Marvin Minsky’s Legacy of Students and Ideas

Deciphering how brains and machines think and learn

Geniuses are meat machines—that’s how Marvin Minsky once characterized human beings—just like the rest of us, only much more efficient. And Minsky, who passed away in January, was certainly one of the most efficient meat machines this century or the last has ever seen.

There are the curriculum vitae facts of his genius: the degrees in mathematics from Harvard and Princeton; the cofounding, with John McCarthy, of the Computer Science and Artificial Intelligence (AI) Laboratory at MIT; the invention of the first head-mounted graphical display, the confocal microscope, and the first randomly wired neural network machine. He was a pioneering computer scientist, cognitive scientist, and roboticist, a fellow of IEEE and of the American Academy of Arts and Sciences, and the recipient of numerous honors and awards, among them the Turing Award, the IEEE Computer Society’s Computer Pioneer Award, and the Franklin Institute’s Benjamin Franklin Medal. He left his mark on every field that captured his interest, moving through several with seeming ease before finding his life’s work: creating a theoretical framework in which to build machines that could understand as well as calculate.

But then there is the much-harder-to-characterize genius of his life as a public intellectual, provocateur, mentor, and friend.

In addition to thinking about how computers could be taught to learn as children do, bootstrapping their knowledge to learn more and more, he wrote about his efforts with verve and wit. Reading “Steps Toward Artificial Intelligence,” written in 1961 and published in the Proceedings of the Institute of Radio Engineers, it’s important to remember that the computers that sparked his inquiries then had mere kilobytes of memory. Minsky’s best-known book, The Society of Mind (1986, Simon & Schuster), laid out how a human mind could emerge from elements, like neurons and neuronal pathways, that have no minds themselves. His articles, books, and interviews are a treasure trove of epigrams and ideas—funny, profound, confounding, and sometimes infuriating.

The most important part of Minsky’s legacy just might be his students and his students’ students. If you take a look at the Mathematical Genealogy Project, you’ll find that he has 36 students and 1,123 descendants in this one knowledge domain alone. One of my colleagues quipped, “It’s kind of like Mozart giving flute lessons.”

Minsky wanted the AI that many find scary and threatening: He wanted to build computers that could actually think, not just machines that appear to think by virtue of their prodigious abilities to grind through masses of data. Steven Cherry, director of TTI/Vanguard, a think tank that sponsored meetings Minsky participated in, says, “He saw the developments of the last few years as steps in the wrong direction. Google and Facebook are exploiting their vast data sets, using deep learning. But Minsky saw this as achieving short-term gains at the expense of solving the real machine-intelligence problem.”

Minsky wasn’t naive about what could happen if machines achieved the capacity for humanlike learning, but he believed that humans would be able to deal with the challenges this would create. He worked on his magnificent obsession, not in tweets, or three-year research-grant sprints, or IPO-funding cycles, but over decades—by thinking one rich thought at a time.

This article appears in the March 2016 print issue as “Marvin Minsky and the Pursuit of Machine Understanding.”