Physicist Offers New Answer for Smart Machines: Think Smaller

Creating computers that can think has long been a Holy Grail of computer science. Usually that means building bigger machines with more processing power, such as IBM’s Deep Blue, the mainframe that defeated world chess champion Garry Kasparov.

But physicist Neil Gershenfeld prefers the opposite approach. The faculty member at the Massachusetts Institute of Technology’s Media Laboratory believes the solution is to make computers very small and sprinkle them throughout the environment.

In his new book, “When Things Start to Think,” Gershenfeld describes coffee makers that know when to brew a fresh pot, cellos that can make music even when the bow isn’t touching the strings and paper that can be reused because it is covered with programmable electronic ink.

onetime philosophy student, Gershenfeld joined the Media Lab in 1991 after studying physics at Swarthmore College and Cornell University and picking up some engineering at Bell Labs. He leads the Media Lab’s Physics and Media group and is co-director of the Things That Think research consortium.

Gershenfeld recently spoke with Cutting Edge about his vision of computers that are designed to accommodate people instead of the other way around.

*

Question: The title of your book is “When Things Start to Think.” What do you mean by “think”?

Answer: I mean thinking in a very general and simple sense. I’m not interested in artificial intelligence. I’m more interested in the intelligence of people than the intelligence of machines. I’m interested in building simple capabilities into objects around us that will help us solve problems.

Q: Can you give me an example?

A: We’re developing little consumer NMRs [nuclear magnetic resonance sensors] that can describe chemistry. You can put one in your refrigerator and it can analyze your milk and say, “Your milk is turning bad.” Then as you walk by, your fridge can tell your shoes, “You’re low on milk and it’s turning kind of funny.” Later, when you’re walking by the store, the store can broadcast, “We’ve got milk on sale.” At that point, your shoes can send a message to your glasses saying, “By the way, you’re running out of milk and you might want to go in and get some.”

Here’s one more example: computers that recognize emotional states. If you have lots of sensor data on body temperature and things like that, you can discover if somebody’s stressed or happy or sad or confused. That has huge implications. If you’re stressed, information given to you quickly helps you relax. If you’re relaxed, information given to you quickly makes you stressed. Something as simple as the pace of information delivery should depend on your emotional state. What you need is some sensors and some inference.

Q: Is there a lot of new technology required to get that kind of data?

A: There’s some, but not a lot. You need better sensors and you also need better mathematics of inference. But that’s not a breakthrough in artificial intelligence. It’s just a bunch of little pieces connected together, getting the right data in the right place.

Computers are essentially blind, deaf and dumb. They don’t have the life experience of a 2-year-old child, so they don’t have a hope of having the simple sort of common-sense understanding of the world that we have. Making machines behave more sensibly around us comes not from a penetrating brilliant insight into what is consciousness, but by giving them access to information about the world.

Q: Is this beyond the limits of traditional computers, the kind that sit on our desktops?

A: Everywhere we look, engineering technology is approaching physical limits. In about two decades, transistors will be down to one atom big, and clocks are already approaching the speed-of-light limit on sending signals. So to progress in computation, you have to think much more deeply not about arbitrary computers, but about how nature can manipulate information. We have to be smarter in taking advantage of what nature already knows how to do.

Q: You’re trained as a physicist. How did you come to be reinventing the notion of a computer?

A: It started after I developed a new sensor to measure how Yo-Yo Ma plays the cello. I thought it was just an amusing hack. But then many other people got interested in using it to solve many other problems. That’s when I began to realize that billions of dollars go into developing smart processors that go into stupid boxes that haven’t changed much since the earliest days of computing.

We learned how to make furniture that can track the position and orientation of a hand in 3-D space around it. Right now, if you want to use a computer, you’ve got to reach out and find its mouse and drag it around. With this, you can just have a nice piece of furniture and the furniture finds you in 2-D or 3-D.

Q: How would somebody use it?

A: It can be used in any way. The computer starts by knowing what you’re doing, and it’s up to you and the furniture to decide what that means. You could become a mouse, you could type, you could point, you could gesture. There’s an infinity of ways to do it.

Q: This sounds like a much more natural interface between computers and people. What’s the catch?

A: What makes that possible is bringing more technology closer to us. I’m asking you to put a computer in your shoe. We’re in an awkward evolutionary stage where we need more of the stuff closer so that it goes away.

Q: When computers are embedded in everyday objects, we’ll all generate a lot more personal data. Do you have any concerns about privacy?

A: It’s a social, not a technological, question. For example, if your shoe has the computing power of an early Cray, if you walk into a store and leave your identity on, somebody comes breezing up and says: “Welcome Mr. Gershenfeld. Here’s the shirt you’re looking for. It’s just your size.” I get great service, and not only that, but they’ll give me a discount because I’ve given them all this great demographic information. Then I go home and get all their junk mail.

Or, I could leave my identity off. I buy the shirt and they never learn a thing about me and it costs me more and I get worse service--and I won’t get junk mail.

It’s not up to me as a technologist to decide how that trade-off should be made. That’s really a social decision. But it’s up to me as a technologist to enable it.

Q: Isn’t it unnatural for people to be so surrounded by so many computers?

A: We do become kind of dependent on computers. But the World Wide Web is designed around a person sitting by themselves typing at a desk. That is a small subset of human experience and leads to the sad situation of somebody at lunch browsing the Web alone. If you think about computing, that’s what it forces you to do--either you use the computer or you live the rest of your life.

By making them go away--by losing the notion of a distinguishable computer--the physical environment becomes your interface to information. Then you can live your life and interact with people. It allows you to be human but not chained to the computer.

Times staff writer Karen Kaplan can be reached at karen.kaplan@latimes.com.