This is part 3 of our AI history series. Check out part 1 and part 2 if you missed them.
In the summer of 1956, a small group of researchers gathered at Dartmouth College in Hanover, New Hampshire. They had received a modest $7,500 grant from the Rockefeller Foundation for an eight-week workshop.
The subject they were studying didn’t even have a proper name; however, a proposal written by the researchers in the previous year opened with this: “We propose that a 2 month, 10 man study of artificial intelligence be carried out... The study is to proceed on the basis of 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.”
John McCarthy, a young Dartmouth mathematician, had coined the term “artificial intelligence” specifically for the proposal, and it has stuck ever since.
The term “artificial intelligence” was as much a business play as a philosophical one, intended to capture government interest – and therefore funding – during the Cold War, and the Dartmouth proposal listed specific subproblems that they were seeking to solve: Automatic computers, programming languages, neuron networks, self-improvement, abstraction.
That list relied on a foundational assumption: That symbolic description would be the medium of intelligence. This meant that humans could write rules to build AI, using formal languages, logical rules, and procedural representations. A rival theory would argue that AI would be achieved by analyzing data and extracting patterns from it. While the former would dominate the field for decades, the latter would eventually win out.
Regardless, the original proposal envisioned 10 people working intensively for 8 weeks throughout the summer. In reality, attendance was staggered, with people dropping in for a week or two rather than staying the whole summer. At any given time, only a handful of participants were present. There was no structured agenda, no formal proceedings, no concluding report.
Still, two demonstrations would prove visionary.
The first was a demo called the Logic Theorist, built by researchers at the RAND Corporation, that could prove mathematical theorems. Applying new reasoning approaches, rather than brute force calculation, it proved 38 of the first 52 theorems from Principia Mathematica, a foundational work of logic. The researchers behind this submitted their result to the Journal of Symbolic Logic; however, the journal rejected it, reportedly because one of the co-authors was a computer.
The second demonstration came from IBM engineer Arthur Samuel, who had been building a checkers-playing program that was able to learn. Rather than being explicitly programmed with every strategy, Samuel’s program analyzed its past games and adjusted its play accordingly. It showed that machines could get better at tasks without a human reprogramming them for every improvement. (Six years later, an improved model would defeat a Connecticut state checkers champion, a step toward the more recent AI models that have beaten chess, Go, and Dota champions.)
These two programs marked different achievements: The Logic Theorist manipulated formal rules to derive conclusions; Samuel’s checkers program learned from experience and improved through feedback. Both approaches would define the field for decades.
While the demos marked progress, they had been prepared outside of Dartmouth and merely shown there, and by the end of the summer of ‘56, little concrete progress had been made. Yet an AI community had emerged, collectively approaching machine learning, game-playing programs, language processing, and other cutting-edge computer domains. From this seed would grow AI as we know it today.
The phrase AI itself compressed many distinct technical problems – pattern recognition, search algorithms, knowledge representation, learning – into a single outcome: A thinking machine.
Beyond focusing researchers within a new domain, aligning around this phrase AI had an added effect: Simple and catchy enough for the media to pick up on, it also outlined a field that scientists could join.
In this way, 1956 set the stage for what was to come: Technological progress, new research, media hype, and a lot of money.
Yet to unlock that, researchers needed to create programs that could do something that looked like intelligence under laboratory conditions. That first wave was about to begin.
If you enjoyed this article, please consider upgrading to a paid subscription. Paid subscribers fund Roca’s journalism. Without you all, there would be no Roca!
Editor’s Note
Thanks for reading. Should we keep writing about the history of AI? Let us know here or in the comments below. We’re curious to hear your feedback.
And, as usual, you can read our latest here:
We’ll see you back here tomorrow.
—Max and Max
I was a couple months old when that happened but the Kiewit Computation Center remained as I was growing up. You couldn't be anymore Cool if you had an outline of Snoopy Kicking a football that was printed with type covering 13 sheets of green and white striped computer paper. P.S I never had one.
I find this history of AI very interesting! I would enjoy further reporting on how we got to where we are now. I think reporting on its origins helps take away some of the fear/doom around this (new?) technology.