Frédéric Kaplan: A.I. Auteur

Dream Jobs 2007

Loan Nguyen

I, Robot:

Frédéric Kaplan’s machines make people think.

In a sly sort of way, Frédéric Kaplan is a subversive. The lanky, soft-spoken French robotics expert, who joined the staff of Switzerland’s prestigious École Polytechnique Fédérale de Lausanne (EPFL) last fall, certainly doesn’t look or act the part. But make no mistake, this man wants to mess with your mind. And he intends to have a lot of fun doing it.

When a visitor recently sat down with Kaplan in the research lab where he oversees a multinational team of think-outside-the-box engineers, it was at a table that exemplifies his latest adventure in intelligent objects: interactive furniture.

”We call it a noise-sensitive table,” he says, running his hand over a plate of frosted glass set in a wooden frame. As he speaks, waves of color generated by embedded light-emitting diodes ripple across the surface, as if a pebble had fallen from his lips into a phosphorescent pond. The table is equipped with hidden microphones that measure the speaker’s voice’s timbre, volume, location, duration, and even—in a future prototype—stress levels. The display of lights conveys a nonverbal, real-time reflection of what is being said.

Working with industrial designers, Kaplan and his team are creating an entire menagerie of such nonintrusive, smart furniture. But their goal isn’t to be merely decorative. They also aim to toy with the way people interact with each other and with technology. After a few minutes, a little blue light goes on just in front of Kaplan, signaling that he has monopolized the dialogue thus far.

”Conversation is very important in everyday life. It is arguably what holds society together,” he notes. ”We wanted to find out what happens if you measure things and show them to people as they are speaking. Does it change the dynamic of the conversation in specific ways? It is potentially something very powerful.”

31: number of Facial Motors in ”Albert Hubo,” a Korean humanoid robot

Kaplan’s team may soon get to test their ideas in a living laboratory, when EPFL completes construction of its futuristic new learning center. The size of three football fields, the learning center will house a library, plus study and recreational facilities for the school’s 6000 students. And Kaplan hopes to furnish it with his experimental creations.

The 32-year-old wasn’t always so sure of what he wanted to do. Even as a child, he had been torn between the worlds of machines and of living things. Although he earned an engineering degree from the École Nationale Supérieure des Télécommunications, in Paris, he says, ”I hesitated a long time between becoming a computer engineer and a biologist.” He even worked briefly at a molecular biology laboratory in Germany before returning to Paris to pursue a Ph.D. in artificial intelligence at the Université Pierre et Marie Curie.

What drew him to A.I. wasn’t figuring out how robots could mimic natural intelligence but rather how biology could learn from smart machines. While Kaplan was still in graduate school, Luc Steels, the director of Sony’s newly opened computer science laboratory in Paris, offered him a junior research position. It seemed the perfect opportunity, and he eagerly accepted.

In the freewheeling atmosphere at Sony, Kaplan helped design the ”brain” of AIBO, the company’s eerily endearing canine robot, and he continued to develop it over nearly a decade. One of his biggest achievements was to program AIBO to get ”bored,” an experiment designed to test the limits of open-ended learning—the holy grail of artificial intelligence. The first step was to write a neural ­network–based program that predicts the effect of a given action—turning one’s head, kicking a ball—and then compares the prediction with the actual outcome, a process well known in machine learning. Kaplan and his colleagues then created a second ”metapredictor” that tells the robot if an action will be easy or difficult and steers it away from either extreme. ”It is motivated by experiencing progress, which we defined as the reduction of error over time,” says Kaplan.

As AIBO learned, it outgrew the limited range of activities available to it. It was ready to play, in other words, but had no suitable toys. Someone mentioned to Kaplan that a group of talented design students at EPFL might be able to help.

”They came up with ideas in a couple of days and finished prototypes within a week,” recalls Kaplan. ”I was amazed.” Soon, AIBO was grappling with a miniature pushcart, a periscopelike mirror, a waterproof swimsuit—and continuing to learn as it played.

Kaplan found himself spending more and more time at EPFL, and last October, he was invited by Pierre Dillenbourg, head of EPFL’s Center for Research and Support of Training and Its Technologies, to oversee a new team there working on interactive furniture. In his new Swiss digs, he ­continues to let his curiosity drive his research. Lately, he’s been delving into epigenetic robotics, a field in which developmental psychologists, roboticists, and computer scientists see what they can learn from one another.

Kaplan continues to be fascinated by what robots and machines can tell humans about themselves. Westerners tend to fear robots, he notes, in stark contrast to the Japanese, who embrace a future in which the machines will be ubiquitous. A good way to challenge such prejudices, he says, is to surreptitiously slip intelligent technologies into people’s surroundings, through objects like his interactive furniture, and then see how they respond.

”I want to unsettle people,” Kaplan says with a mischievous grin. ”I am exploring machines that are not necessarily like servants but more like tricksters, capable of surprising us and making us think and behave differently.

Related Stories