Artificial Intelligence
Introduction
Artificial intelligence is a term made up in 1955 to avoid a specific person.
John McCarthy was at Dartmouth, putting together a funding proposal with Marvin Minsky at Harvard, Nathaniel Rochester at IBM, and Claude Shannon at Bell Telephone Laboratories. The work they wanted to do was cybernetics. Everyone knew it was cybernetics. But calling it cybernetics meant dealing with Norbert Wiener.[1]
So they picked a different name, and we've been living with it ever since.
Why not cybernetics
Wiener taught at MIT and had named the field himself in Cybernetics (1948), the first book on electronic computing to reach a wide audience. He was working on how to control and automate systems. During the war he built one of the first automated weapon systems, which is worth remembering now that we worry about AI-guided weapons as if the pairing were new.[1]
By the early Cold War he was the loudest voice in any room about the future of computing. Sometimes literally. If a computing conference was happening anywhere near Cambridge, Wiener had a habit of turning up and lecturing everyone at length.[1]
You can't hold a conference about computing and keep out the man who named the field. You can hold a conference about something else. McCarthy said later that one reason for inventing the term was "to escape association with cybernetics," and that he didn't want to accept Wiener as a guru or argue with him.[1]
The proposal asked for a two month, ten man study. The workshop ran the following summer, in 1956. Selfridge, Rochester, Minsky, McCarthy, Solomonoff, Milner and Shannon are the ones in the photograph.[1]
Cybernetics is named after steering. It points at a mechanism. Artificial intelligence is named after a human quality, and it points at mental concern, reaching deeply into the history of the industrial revolution, and that human thinking might be industrialized next.
That difference has done a lot of work over seventy years. Because the term was invented to draw a line rather than to describe a method, it got defined loosely enough that almost any automated system can be called AI. Which is exactly why it keeps selling.[1]
It also means every public conversation about the field starts on the wrong footing. Say intelligence and people think about souls, about being replaced, about how long any of this lasts. They panic first and look at the actual system second, if at all.[1]
Same cycle, different decade
The people at that first workshop raised millions to build a machine as smart as a person. They figured one generation would do it.[1]
Two generations later, the pattern is easy to recognize:
- A research lab makes something that works.
- Excitement forms at the edges, among marketers and paid evangelists rather than the people who built it.
- Expectations inflate past what the technique can do.
- Executives ask engineers to deliver the inflated version.
- The systems keep doing the nerdy automation tasks they're good at.
- Everyone feels let down.
The 2014 round was about voice assistants. Siri, Alexa, Cortana, with Hollywood supplying the version people actually pictured, Jarvis in Iron Man and the Scarlett Johansson voice in Her.[1] That wasn't new either. René Barjavel's La Nuit des temps (1968), published in English as The Ice People, has the protagonist attached to Blanche, the voice of his car's computer.[1]
Around that time I started getting pitches from big companies who wanted magic systems with AI sprinkled on top. When I asked what the system needed to do, it turned out they weren't interested in AI. They wanted to look current. They'd read somewhere that this mattered and hadn't gone further than that, and they got impatient when I tried to walk them through what building it would actually involve.[1]
A consultant at an arts festival once told me I should be talking about AI more, since people are frightened of it and fear pays well. I like giving talks. I'm not interested in selling people their own panic back to them.[1]
Voice assistants didn't vanish. They settled into the places where they genuinely help, mostly directions while you're driving, which works because your hands and eyes are already busy and the voice sits in the periphery.[1]
Large language models will run the same course. Some real uses will stick. From inside the noise it's too early to say which ones.[1]
A better name
"Alongside technologies" is less catchy. That's the point. This stuff shouldn't be catchy, and the phrase says the true thing, which is that these systems need a person guiding them to work at all.[1]
An honest name for this field wouldn't need to:
- Borrow a human quality
- Promise something general
- Hold up as marketing
- Hide what the system actually does
What this means for design
See IV. Technology should amplify the best of technology and the best of humanity. The automations worth building take what machines are good at and what people are good at and use both. Trying to make a machine into a person has never gone well, and it isn't going to.[1]
Name a field after your rival and you spend seventy years arguing about the wrong thing. Name a machine after a person and it gets measured against one.
References
Sources
- Case, Amber. "Inside the Very Human Origin of the Term 'Artificial Intelligence' and Its Seven Decade Boom/Bust Cycle." Medium. [URL]
- McCarthy, John, Marvin L. Minsky, Nathaniel Rochester and Claude E. Shannon. "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence." 31 August 1955.
- Wiener, Norbert. Cybernetics: Or Control and Communication in the Animal and the Machine. MIT Press, 1948.
