AI History
AI has grown in cycles of optimism, funding cutbacks, and renewed progress, moving from hand-coded logic in the 1950s to today's data-driven deep learning systems.
Before the Name: Turing's Question
In 1950, mathematician Alan Turing published a paper asking whether machines can think. Rather than trying to define 'thinking' directly, he proposed a practical test now known as the Turing test: if a human judge, chatting by text with both a person and a machine, cannot reliably tell which is which, the machine could be said to be exhibiting intelligent behavior. This gave researchers a concrete way to talk about machine intelligence years before the field had a name.
1956: The Term Is Coined at Dartmouth
The phrase 'artificial intelligence' was coined by John McCarthy, who organized a summer workshop at Dartmouth College in 1956, along with Marvin Minsky, Nathaniel Rochester, and Claude Shannon. Attendees included Allen Newell and Herbert Simon. The workshop proposal made a bold claim: that every aspect of learning or intelligence could, in principle, be so precisely described that a machine could be made to simulate it. That claim marks the formal starting point of AI as a named research field.
- Logic Theorist (1956), by Newell, Simon, and Shaw, proved mathematical theorems and is often called one of the first AI programs
- The General Problem Solver extended the idea of searching step by step toward a goal
- ELIZA (1966), by Joseph Weizenbaum, used simple pattern matching to imitate a psychotherapist in conversation
- Early researchers were openly optimistic, with some predicting human-level machine intelligence within a couple of decades
Expert Systems: AI's First Commercial Wave
Through the 1970s and 1980s, the dominant approach was the expert system: a program that captured a human specialist's knowledge as a large set of if-then rules. MYCIN suggested diagnoses and antibiotics for bacterial infections, and XCON (also called R1) helped Digital Equipment Corporation configure computer orders, reportedly saving the company significant money. Their commercial success fueled a boom in AI companies and specialized 'Lisp machine' hardware built to run these rule-based systems efficiently.
The AI Winters
From Hand-Written Rules to Deep Learning
Through the 1990s and 2000s, attention shifted away from hand-coding knowledge as rules and toward statistical methods that learn patterns from data instead, an approach covered algorithm by algorithm in this site's Machine Learning tutorial. Growth of digital data and cheaper computing made this shift practical at a much larger scale than before.
Neural networks had existed since the 1950s but were long considered impractical for hard, real-world problems. That changed in 2012, when a deep neural network called AlexNet dramatically outperformed competing approaches in the ImageNet image recognition competition. The result, made possible by more data, more computing power (particularly GPUs), and refined training techniques, is widely credited with reigniting mainstream interest in neural networks. Progress accelerated in the years that followed, including AlphaGo's 2016 victory over top Go player Lee Sedol and rapid advances in speech recognition, translation, and large-scale generative language models.
