Marvin Minsky: in search of the smart machine

A pioneer in computer research since the mid 1950s, Marvin Minsky is often mentioned in the same breath as Artificial Intelligence.
Currently a professor at MIT’s Department of Electrical Engineering and Computer Science, Minsky is closely involved in that university’s hotbed of new information technology, the Media Lab.
    Electric Word’s Colin Brace recently came face to face with Minsky in his heavily-trafficked office on the fourth floor of the stunning Media Lab building.

“I’d like to propose the four-year software patent.” Minsky pauses to reconsider. Better yet: “If you get a really good idea in programming, you should have the exclusive rights to it for as long as the machine you’ve designed it for lasts.” Minsky chortles. We’d been talking about the state of the art of the computer interface, something he has been inextricably involved with throughout his long career in computer science. To him, though, the attempts of industry giants like Apple and Lotus to protect their programs’ human interfaces are not simply inane but highly damaging to the further development of this crucial aspect of computer design.

PATENT? COPYRIGHT?
Minsky sees current computer interface development as fragmented. “Some interfaces have all kinds of wonderful improvements - but they don’t get incorporated by the competitors. Under the present legal situation, it is impossible to have incremental advances. If you get a good idea, someone else can’t just use it.
    “The question is: will the public stand for all the lawyers?” How long will we balance one kind of protection against another? Like women wanting the streets back — at some point, programmers will want their programs back. Will they be liberated? I don’t know.
    “In the computer field, we got used to the idea that a patent lasts 17 years and a copyright lasts 40 or whatever. We are stuck on this old idea that progress is slow. But maybe we should be more flexible. In some areas of computer science, what you learn in four years of college is obsolete four years later.
    ”In that case, the protection rules should have a lifespan of four years insteadof 17. How long is a copyright? 100 years with renewals? Long enough to be infinite in the case of software.
    ”We forget the reasons why we have copyright and patent laws. Some of them are to make sure the inventor gets some reward for his invention, some are there to provide incentives to inventors, and some are so that people can recover their capital. All kinds of different reasons are mixed up here, and we have to sort out which ones should be supported and which are doing more harm than good.
    ”In the case of the look-and-feel patents, I find these particularly obnoxious because many of my students invented these techniques - the different menus and displays. Twenty-five years later, why should some company get an iron grip on some combination of these features invented by someone else? It doesn’t seem right.” Nonetheless, Minsky agrees with Jef Raskin about the primitive state of present interface design. “Raskin’s right.”
    Current operating systems don’t see what the person is doing. Let’s hope their lifespan is short.” Asked about spinoffs from AI research on computer interface design, Minsky replies: “AI research, as I see it, is essentially the same as advanced computer science. Namely AI people try to get computers to do what computers don’t do yet. Almost everything is a spinoff in some sense.
    “Wordprocessing itself came from AI labs, because these people realized they might as well be using computers for editing text as well. That’s not usually considered AI. But as I recall, the first editors were done in AI groups. Likewise the first networks and timesharing systems.” Minsky points out that a wide variety of vision and pattern recognition systems, expert systems, and optical character recognition systems are currently available. “’You can now put an OCR system into a PC, but in the ‘60s, that was considered to be AI research.”

COMMON SENSE REASONING
Machine translation is the Holy Grail of the language industries, but Minsky sees the fundamental technological problem as broader, namely in terms of syntactic versus semantic understanding.
    MT systems, he says, “will always be crude until the machine understands what the words mean. A good machine translation system will have to wait for a commonsense reasoning system.” Referring to the Cyc project to develop a database of commonsense knowledge at MCC in Austin, Texas, Minsky would like to see perhaps five such projects going on around the country - maybe more than that around the world. He would ensure that they didn’t communicate with each other.
    “The worst thing to do is to share ideas when you’re not sure the ideas are good. These groups should work on their own and not try to be compatible with other groups.” Conversely, he doubts whether technology will affect natural language much. “For programming, natural languages are really pretty bad because it’s very hard to express connections between things. We don’t have enough pronouns in language for expressing complicated relationships.”

CAD FOR ALL AGES
Where does Minsky see applications emerging for this? Within education or commerce?
    “That’s a nice issue. In the last century or two, we’ve gotten used to the idea that education should be separated from professional skill. Certainly in some education systems, we discourage children from learning too much about any particular thing.
    “I think this is a disaster. I meet young people who’ve been to school a lot, and they’ve learned a little bit about many things.
But they’ve never learned what it’s like to become good at anything. In my view, there shouldn’t be such a clear distinction between educational and professional software.
    “Young children should be encouraged to use simplified versions of adult systems, like CAD systems and simulation languages. There ought to be continuity. Apprenticeship is the best way to learn a skill — you actually start out by helping someone else with areal problem.
    ”Sometimes, it’s hard for the expert to find suitable things for the beginner to do. Eventually, we get tired of that, and we make schools where all the children uniformly do things that are so terribly oversimplified that they’re no use to anyone at all. We ought to move back a little. With the smart software of the future, there’ll be a good chance for kids to experience doing substantial projects.
        lright>If kids take biology, they ought to be able to make simplified animals and see how they behave, and mutate them modify their behavior- and get some idea of what kind of mechanisms lead to lifelike behavior. Mike Travers here is trying to develop an animal programming system, a simulation system usable bychildren. It’s part of Apple’s Vivarium project.”
    Minsky turns to the graduate student, sitting by a Mac: “Can you turn on an ant?” A bevy of ants starts trudging across Mike’s screen in search of food.
”The idea is to develop a language to enable you to create your own animal. You can say: ‘If an ant had this behavior, would it survive in that environment?” Minsky sees the current plague of computer viruses in biological terms, as software catching AIDS. If indeed biological evolution is spurred on by parasites, will viruses cause software to evolve? “It may not improve it. In fact, it will make it less flexible - harder to link things and make versatile systems.”
    But, he adds with a smile: “If everyone were to write their own operating system, it would induce some genetic variability! That’s how sex evolved, at least that’s the current theory!”