Connectionists use learning rules in big networks of simple components -- loosely inspired by nerves in a brain. Connectionists take pride in not understanding how a network solves a problem.

Stork: Is that really so bad? After all, the understanding is at a higher, more abstract level of the learning rule and general notions of network architectures.

Minsky: Yes, it is so bad. If you just have a single problem to solve, then fine, go ahead and use a neural network. But if you want to do science and understand how to choose architectures, or how to go to a new problem, you have to understand what different architectures can and cannot do. Connectionist research on this problem leaves much to be desired.

Stork: What were you doing in the mid-sixties?

Minsky: I was interested in heterarchical systems for AI -- ones that consisted of many interacting parts -- applied to the same environment as Winograd. We wanted to solve robot problems and needed some vision, action, reasoning, planning, and so forth. We even used some structural learning, such as was being explored by Patrick Winston.

After about 1970, Winograd and his students tried to add more stuff. A major problem with AI research then was that Ph.D. dissertations were huge programs, and only the writer knew how they worked. Winograd thought you needed a second database, on how the design was done, so that all the theses could be unified. But we didn't have a good tool for knowledge representation. So Winograd and colleagues developed KRL -- knowledge representation language -- but this didn't work out, and the project fizzled. It turned out that it took as much time to cast a dissertation into KRL as it took to write the dissertation program in the first place! Thus this line of research was declared a failure, as summed up by Winograd's statement to the effect that you can't write big, smart programs.

Stork: But before that, again back to the mid-sixties, wouldn't you agree that the field was quite optimistic? After all, in a Life magazine article, you were quoted as saying "In from three to eight years we will have a machine with the general intelligence of an average human being. I mean a machine that will be able to read Shakespeare, grease a car, play office politics, tell a joke, have a fight. At that point the machine will begin to educate itself with fantastic speed. In a few months it will be at genius level and a few months after that its powers will be incalculable."

Minsky: Oh, that Life quote was made up. You can tell it's a joke. Herbert Simon said in 1958 that a program would be chess champion in ten years, and, as we know, the IBM group has done extremely well, but even now [Deep Blue] is not the undisputed champion.

As for optimism, it depends on what you mean. I believed in realism, as summarized by John McCarthy's comment to the effect that if we worked really hard, we'd have an intelligent system in from four to four hundred years.

Stork: What was your relationship to the filming of 2001?

Minsky: Stanley Kubrick knew we had good graphics around MIT and came to my lab to find out how to do it. We had some really good stuff. I was very impressed with Kubrick; he knew all the graphics work I had ever heard of, and probably more.

Stork: Did he use any of your images or image techniques directly?

Minsky: No. At that time, all our images were 512 by 512 pixels -- far too coarse for film audiences, and, more importantly, far too coarse for what we all knew would be available by the year 2001. Our images were on a twenty-inch raster-scan TV screen, before good color rendition. Kubrick was quite secretive, however, and I never really knew just what he would or wouldn't be using from MIT.

Stork: What kinds of images did you show Kubrick?

Minsky: We were showing him spectacular, out-of-this-world fractal images -- you know, where a simple function was iterated again and again. For instance, one could start with a simple line segment, and then at one end, splice on two shorter line segments at some angle. At the end of each of those you add two more, and so on and so on. Pretty soon you have a simple tree. You can make some spectacular fireworks this way too.

The problem was that none of these were images that a HAL would be showing to the crew, so Kubrick never used any of our images.

Stork: Were you looking at such fractals just for fun? What was the research?

Minsky: In 1963 I had made some sketches for a "recursive structure" based on my fractal and bifurcating work. It looked so humanoid. We never built such a robot, but we used it in simulation.

Stork: What was it like on the set in England?

Minsky: The moon set at Shepperton, south of the Thames near London, was very impressive -- very realistic. I remember clearly the excavation set with its lights and the monolith.