Marvin Minsky at TEDMED3: what we need from artificial intelligence

In the time I set aside today to keep working on my biography of Richard Saul Wurman, I pulled some videos off of a CDROM that was shipped to attendees of RSW’s third TEDMED conference, in 2003. The image and sound quality are what you’d expect from analog-to-digital conversion circa 2003. But the content is as good as it gets. Most of these videos have never been published, beyond the private edition CDROM package that attendees received after the conference.

Of particular interest to contemporary IA (as well as AI) Theory is this session video featuring MIT AI Lab co-founder and Japan Prize winner Marvin Minsky. Here’s just a few moments of what he said from the YouTube transcript. Full video of his session is embedded below.

What do we know from doing project work with clients that we could bring to bear on the wicked problems facing the world today?

What we've got to do in our machine is recognize different kinds of problems, and then turn on different ways to think.

There's nothing wrong with neural networks: they can solve certain kinds of problems but not others.

There's nothing wrong with logic: it can solve mathematical problems, but not real life problems.

There's nothing wrong with evolution, where you mutate a program and reward it if it solves the problems. Evolution is good for making germs and mice and rats and once in once in a million years it makes something like a person, but that turns out not to work, and they shortly disappear.

And so evolutionary programs are dramatic, but they're certainly not the answer, and they're very slow.

But you see what I'm saying. is that we need a machine that has ways to recognize obstacles, and it has different ways to think, and that's what's never been done, and that's what we're starting a new project to do. To take all of the different things that solve a problem here and there, but can't have the generality of the three-year-old child, and organize them in one machine.

And believe it or not, this machine is going to look like the architecture of Sigmund Freud