Yoshua Bengio are widely recognized as the "Godfathers of Deep Learning."

2. When and where was the term "Artificial Intelligence" coined?

Computer scientist John McCarthy coined the term "Artificial Intelligence" in his 1955 proposal for the 1956 Dartmouth Summer Research Project on Artificial Intelligence held at Dartmouth College in Hanover, New Hampshire.

3. What was the very first artificial intelligence program?

There is no universally accepted single first AI program because the field did not yet have a settled definition. Early game-playing programs appeared in the 1950s, while Logic Theorist, developed by Allen Newell, Herbert Simon, and Cliff Shaw in the mid-1950s, is often cited as one of the earliest landmark AI programs because it could prove mathematical theorems.

4. What were the "AI Winters" and why did they happen?

The AI Winters (roughly 1974–1980 and 1987–1993) were periods when government and corporate funding for AI collapsed. They occurred because early rule-based systems could not handle real-world ambiguity, lacked common-sense knowledge, could not learn automatically from data, and hit severe computing hardware limits after overpromising fast results.

5. How did early Expert Systems differ from today's Machine Learning?

Classic expert systems relied primarily on explicitly encoded knowledge and rules, while machine-learning models estimate patterns from data during training. Both approaches have limitations, and modern systems sometimes combine rules with learned models.

6. Why was the 2017 Transformer paper so important for modern AI?

The 2017 paper "Attention Is All You Need" introduced the Transformer architecture, which showed how attention-based sequence modeling could improve training parallelism. Later scaling advances helped turn Transformers into a foundation for many modern language models, making modern Large Language Models like ChatGPT possible.

7. Why did artificial intelligence suddenly advance so fast in the 2020s?

Rapid progress in the 2020s reflected several interacting long-term developments: massive digital datasets from the internet, the parallel computing power of modern GPUs and AI chips, and scalable deep-learning architectures like the Transformer and diffusion models.

8. What is the next major chapter in the history of AI?

Following the rise of conversational chatbots, AI is moving into multimodal reasoning, agentic workflows, scientific discovery (such as protein design and materials science), Physical AI in robotics, and formal AI governance and safety standards.

• Alan Turing, “Computing Machinery and Intelligence” (1950) — the original paper introducing the imitation-game discussion.
• McCarthy, Minsky, Rochester, and Shannon, Dartmouth research proposal (1955) — the proposal that named the field before the 1956 workshop.
• IBM, History of Deep Blue — the 1997 match with Garry Kasparov and the system's specialized search approach.
• Krizhevsky, Sutskever, and Hinton, ImageNet classification (2012) — the AlexNet research paper and competition results.
• Vaswani and colleagues, “Attention Is All You Need” (2017) — the original Transformer architecture paper.

These sources provide a starting point for checking key dates and technical milestones. Historical accounts may differ in how they define a field's beginning or the boundaries of each AI winter.

Primary Sources and Historical References

Conclusion

Looking back across the history of artificial intelligence reveals one unmistakable lesson: overnight tech revolutions are almost always 70 years in the making. Every time you prompt an AI assistant, translate a document, or use smart medical software today, you are interacting with a chain of human ingenuity stretching from Alan Turing's 1936 paper and the 1956 Dartmouth campus to modern GPU clusters and Transformer neural networks. Understanding that journey helps us look past short-term hype—and use modern AI with clear eyes, realistic expectations, and responsible human judgment.