Rod Brooks has two small tasks in mind as he leaves the MIT academic bureaucracy he’s known for the last decade and dives back into hands-on science. “One is to restructure the world’s labor markets, the other is to discover the meaning of life.”
Last week, when I reported on Brooks stepping down as director of MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) to return to the lab bench, I promised to follow up with more details. Here’s that report, direct from the computer-science visionary’s cavernous robot-and-tool-filled new office in the “nose” of MIT’s Stata Center. It’s a quiet place now that the school year has ended. “All my students have graduated, so I’m starting with a fresh slate,” Brooks says. The renowned robotics expert hasn’t taken on new students yet because he is still defining the next steps in pursuing his two quests. But, of course, he does have some ideas.
Task number one, reshaping the world’s labor markets, is really a robotics problem. The goal is to achieve the same sort of transformation in robotics as personal computers made possible in computing. Or, as Brooks puts it, “I’m trying to do for manual workers what PCs did for information workers, i.e., let ordinary manual workers become their own information engineer and increase their own productivity.”
Back around 1980, he explains, computers were back-room behemoths. “Office workers ‘used’ computers by poring through fan-folded printouts of sales and stock reports. If they wanted a different analysis than what was delivered they had to defer to systems analysts, programmers, punch-card operators, and computer operators,” he says. “They didn’t have direct control.” But then came the PC, with software applications such as spreadsheets that automated simple tasks. Office workers, says Brooks, “became their own automation engineers, and they increased their own productivity.”
Robots today are essentially where computers were nearly 30 years ago, Brooks says. Industrial robots do dangerous tasks in factories and other sites, but to program them takes months. If you want to change the process, it means relying on professional automation engineers. Robots and human workers typically don’t mix.
That’s what Brooks wants to change. “My use case is a bakery in Cambridge, three or four people. They go on the web, click, click, click, they order a robot for 2000 bucks.” Then, when the robot arrives, “they don’t read a manual, because no one reads a manual.” Yet, two hours later, using a combination of natural language and physical demonstration of the job to be done, the bakery workers have programmed their new “hire” to perform simple tasks that they used to do manually. “They won’t even think of it as programming,” envisions Brooks, who adds that the robot doesn’t have to perform jobs flawlessly since the workers would be there to fine-tune things.
Thus begins the revolution. Ultimately, Brooks hopes, by empowering U.S. workers and increasing their productivity, robots will make U.S.-based manufacturing more attractive than outsourcing in many cases. “This can be a force for evolving in a different direction from the way we have been heading the last 50 years. That’s my plan.”
Brooks’ second plan, exploring the mathematics of living systems to create a revolution in electronics, might be even more ambitious. What, he wonders, makes living systems so adaptable? How do they self-organize and maintain themselves? This overlaps a bit with Artificial Life, a subset of computer science in which people create algorithms to try to mimic Darwinian evolution. But where those efforts have fallen short, Brooks says, is in capturing the flexibility of living things. For example, he notes that scientists can cut out the brains of flatworms and put them in backward, and the brains still function. Similarly, a human liver can be transplanted into another person, with the liver then growing and adapting in the new host.
Programming to date doesn’t have anything close to that adaptability. Can we rethink computer science to make algorithms far more adaptable? Brooks asks. If so, a person might be able to take a 1992 Apple Macintosh application and run it on a 2007 PC, and vice versa. “In computer land that doesn’t work. They’re so brittle,” says Brooks. “Well, maybe if we can figure out how living systems do it, then perhaps we can make our software not so brittle.”
