Saturday, July 23, 2011

Robot Emotional Intelligence

The emotionally clueless computer or robot has been a science fiction trope for quite some time. AI entities are shown to have incredible math skills, but have serious trouble understanding, predicting, and interacting with humans.

I had always assumed that this was justified. After all, humans do not have brain circuits specifically for things like math or logic, we have to literally rewire our brains (with practice and study) to do basic math problems. But we are inherently social animals; with brain circuits specifically meant for things like reading the facial cues of other people. Replicating those circuits in a computer should be a very hard problem, like machine vision or speech recognition.

It turns out that it is not. With relatively little effort, a small team of scientists has made a device that can read human facial expressions better than humans can. Right now, the device takes the form of a pair of glasses that gives information to a person wearing them. But it would be trivial to add this functionality to any computer that interacts with people, and the face-reading can probably be improved even more.

What this means is that by the time we get computers or robots that can talk with us, they will have an emotional intelligence greater than most humans. Their powers of insight and perception will be as far above that of people as their math skills. They will know from your expression if you are thinking, agreeing, concentrating, interested, confused, or disagreeing, and will adjust their behavior accordingly.

Combine this with the fact that it is very easy to get people to emotionally bond with robots, and the fact that the robots will be programmed to obey you, and most people will probably prefer the company of a well-programmed robot to the company of a human. Household robots, even the first generation of them, will not act like cold robotic automatons. The marginal cost of adding things like emotion reading and very expressive faces will probably be low, compared to the cost of something that can understand speech and interact with a complex environment.

The first generation of robots will probably be more like Forrest Gump than Data. They will be dumb, and they will admit to being dumb, but the people they interact with will love them because they are emotionally responsive and friendly and never get upset or flustered.

The more I think about this, the more it scares me. If normal humans no longer have a comparative advantage at interacting with other humans, then what role will they play in a future economy with lots of robots? At the moment, the only thing computers have serious trouble with is metacognition and game-theoretic stuff like poker. Computers are horribly bad at poker and any skill similar to poker, and there is no sign that this will change any time soon. This is good news for our ability to control and manipulate them, but bad news for people in a social system that does not handle unemployment well.

No comments: