The Men Who Saw Tomorrow

Before artificial intelligence became an app, a business, a laboratory race or something millions of people could speak to directly, it was a question.

Could a machine do something that looked like thinking?

In 1950, British mathematician Alan Turing published a paper called Computing Machinery and Intelligence. He opened with a question that would follow computer science for generations: can machines think?

Turing understood immediately that the question was difficult because people could argue forever about the meaning of the words machine and think. So he changed the problem.

Instead of demanding a perfect definition of intelligence, he proposed what became known as the imitation game. If a person communicated through text and could not reliably determine whether the other participant was a human or a machine, then the machine's behavior deserved serious consideration.

Turing did not build the artificial intelligence systems people use today. He did something more fundamental.

He made machine intelligence a question that could be investigated.

Giving the Question a Name

Five years later, another step was taken.

In 1955, John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon prepared a proposal for a summer research project at Dartmouth College.

The proposal used a name that would eventually travel around the world: artificial intelligence.

Their central idea was bold. They proposed that aspects of learning and intelligence might be described precisely enough for machines to simulate them.

That was not the same as claiming that machines were already intelligent. It was a research program.

If intelligence had structure, perhaps parts of that structure could be expressed mathematically.

If reasoning followed patterns, perhaps machines could manipulate those patterns.

If learning involved changing behavior after experience, perhaps a machine could be designed to change too.

Dartmouth, 1956

In the summer of 1956, researchers gathered at Dartmouth for the Summer Research Project on Artificial Intelligence.

That meeting is often treated as one of the founding moments of AI as an organized research field.

The people involved did not all agree on how intelligence worked. That disagreement became part of the history.

Some researchers concentrated on symbolic reasoning: representing knowledge through rules, symbols and logical operations.

Others explored learning, perception, neural ideas, language and problem solving.

What united them was not one finished theory of intelligence.

It was the belief that intelligence could be studied computationally.

Making Reasoning Executable

Around the same period, Allen Newell and Herbert Simon were developing programs that attempted something remarkable for the time: using computers to perform forms of symbolic reasoning.

Their Logic Theorist program demonstrated that a computer could search through possible steps and produce proofs for mathematical statements.

This mattered because computers were beginning to move beyond arithmetic.

A machine could be instructed to manipulate symbols according to rules and search through possible solutions.

That idea became one of the foundations of early artificial intelligence.

The machine was not simply calculating a larger number.

It was participating in a process that people associated with reasoning.

Building a Home for Artificial Intelligence

The idea continued spreading into universities and research laboratories.

At MIT, John McCarthy and Marvin Minsky helped establish a coordinated artificial-intelligence research group in 1959.

Researchers began attacking problems in vision, language, planning, robotics, mathematics and learning.

They were not building one machine called artificial intelligence.

They were breaking intelligence into problems.

How does a system recognize an object?

How does it decide what action to take next?

How does it represent knowledge?

How does it learn from experience?

How does language connect words to meaning?

Those questions still exist.

They Did Not Predict ChatGPT

It is easy to look backward from modern AI and imagine that the pioneers saw everything coming.

They did not.

They did not predict the exact architecture of today's language models, the internet that would supply enormous digital datasets, modern graphics processors, smartphones, cloud computing or the precise form of conversational AI.

Their predictions were sometimes right, sometimes wrong and sometimes far too optimistic about how quickly progress would happen.

But they saw something deeper.

They saw that intelligence itself could become an engineering problem.

Reasoning could be studied.

Learning could be modeled.

Language could be represented.

Machines could search through possibilities.

And a computer might someday do more than calculate.

The Future Arrived in Pieces

The road from Turing's question to modern artificial intelligence was not smooth.

There would be breakthroughs, disappointments, competing theories, funding collapses and periods when many people stopped believing the promises.

But the central questions survived.

Could machines learn?

Could machines reason?

Could machines understand language?

Could machines solve problems that once seemed to require a human mind?

Seventy years later, those questions are no longer confined to research laboratories.

Ordinary people ask machines questions every day.

Students use them.

Workers use them.

Artists use them.

Scientists use them.

Businesses use them.

Governments use them.

And the argument about what machine intelligence means is still not finished.

The men who helped create the field did not see the exact future.

They saw the road.

Then they started walking.

The next chapter begins with what happened when that road became colder than almost anyone expected.