My wonderful AI predictions
In 1970 I wrote the introduction to a TV documentary on artificial intelligence. It was very clever — but completely wrong!
As a preamble I need to tell you this: when my originally intended career as a teacher of philosophy ground to a halt (I realized that there would be no vacancy for me at university to get a post — ever) I embarked on an alternate path. I became a science journalist, making documentaries for German TV. That is described in this article, which might be of interest.
Well, after making half a dozen documentaries on science subjects — for instance on how the pyramids of Egypt were built, on superconducting materials, Rubik’s cube, debunking astrology — I was commissioned by a major publishing house to research the current state of “Artificial Intelligence”, an emerging field that was still in inverted commas. The publishers gave me funding to travel around the world, visiting all the major AI labs in the U.S. and Japan, and talk with all the famous AI pioneers. I interviewed Claude Shannon, Marvin Minsky, Ray Kurzweil, John McCarthy, Raj Reddy, and a number of others who were actually building AI, as well as thinkers and philosophers who had written on the subject, like Daniel Dennett, Hubert Dreyfus and John Searle.
Before I continue with today’s discourse I need to tell you about the last two. I visited Searle because I had written my university thesis on his work in Speech Act Theory. When he discovered I was researching Artificial Intelligence, a subject he himself was deeply interested in, he refused to discuss speech acts. We talked about AI and nothing else.
At the end of an hour-long fairly adversarial debate he took me to lunch and asked the famous Hubert Dreyfus, a leading critic of AI, to join us. That led to even more heated discussion, but one we all clearly enjoyed. At some stage I told Dreyfus that he had to be careful not to declare, in his writings, that computers would never be able to do something they were already doing (Searle guffawed). In the end Hubert gave me a copy of his seminal book with a little dedication, calling our discussion “the most philosophical interview I ever took part in”.
I was able to debate the subject meaningfully with these two famous philosophers because I had just been through all the AI labs in America and had become quite knowledgable. And after touring the AI labs in Japan as well I became a recognized “expert” on the subject in Germany. And as such I was commissioned to work on a TV documentary on Artificial Intelligence.
The first task I undertook was to write an introductory sequence — and I delivered the script for that. I proposed that the documentary should start by following a man who drives to the university campus of Stanford, goes to the cafeteria in the first floor, where he fetches himself a coffee. He spots a friend in the far corner of the room and goes over to join him. “You look worried,” he says. “Yes,” says his friend, “I am stuck on this theorem and cannot find a solution.” He describes the equations that have failed, to which our subject says: “Why don’t you try ….” (giving different equations). “Wow, that is brilliant,” says his friend. “I have been mulling over it for weeks, and you solve it in seconds!” The message was: here is something no computer will ever do (we were going to talk about Hubert Dreyfus’ book).
The idea was that we would repeat the intro sequence, telling viewers what computers actually would never be able to do: steer a car through traffic, park it safely, traverse the campus lawn on two feet, climb the stairs, operate the coffee machine, spot and recognize a friend in the far corner of the room, walk over, see that he was unhappy, understand his words in the noisy cafeteria, comprehend what he was saying and react to it meaningfully. Oh, and the part where it finds a brilliant solution to his mathematical problem: that was easily possible. In fact, it had already been done —I gave a problem that computers had recently very cleverly solved.
My intro was well received, everyone thought it was very clever, summarizing what we were going to show perfectly. The film was never made, for planning and budget reasons. For this, in retrospect, I am thankful. Because I was wrong. Boy, was I wrong.
Today, forty years later, it turns out, computers can do everything I had deemed impossible. We are on the verge of seeing ubiquitous self-driving cars (I now predict that in twenty years humans will not be allowed to drive — too dangerous). You can already buy a vehicle that will park itself more reliably than you could ever do — I have driven one that parallel parks into spaces I could never manage. And walk over a campus lawn, climb stairs?
What about spotting and recognising a friend from a distance in a cafeteria? Modern face recognition is scary. I have written about this: I store most of my pictures in Google Photos. Periodically it will show me a person in random images from my photo database, asking me to confirm if they are of the same person. It is learning to recognize that person.
After ten or so images it knows me, for instance, and searching for “Frederic Friedel” will show all pictures in which I appear. In the above example it is showing me a tiny blurred reflection of myself in the glass window behind the lady I am taking the picture of (click to enlarge). It found that! The face recognition software is truly awesome — hard to believe.
Voice recognition and understanding language? In 1979 Raj Reddy, the leading expert in this area, assured me that it would never happen. Today my Android smartphone disproves that. Sometimes it will speak from my pocket, answering a question it thought I had addressed at it: “The state of Georgia has an area of 59,425 square miles and a population of 10,617,423…” And everyone is used to asking Amazon Echo for help in everyday things: “Alexa, when is the next train into town?” Currently I am in the process of switching from typing to dictating text. This “voice notepad” or Dragon Naturally Speaking do an excellent job.
So in my intro to the AI documentary I got almost everything wrong. My only excuse is that it was forty years ago, and everybody was mistaken about Artificial Intelligence — nobody saw what was coming.