IBM Explores How Affective Computing Can Benefit Society
If only our computers could tell us, “Hey, you’re tired. Go home and take a nap.” Or, “That email you’re about to send … you might want to rethink that.” Or, “We’re sorry for the mistake with your order. How can we make it right?” This insightful and empathetic computing—otherwise known as “affective computing”—isn’t all that far out and, in some cases, already exists.
“When you ask whether computers are going to be better than humans at detecting emotion, it’s really dependent on the individual emotional situations.” —David Konopnicki, manager of information retrieval at IBM Research–Haifa
In fact, according to David Konopnicki, manager of information retrieval at IBM Research–Haifa and leader of IBM research into affective computing, affective computing can be quite beneficial, whether it’s helping people during customer service calls, at school or while driving their cars.
Q. What is affective computing?
A: Affective computing has several definitions, but I think the best is “computing that’s influenced by human emotion.” We want the computer to detect what human emotions are, take human emotions into account in computation and sometimes simulate human emotion as part of the computer’s output.
Ideally, we would like the computer to reply in a natural way, as if it has emotions. Think of a conversational system where I can chat with an agent—which is a computer—to describe my problem. When the computer discovers, for example, that a mistake was made, it should be able to apologize, say thank you and sometimes give you time before answering so you can complete your sentences. The entire system should be empathetic to the user.
This is an example of a computer used for customer service, but you can easily imagine other cases. For example, if the computer is used for educational purposes, it could encourage users to help them overcome difficulties.
Q. When you’re interacting with a system such as this, how does it know how to respond?
A. Most existing applications for affective computing are based on capturing what the emotion is, and recording and reporting that emotion. An example would be when you speak with a human agent in a customer service center and the call is being recorded. A computer can analyze the recording to provide a score as to whether, for instance, people are getting angry. Most of the existing applications are centered around reporting on emotions.
Our research goes one step further. It works after a system has detected an emotion and helps optimize the response of the system to this emotion. For example, if we talk about customer service, we’re looking at two things: We’ll analyze a real discussion that has taken place, what the answers of the human service representative were and whether those interactions went well. Based on that, we can teach the computer how to behave. That’s one set of data we have.
The other set of data are books that exist from training centers. We try to take this body of knowledge and teach the computer how to use that. Say you were working in a call-in center in a department store. You would learn how to behave and how to respond most effectively to the customer you’re dealing with. This is the same type of thing we’re doing. We’re learning from, first, examples of actual discussions and, second, the experience of existing centers we can find in groups.
comments powered by