# History of Artificial Intelligence

Artificial intelligence (AI) is both an engineering discipline and a scientific field aiming to understand and reproduce intelligent behavior in machines. Its history stretches from ancient myths of artificial servants to the modern era of large language models and general-purpose systems.

This article is the chronological backbone of the AI Pedia. It interlinks with [Hardware and Software](hardware-software.md) and [Philosophy & Society](philosophy.md) throughout.

## 1. Preconditions and Foundations (Antiquity–1940s)

Long before computers, humans imagined artificial minds:

- **Ancient automata and myths.** Greek myths (Hephaestus' golden servants), Jewish legends of the Golem, and medieval clockwork figures all prefigured the idea of a constructed intelligence.
- **Formal reasoning.** Aristotle's syllogisms, Ramón Lull's "thinking machines," and Leibniz's *characteristica universalis* laid groundwork for symbolic computation.
- **Computability theory (1930s).** Alan Turing's machines, Alonzo Church's lambda calculus, and Gödel's incompleteness theorems defined what it means to compute—and, by extension, what a machine might compute.

## 2. Cybernetics and the Birth of the Field (1940s–1950s)

- **Cybernetics.** Norbert Wiener's *Cybernetics* (1948) framed control and communication in animals and machines, uniting feedback, information, and purpose.
- **Turing's question.** In *Computing Machinery and Intelligence* (1950), Alan Turing posed the question "Can machines think?", introduced the **Turing test**, and predicted machines would one day play chess and converse.
- **The Dartmouth Workshop (1956).** John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon coined the term **artificial intelligence**. Their proposal defined the goal: "every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it."

## 3. Symbolic AI and the "Golden Age" (1956–1970s)

Early AI was dominated by symbolic, rule-based approaches:

- **Search and problem solving.** Programs like *Logic Theorist* (Newell & Simon, 1956) and *General Problem Solver* proved that machines could reason over symbols.
- **Game playing.** Arthur Samuel's checkers program (1959) pioneered machine learning through self-play.
- **Lisp and the Lisp machine.** McCarthy's Lisp (1958) became the lingua franca of symbolic AI research.
- **Cognitive modeling.** The *Perceptron* (Rosenblatt, 1958) introduced early neural networks, but Marvin Minsky and Seymour Papert's *Perceptrons* (1969) highlighted its limitations, cooling enthusiasm for neural approaches for years.
- **Robotics and planning.** Shakey the robot (1966–1972) at Stanford demonstrated autonomous planning and perception.

## 4. Expert Systems and the First AI Winter (1970s–1980s)

- **Expert systems.** Rule-based systems like **MYCIN** (1975) and **XCON** encoded human expertise in narrow domains.
- **Knowledge representation.** Formalisms such as frames, semantic networks, and logic programming (Prolog, 1970s) aimed to encode world knowledge.
- **Funding collapse (AI winter).** Overpromising and underdelivery, combined with the LISP machine market crash (1987), led to reduced funding and interest. This is the first **AI winter**.
- **Fifth-generation project.** Japan's 1982 Fifth Generation Computer Systems project spurred Western response (e.g., the UK Alvey program) but ultimately failed to deliver broad logic-based intelligence.

## 5. Statistical Learning and the Quiet Resurgence (1980s–2000s)

- **Neural networks revived.** The **backpropagation** algorithm (Rumelhart, Hinton, Williams, 1986) enabled multi-layer training. Hopfield networks and Boltzmann machines advanced theory.
- **Machine learning as engineering.** Decision trees, support vector machines, Bayesian networks, and ensemble methods became standard tools.
- **Speech and vision.** Hidden Markov models dominated speech recognition; computer vision advanced with feature-based methods.
- **The Internet changes data.** Search engines, recommendation systems, and spam filters made machine learning economically central.
- **IBM Watson (2011)** beat human champions on *Jeopardy!*, demonstrating applied question-answering.

## 6. The Deep-Learning Revolution (2006–2020)

- **Deep learning breakthrough.** Geoffrey Hinton's 2006 work on deep belief networks, followed by AlexNet's ImageNet win (2012), showed that deep neural networks could outperform hand-crafted features.
- **Hardware acceleration.** GPUs (initially for graphics) proved ideal for matrix-heavy neural networks, leading to the rise of CUDA-based frameworks. See [Hardware and Software](hardware-software.md).
- **Frameworks.** Theano, Caffe, TensorFlow (2015), and PyTorch (2016) democratized model building.
- **Domain successes.** Convolutional networks transformed image recognition; recurrent networks and then **Transformers** (Vaswani et al., 2017) transformed sequence modeling and NLP.
- **AlphaGo (2016)** defeated Lee Sedol, showcasing deep reinforcement learning on the game of Go.

## 7. The Generative and Agentic Era (2020–present)

- **Large language models (LLMs).** Transformer-based models such as GPT-2, GPT-3, GPT-4, and open models like Llama demonstrated few-shot learning, code generation, and general text understanding.
- **Multimodal models.** Systems now process text, images, audio, and video jointly.
- **Diffusion models.** Text-to-image systems (DALL·E, Stable Diffusion) and text-to-video systems reshaped creative fields.
- **Agents and tools.** Models increasingly operate as agents—planning, using tools, browsing, and taking actions. See [Recent Topics](recent.md).
- **Debates on AGI and alignment.** Claims about general intelligence and safety have become central to public discourse. See [Philosophy & Society](philosophy.md).

## 8. Recurring Themes and Lessons

- **Cycles of hype and winter.** AI has repeatedly overpromised, then corrected. Understanding these cycles is essential for sober assessment.
- **The battle of paradigms.** Symbolic versus statistical (neural) approaches have alternated in dominance; modern systems often hybridize them.
- **Scale and data.** Progress has increasingly come from data, compute, and scale rather than new algorithmic ideas alone.
- **Intelligence is embodied and social.** Cognition is shaped by hardware, environment, and interaction—a theme explored in [Philosophy & Society](philosophy.md).

## 9. Sources and Further Reading

- Turing, A. M. (1950). *Computing Machinery and Intelligence.* Mind 59(236):433–460.
- McCarthy, J. et al. (1955). *A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence.*
- Newell, A. & Simon, H. (1972). *Human Problem Solving.*
- Rumelhart, D., Hinton, G., & Williams, R. (1986). *Learning representations by back-propagating errors.* Nature 323:533–536.
- Vaswani, A. et al. (2017). *Attention Is All You Need.* NeurIPS.
- Russell, S. & Norvig, P. (2021). *Artificial Intelligence: A Modern Approach*, 4th ed.

## Sub-Articles

- [Preconditions and Foundations](sub/preconditions.md)
- [Cybernetics and the Birth of AI](sub/cybernetics.md)
- [Symbolic AI and the Golden Age](sub/symbolic-ai.md)
- [Expert Systems and the First AI Winter](sub/expert-systems.md)
- [AI Winters](sub/ai-winters.md)
- [Statistical Learning Resurgence](sub/statistical-learning.md)
- [The Deep-Learning Revolution](sub/deep-learning-revolution.md)
- [The Generative and Agentic Era](sub/generative-agentic-era.md)
- [Transformers](sub/transformers.md)

See the [full sub-article index](sub/index.md).

*Continue to [Hardware and Software](hardware-software.md), [Philosophy & Society](philosophy.md), or [Recent Topics](recent.md).*