Why was AI created? The real motives behind it (2026)

AI was created because a handful of mathematicians wanted to find out whether thinking could be described so precisely that a machine could copy it. That curiosity came out of wartime work on code-breaking and early computers, and governments paid for it because they hoped it'd give them automatic translation, better defence systems and an edge in the Cold War. So the honest answer to "why was AI created?" has two halves: the scientists' reasons and the funders' reasons, and they weren't always the same.

In August 1955, four researchers asked the Rockefeller Foundation for $13,500 to spend a summer at Dartmouth College working on what they called "artificial intelligence". Their proposal is the closest thing the field has to a mission statement. If you want the names and dates, our guides to who created AI technology and how long AI has been around cover the timeline. This piece is about motives: what the founders were actually trying to do, who paid them, and what those goals tell you about the tools you use now.

Short answer: why AI was created

AI began as a scientific bet. Alan Turing asked in 1950 whether machines could think; John McCarthy and his co-authors wrote in 1955 that "every aspect of learning or any other feature of intelligence" could in principle be simulated by a machine. Military and government money then turned that bet into a research field, mainly because officials wanted practical payoffs such as translating Russian and supporting defence planning.

The background: four motives behind AI

I find it helpful to separate the reasons into four strands. They overlap, and the same people often held several at once, but each one left a different mark on the field.

Turing's question: can machines think?

Alan Turing's 1950 paper in the journal Mind opens with the line "I propose to consider the question, 'Can machines think?'" His motive was partly philosophical. He thought the question as asked was too vague to be useful, so he swapped it for a practical one: could a machine hold up its end of a typed conversation well enough to be mistaken for a person?

What's easy to miss is that Turing also had a plan for building such a machine, and it looks surprisingly modern. In the same paper he wrote: "Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child's?" The idea was to build something with "rather little mechanism, and lots of blank sheets" and then educate it. That's close to how today's machine learning works, even if the details differ hugely. He closed with a line I'd put on the wall of any AI lab: "We can only see a short distance ahead, but we can see plenty there that needs to be done."

Wartime code-breaking at Bletchley Park

Turing's interest didn't start in 1950. During the Second World War he was a leading code-breaker at Bletchley Park, and according to Jack Copeland's history at the Turing Archive, colleagues there remembered "numerous off-duty discussions" with him about machine intelligence. Donald Michie, who later founded a machine intelligence department at Edinburgh, recalled Turing talking about computers that could learn from experience and solve problems by searching through possible answers with rules of thumb. He used chess to illustrate the ideas.

The link between code-breaking and AI makes sense when you look at the work. Breaking ciphers meant searching enormous spaces of possibilities quickly, and it showed a generation of mathematicians that machines could do jobs that had looked like pure human cleverness. By 1947, Turing was telling a London audience, "What we want is a machine that can learn from experience."

The war shaped the Dartmouth group too. In the 1955 proposal, Nathaniel Rochester of IBM is described as having worked on radar for seven years, and Claude Shannon's listed work includes cryptography. Wartime projects built both the first computers and the people who wanted to push them further.

The Dartmouth proposal's seven aims

The Dartmouth proposal, dated 31 August 1955, is short and very readable. Its central line is the "conjecture that 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." It then sets out the plan: "to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves."

The authors listed seven problems to work on: automatic computers, programming a computer to use language, neuron nets, the theory of how big a calculation is, self-improvement, abstractions, and randomness and creativity. Read today, that list is uncanny. Language models, neural networks and machines that improve themselves are still the headline topics seventy years later.

One sentence in it aged badly, though. The authors argued that "the major obstacle is not lack of machine capacity, but our inability to write programs taking full advantage of what we have." I think history proved them mostly wrong on that point; computing power turned out to be one of the biggest limits for decades.

There was a second goal alongside building machines. Philosophy professor James Moor, quoted by Dartmouth, says the attendees "wanted to lay out a framework to better understand human intelligence." Building a thinking machine was, for some of them, a way to study the mind. Even the name had a motive behind it: McCarthy chose "artificial intelligence" partly to stay neutral and avoid the existing label "cybernetics", which, as the workshop's history notes, might have meant accepting Norbert Wiener as a guru or arguing with him. (Academic politics, in other words, gave us the most famous name in tech.)

Cold War money and its strings

Curiosity doesn't pay for computers. Governments did, and their reasons were more practical. In January 1954, Georgetown University and IBM ran a public demonstration that translated more than sixty Russian sentences into English using an IBM 701. Going by accounts of the Georgetown-IBM experiment, it was "conceived and performed primarily in order to attract governmental and public interest and funding", and the system had only six grammar rules and 250 vocabulary items. A US Department of Defense official had already offered to finance machine translation at a 1952 MIT conference.

The big money arrived in the 1960s. In June 1963, MIT received a $2.2 million grant from the new Advanced Research Projects Agency (ARPA, later DARPA), and ARPA went on to provide about $3 million a year, with similar grants to Carnegie Mellon and Stanford, according to Wikipedia's history of AI. ARPA's director J. C. R. Licklider believed the agency should "fund people, not projects!", which gave researchers unusual freedom.

