Takeo Kanade is the 2008 recipient of the Bower Award and Prize for Achievement in Science from Philadelphia's Franklin Institute. The award, which includes a gold medal and a cash prize of $250,000, honors Kanade for his "visionary leadership and scientific accomplishments in the design of perceptual robotic algorithms and systems that function in the physical world." He recently spoke with Link Managing Editor Jason Togyer about his work and offered advice for aspiring technologists.
Quality of Life Technology
The most exciting part of my current research is being done in the Quality of Life Technology Center. Imagine you could create a system that would take patients from bed in the morning, feed them, dress them, turn on the TV, and handle their hygiene needs. It would be an enormously helpful machine--and most people wouldn't want to live with a machine like that.
Most people want to be able to take care of themselves as much as possible. What quality-of-life systems should do is the difference between what you want to do and what you are physically able to do--no more, no less. They should increase the amount of human involvement, not reduce it.
In the past, roboticists wanted to develop autonomous systems--robots able to act without human help. They had the implicit goal of reducing human input. I'm very interested in developing technology that can think about human needs non-invasively.
First-Person vs. Third-Person
Computers should be able to think intelligently about what people need, but knowing that from an outside-in viewpoint is almost impossible. Suppose we want to know what you're doing right now. We could put a camera in the room and observe you, but that's a third-person viewpoint, and it's pretty tough to get useful information, because the camera has to have the right viewing angle at all times. Take the way you're sitting--you may be writing or sleeping; depending on where the camera is, it might be hard to tell, because you have the same pose.
What if you wore a little camera? Now the computer is looking at things from your point of view and it's easy to tell what you're doing. I call it "inside-out, first-person vision." Most of the time, your gaze is focused on the most important thing that you need. After all, when we watch movies, from time to time we see things from the character's point of view, and we can easily understand what he's about to do.
Balancing Help and Privacy
We're very aware of privacy issues. That's one of the problems with putting a camera in a room--you feel like you're being watched all the time. But a camera that looks at things from your point of view is more like an assistant. For instance, some people put on gloves before they drive somewhere; they feel that helps them better hold the steering wheel. What if you put on a small camera instead, so that it was actually sharing the complete view that you have? The most intelligent machine in the car is you. Why shouldn't the computer be helped by you?
In factories, one of the newer trends is not to completely have a robot do assembly, but to have a robot as a helper to the human assembler. A robot can be a very, very effective helper if it provides tools or parts, or inspects the assembly while the component is being assembled. Let's say you're wearing a camera. Now the computer is watching from your perspective--you know what you're assembling, and the computer knows what you're assembling. Maybe it can spot something that you've forgotten.
We may have the wrong concept of "new." "New" is not itself a virtue. "Usefulness" is a virtue. Doing "new" things is almost trivially easy--there are probably many more new things done every day than useful things. What's important to a concept is, does it do anything useful for us? Does it solve a problem meaningfully?
Then, when you come up with an idea, you have to sharpen it, and that takes work. The only way to test a concept is to try it. Students will say in a presentation, "Parameter A is best when it's a small number near zero, so I set it to 0.8." I'll ask, "You say A is better small, so then why not go all the way to zero, and see what happens?" You have to extend your ideas, sometimes to extremes.
Too often, people don't go to these extremes, or even do a "thought experiment" to figure out what would happen at the extremes. They have to actually follow these things through, either by thinking about it, or by actually doing it. One of my favorite sayings is, "Think like an amateur. Execute like an expert." Most people do the other thing--they think they're an expert, and execute like an amateur! That's a recipe for failure.