Artificial intelligence may feel like a defining technology of the 2020s, but the phrase “artificial intelligence” is far older than ChatGPT, smartphones, or even the modern computer industry.
The term was introduced 71 years ago, in 1955, by American computer scientist John McCarthy in a proposal for a summer research project at Dartmouth College in New Hampshire.
The proposal, dated August 31, 1955, was co-authored by McCarthy, Marvin L. Minsky, Nathaniel Rochester, and Claude E. Shannon. It laid out a remarkably ambitious question for its time: Could machines be built to perform tasks that appeared to require human intelligence?
The researchers proposed studying the idea during a small summer workshop in 1956.
What they imagined was primitive compared with today’s generative AI, but the basic question remains strikingly familiar.
Who coined the term “artificial intelligence”?
John McCarthy is widely credited with coining the term “artificial intelligence” in 1955.
McCarthy was a computer scientist and mathematician who later became one of the most influential figures in the early development of AI research.
His 1955 proposal for the Dartmouth Summer Research Project on Artificial Intelligence used the phrase explicitly and gave the emerging field a name.
The proposal was written with three other researchers: Marvin Minsky, Nathaniel Rochester and Claude Shannon.
Each would go on to become a major figure in computer science.
What did the 1955 Dartmouth proposal actually say?
The document proposed a two-month study involving 10 researchers during the summer of 1956.
Its central claim was extraordinarily ambitious for the era: the researchers argued that aspects of learning and intelligence could, in principle, be described precisely enough for a machine to simulate them.
The proposal suggested that researchers should investigate whether machines could:
- Use human language
- Form abstractions and concepts
- Solve problems typically handled by humans
- Improve their own performance
- Develop forms of reasoning and learning
The researchers were essentially asking whether intelligence could be broken down into sufficiently precise processes that a machine could reproduce them.
That basic idea sits at the heart of modern AI.
The machines were primitive. The ambition was not.
Did AI exist before the term was coined?
Yes.
The field itself did not suddenly appear when McCarthy wrote the words “artificial intelligence.”
Researchers had already been exploring machine reasoning, computation and the possibility of mechanical intelligence for years.
One of the most important figures was British mathematician Alan Turing.
In 1950, Turing published his landmark paper “Computing Machinery and Intelligence,” opening with the question: “Can machines think?”
Rather than getting trapped in a philosophical argument over the exact meaning of “think,” Turing proposed a practical test.
What was Alan Turing’s “Imitation Game”?
Turing imagined a situation in which a human communicated with another human and a machine through written messages without seeing either participant.
The human evaluator would then try to determine which participant was the machine.
If the machine could produce responses convincing enough to be mistaken for a human, Turing argued, that would provide a practical way to discuss machine intelligence.
The experiment became known as the “Imitation Game” and later became widely associated with the Turing Test.
Turing’s approach shifted the discussion away from an abstract question about whether machines genuinely “think” and toward observable behavior.
That was a major conceptual step toward the AI field that emerged in the following decade.
What happened at Dartmouth in 1956?
The 1955 proposal led to the Dartmouth Summer Research Project on Artificial Intelligence in 1956.
John McCarthy organized the workshop, which brought together a small group of researchers in Hanover, New Hampshire.
The meeting is widely regarded as a foundational moment in AI as a formal field of academic research.
The researchers did not have anything resembling modern AI systems.
There were no large language models, neural-network chatbots or image generators.
Instead, they were exploring whether computing machines could learn, reason, use language and perform tasks associated with human intelligence.
The field would spend decades moving toward those goals.
What did the early AI researchers want to build?
The Dartmouth proposal contained a surprisingly broad wish list.
Among the subjects it proposed studying were automatic computers, programming computers to use language, neural networks, self-improvement, abstraction, randomness and creativity.
Some of those ideas sound strikingly modern.
Neural networks, for example, are now central to modern machine learning. Self-improvement and machine reasoning remain major areas of AI research.
The difference is that researchers in the 1950s were operating with radically less computing power and far less data.
Their challenge was not simply to improve AI systems.
They first had to prove that many of the concepts were computationally possible at all.
Why was the Dartmouth project so important?
The Dartmouth workshop helped establish AI as a distinct area of research.
Before that period, scientists were working on related questions across mathematics, logic, cybernetics, neuroscience and computer science.
The Dartmouth proposal helped give those efforts a common identity.
That mattered because naming a field can shape how researchers organize their questions, secure funding and build institutions around them.
The phrase “artificial intelligence” gave scientists a framework for investigating whether machines could reproduce aspects of human cognition.
It also established a research agenda that continues to influence AI more than seven decades later.
How different was 1950s AI from today’s AI?
The difference is enormous.
Early researchers often worked with computers that had tiny amounts of memory by modern standards and operated at speeds that seem almost impossibly slow today.
They relied heavily on explicit rules and symbolic reasoning.
Modern AI, by contrast, often uses neural networks trained on enormous datasets and accelerated by specialized hardware.
Large language models can generate essays, write software, summarize documents and hold conversations. Computer-vision systems can identify objects and analyze images. Generative models can produce photographs, video, music and synthetic voices.
Yet the underlying ambition has a familiar ring.
Researchers are still trying to build machines that can perform tasks associated with human intelligence.
Did the early researchers predict modern AI?
In some respects, remarkably well.
The Dartmouth proposal mentioned language, learning, abstraction, neural networks, self-improvement and creativity.
Those subjects now sit at the center of modern AI research.
But the researchers were far too optimistic about how quickly some of these goals could be achieved.
Many believed substantial progress could come within years.
Instead, AI development moved through periods of excitement and disappointment, including several “AI winters” in which funding and public interest declined after ambitious promises failed to materialize.
The latest AI boom is therefore part of a much longer cycle.
Why did AI take so long to reach its current stage?
The idea of artificial intelligence arrived decades before the technology needed to make it practical.
Three major ingredients eventually changed the field: computing power, large amounts of digital data and advances in machine-learning techniques.
The rise of graphics processing units and other specialized processors made it possible to train increasingly large neural networks.
The explosion of internet-scale datasets provided enormous quantities of material from which models could learn.
At the same time, breakthroughs in machine learning, particularly deep learning and the transformer architecture, dramatically improved the ability of AI systems to process language, images and other forms of information.
The result was a technological acceleration that researchers in the 1950s could scarcely have experienced firsthand.
Why does the 1955 date matter today?
August 31, 1955, is more than a historical footnote.
It marks the moment when a group of researchers formally proposed treating machine intelligence as a research problem with its own name and agenda.
The term itself has survived every major transformation in computing.
It was used during the era of room-sized mainframes, expert systems, personal computers, the internet, smartphones and now generative AI.
Today’s systems may look radically different from what McCarthy and his colleagues imagined.
But the central question has remained remarkably stable:
Can machines perform forms of learning, reasoning and problem-solving that humans associate with intelligence?
More than 70 years later, scientists are still trying to answer it.
The bottom line
The phrase “artificial intelligence” was coined in 1955 by John McCarthy in a proposal for a summer research project at Dartmouth College.
The proposal, co-authored by McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon, envisioned machines capable of using language, forming concepts, solving problems and improving themselves.
The Dartmouth workshop that followed in 1956 is widely considered one of the founding moments of AI research.
Alan Turing had already asked whether machines could think in his 1950 paper “Computing Machinery and Intelligence”, but the Dartmouth project gave the broader research effort a name and a formal identity.
More than seven decades later, the technology has moved far beyond the simple computers of the 1950s.
Yet the question at the center of AI has barely changed.
Humanity is still trying to determine how much of intelligence can be understood, modeled, and ultimately reproduced by a machine.
