“When you have a complex set of goals that a system needs to achieve,” says Rosalind Picard, “and there is limited time, and you have limited resources, and the inputs are unpredictable—with the model constantly changing, so you can’t just model the space once and solve the problem—then you need regulatory mechanisms to help the computer decide what to do next.” Moreover, she adds, a computer needs to know whether “if it is doing one thing, it should drop that and try something else, because there are other things more important.” In short, it needs judgment, something to guide its behavior. And human judgment is informed by emotion, says Picard, who heads MIT’s Affective Computing Group. “If we want a computer to have those abilities, we are going to have to effectively give it an emotional system.”

In his thesis, “Natural Intelligence in Artificial Creatures,” Swedish computer scientist Christian Balkenius makes the point that when Pavlov did his famous study—in which dogs salivated at the sound of a dinner bell—he didn’t test dogs that weren’t hungry. Since a dog won’t salivate if it isn’t hungry, Balkenius points out, Pavlov’s results are compromised. He didn’t control for motivation.

Until recently, Picard and her group focused on making computers more responsive to the feelings of humans, but now she’s calling for a major new initiative. One of her inspirations for that is the work of Antonio Damasio, whose book Descartes’ Error: Emotion, Reason, and the Human Brain describes his experience treating emotionally impaired patients. Those patients, who seem strangely unemotional, suffered from a marred rational decision-making that Picard found oddly similar to the brittle expert systems of classical AI.

Discussions at MIT rarely go far without reference to science fiction. Picard alludes to the popular television show Star Trek: The Next Generation, in which the android Commander Data can turn his emotions on or off at will. Says Picard, “He clicks off his emotion chip and he functions better than anybody on the ship, right? And then he clicks it on only when he wants to write poetry or be more human. That’s wrong. If we click off our emotion ‘chip,’ so to speak, we become like Damasio’s emotion-impaired patients—who instead of being more rational actually have trouble with day-to-day decisions such as when to schedule an appointment. They invest in the wrong things and lose all their money, and keep doing it. They behave irrationally.

“What we don’t see,” says Picard, “is that emotions inside us are regulating and biasing the rational decision making in everything we do, coloring all of our information…If we really clicked off those emotions, we would have some serious problems.” Through her reading of Damasio and the work of others in neuroscience, cognitive science, and psychology, Picard came to the view that learning, flexibility, and creativity in thinking can be seriously hampered by a lack of emotion. Looking back, she recalls, “Basically what I said when reading it was, ‘Gosh, this is just what computers do. They malfunction in the same ways as these people whose emotions are disconnected.’”

That’s one reason why “most of these machines make terrible assistants,” she says. “They are fine for use as a hammer or direct communication tool, but when you want to have them do some autonomous task for you, they’re terrible…We’ve erred on the side of building these neurologically impaired patients.”

Tags: affective computing, Antonio Damasio, christian balkenius, Commander Data, Pavlov, robotics, Rosalind Picard

Email this to a friend | Add to del.icio.us | Add to digg |
Linkology by Antony (all links are unofficial) | Rights Info

Rating: 0, from 0 votes

Posted on May 19th, 2006 @ 3:00 pm