If you want to make a difference, don’t follow the crowd, Marvin Minsky advises today’s students. Don’t go into the most popular field. “That could be a disaster. When I started to work on artificial neural networks, only four other researchers were involved with this field. But today, there are many thousands of them. Interesting discoveries come only every few years—so each researcher has less than one chance in 1000 of making significant contributions,” Minsky said.
“Almost everything I’ve worked on was a success because it was done in unexplored areas,” said the man known for major advances in artificial intelligence, neural networks, cognitive psychology, and the basic theory of computations. He invented the confocal scanning microscope in 1955, whose value was finally recognized 30 years afterward.
Minsky also was involved with developing advanced technologies for symbolic mathematical computation as well as theories about semantics and knowledge representation, machine perception, and computer graphics. His pioneer work in robotics and telepresence influenced the work of many others in space exploration and other areas.
STUDYING PSYCHOLOGY
This desire to make a difference helped Minsky choose his career path. “As a student, I was very good at mathematics and was ‘adopted’ by Andrew Gleason and Claude Shannon—both of them world-class mathematicians. But soon I realized that one person great in that field can do more than a thousand good practitioners,” said Minsky.
“I also was a good physicist. But after doing some work with Richard Feynman and Edward Purcell, I realized that physics had no need for me. Then I encountered the ideas of some early great psychologists, such as William James and Sigmund Freud, and started to think about psychology—and then met George A. Miller and J.C.R. Licklider—two young professors of psychology who were pioneering the new field that is now called cognitive psychology,” he said.
“I found that my ideas seemed to be on their level—and that few other people were doing such things—so that’s what I started to focus on. And in 1950, George Miller obtained funding for me to build the first neural network learning-machine,” Minsky said.
So eventually Minsky focused on the relation between ideas about psychology and computing machines. Later, Licklider went to the Advanced Research Projects Agency (ARPA) in Washington and promoted the development of time-shared computers, promoted and established funding for the Internet, and supported Minsky’s new Artificial Intelligence Laboratory at the Massachusetts Institute of Technology (MIT), where he collaborated with John McCarthy, another early AI pioneer, and Seymour Papert, with whom he developed many new concepts about how children learn and think.
“I was fortunate to usually have one great collaborator to work with during each stage of my career,” said Minsky. “It is wonderful to find someone who thinks mostly like you do, but a little differently. Then, when you get stuck, you can bring them your problem and they suggest a different way of looking at it.”
As an undergraduate in the late 1940s, Minsky was hanging around Harvard’s computer laboratory. The program of that early computer could only be changed by someone manually pulling plugs and wires and then re-connecting them. One day a student asked why a computer couldn’t change its own program. “The director said that would be a waste of time, because computers should be producing results. That’s an example of someone’s imagination getting stuck,” said Minsky.
“So, the idea of automatic programming got developed by others, and about 10 years later I wrote an article on how a computer could prove theorems in geometry and calculus (instead of just doing arithmetic) and eventually started a project to get computers to do higher-level symbolic mathematics. Under the guidance of a great graduate student, Joel Moses, this developed into a system called MACSYMA that soon was in use around the world,” he said.
SURROUNDED BY THE BEST
Minsky credits much of his success to his professorship at MIT, where he oversaw a “whole generation of young students, many of whom didn’t fit anywhere else,” he said.
“They were brilliant freshmen obsessed by computers, and I also got them interested in artificial intelligence. Every time I suggested something, it would soon become a working system. For example, this led to the first robot with a hand and eye. Its vision programs could recognize structures made of wooden blocks, and its hand could reach out and replicate that structure with other blocks of similar sizes and shapes,” he said.
“Today we see many students making similar kinds of robots, but they’re usually repeating earlier projects instead of doing something new,” said Minsky. “I’m a bit disappointed that so few of them feel they should do something new, instead of doing what’s popular.”
CAN MACHINES SHOW EMOTION?
Minsky now focuses on AI combined with common sense. “Programs can make airline reservations, and assemble products in factories, but no computer can yet read a children’s story and understand what its sentences mean,” he said. “No machine yet has the common sense of a typical three-year- old, so that it can look around a room and say, ‘That’s a chair, that’s a bookshelf, and there is a bottle of orange juice.’”
Another challenge is trying to make smart machines that feel. In his book, The Emotion Machine, Minsky says emotions, intuitions, and feelings are not distinct things, but different ways of thinking. He examines them to show how our minds grow from simple thoughts to more complex ideas and predicts that machines will one day be as conscious as humans are, once they begin to build useful models of what has been recently happening inside their own mental processes.
“Today, most psychologists try to imitate physicists and keep trying to find a few simple ‘general laws’ for explaining how human minds work. However, I think that this is a bad mistake, because we know that each brain has evolved several hundred different parts, each of which works in somewhat different ways,” said Minsky. Thus, the book suggests how a thinking and feeling machine could be built using many different methods. Holding up progress, laments Minsky, is the disappearance of funding.
“Basic research in the United States is in terrible shape because budgets have been going down, and most of our formerly great basic research labs have disappeared— forcing many of our most promising young potential scientists to move into product development and financial activities,” he said. “Our leaders failed to recognize that much of our prosperity grew out of basic research that was typically done 20 years before those products appeared. So today, the U.S. economy is floundering—and our conceptual growth is moving to other continents.”
NO TIME FOR LEISURE
“I don’t have much time for hobbies or pastimes,” said Minsky, “because I keep thinking most of the time. I’m always impelled to uncover new reasons for things and how to make them better—or simpler.” Minsky credits science fiction with inspiring him. “I’ve partly lived in the world of writers like Asimov, Benford, Heinlein, Niven, Sturgeon, and Vinge—all of whom became friends of mine. Lately, I’ve been inspired by ideas from Greg Egan and Robert J. Sawyer.”
When stuck at developing technical theories, he likes to write music, particularly classical fugues. “Fugues use counterpoint, in which several different things happen at the same time, so you have to make yourself think several different thoughts at once. If I had enough time, my goal would be to write a quartet as good as Beethoven’s Opus 131.”
Looking back, Minsky cites three achievements as his most significant. “In 1955, I invented and built the first confocal scanning microscope. I did not stop to commercialize it, but eventually it revolutionized microscopy in laboratories around the world,” he said.
“In seeking to understand the nature of computing, I discovered several new, remarkably simple but ‘universal’ machines—that is, ones that can do all kinds of computations— including one that can do nothing more than increase or decrease either of just two numbers,” he said.
“In the field of psychology, my two books The Society of Mind (1985) and The Emotion Machine (2000) describe many new theories and ideas about how the processes in our brains might be organized into multiple levels of structures that I called K-lines, Frames, Panalogies, and the Critic-Selector model of Thinking. I’m sure that these could also be used in machines to achieve some aspects of human intelligence.”