That freedom didn't last. After the Mansfield Amendment of 1969, DARPA came under pressure to fund mission-oriented research with clear military goals, such as autonomous tanks and battle management systems. Translation money had already dried up: after a critical 1966 report from the ALPAC committee, the US National Research Council ended support having spent $20 million.

Real examples: the motives side by side

The table below is our own summary of who wanted what from early AI, built from the sources linked above. It's simplified, but it shows how different the goals were.

The relationship between the scientists and the funders reminds me of a polar expedition: the explorers want to reach the pole, while the navies and newspapers paying for the ships want charts, prestige or a good story. Both sides get something, but rarely exactly what they signed up for. Early AI worked the same way. Researchers wanted to understand intelligence; funders wanted translators and weapons systems, and each side's patience ran out at different times.

Benefits and limitations of the founders' goals

The broad, curious goals had real upsides. Because the Dartmouth group aimed at intelligence in general, they put neural networks, language and self-improvement on the agenda in 1955, and those ideas survived long enough to pay off. ARPA's hands-off money in the 1960s built the four university centres (MIT, Stanford, Carnegie Mellon and Edinburgh) that trained much of the field.

The downside was overconfidence. Researchers predicted a fully intelligent machine within 20 years, and funders believed them. When Britain's 1973 Lighthill report reviewed the field, it criticised AI's failure to reach its "grandiose objectives", which led to deep cuts in UK research. Hans Moravec later blamed the crisis on his own colleagues: "Many researchers were caught up in a web of increasing exaggeration."

There's also a mismatch problem that I suspect gets less attention than it should. A scientist who wants to understand thought and an agency that wants a working translator by next year will judge the same result very differently. Japan's 1981 goal to "carry on a casual conversation" wasn't reached for roughly 30 years. That doesn't make the goal foolish; it made the timeline wrong, and the money left before the science arrived.

Practical takeaways for you

Knowing why AI was created is surprisingly practical. Here's how I'd use it:

• Ask who's paying for a tool and what they want from it. In the 1960s it was defence agencies; today it's mostly companies that need subscriptions, ad revenue or enterprise contracts. The goal shapes the product.
• Treat big promises with patience. The founders thought a summer could produce "a significant advance", and the language problem took about seventy years. Today's timelines probably deserve the same caution.
• Remember that "AI" covers very different goals. A spam filter, a chess engine and a chatbot came from different motives, which is why our guide to the types of AI is a useful companion to what AI is.
• Look past the label. Some products use "AI" as a marketing word for simple automation, much as the 1954 translation demo was tuned to impress funders.

Next: the same motives, far more money

The motives haven't changed as much as you'd think. Today's biggest labs describe their goals in language the Dartmouth group would recognise. OpenAI's charter says its mission "is to ensure that artificial general intelligence (AGI)... benefits all of humanity", defining AGI as highly autonomous systems that outperform humans at most economically valuable work. Google DeepMind, which started in 2010 with the aim of building general AI systems, now puts it on its about page as: "Our mission is to build AI responsibly to benefit humanity."

The money, meanwhile, has grown beyond anything ARPA imagined. Stanford's 2026 AI Index puts US private AI investment at $285.9 billion in 2025, more than 23 times the $12.4 billion invested privately in China, and it reports that national AI strategies and state-backed supercomputing are expanding too. So the old mix is back: scientific ambition, commercial money and national competition, all pulling at once.

Where do I land? I think curiosity is still the engine, but commercial and strategic pressure now sets the pace, much as defence money did in the 1960s. If the pattern holds, I'd expect another round of disappointment wherever promises outrun results, though probably not a full winter, because this time the tools already earn money. That last part is a guess, not a forecast. If you want to follow the argument further, our explainers on what AGI means and the future of AI pick up where this one stops.

FAQ

How long has AI existed?

As a named field, AI is about 70 years old. John McCarthy and three co-authors coined the term in a 1955 proposal, and the Dartmouth workshop that founded the field ran in summer 1956. Turing's "Can machines think?" paper came earlier, in 1950, and wartime discussions at Bletchley Park go back to the 1940s.

When did AI first come out?

As a research field, in 1956. For the general public, much later: most people noticed it on 30 November 2022, when OpenAI released ChatGPT, though spam filters had used AI for years.

Who created AI?

No single person created AI. John McCarthy named the field with Marvin Minsky, Nathaniel Rochester and Claude Shannon in their 1955 Dartmouth proposal. Alan Turing laid the groundwork in 1950, and Allen Newell, Herbert Simon and Cliff Shaw wrote what's often called the first AI program. Our full guide covers who created AI technology.

How was AI made?

Early AI was made by writing programs that followed hand-crafted rules, such as logic and search, on room-sized computers that filled whole labs. A rival approach, loosely modelled on neurons, let machines learn from examples instead; Frank Rosenblatt's perceptron did this in 1958. That learning approach, helped by huge datasets, faster chips and the 2017 Transformer design, is how today's chatbots and image generators are built.

Who developed AI?

For its first decades, AI was developed mainly at universities such as MIT, Stanford, Carnegie Mellon and Edinburgh, with much of the money coming from the US defence research agency ARPA. Today, most development happens at companies such as OpenAI and Google DeepMind, alongside university labs and government-backed projects around the world.