Professor Stacy Marsella discusses "Computationally Modeling Human Emotion," a Contributed Article in the December 2014 CACM (cacm.acm.org/magazines/2014/12/180787)
00:00 [VOICEOVER] There's a long history of computers faking emotional intelligence. But in recent years, they've truly become adept at judging our emotional state from such signals as posture, tone, and facial expression. We looked at this field of "affective computing" last January.
00:22 [VOICEOVER] Now researchers are trying to give computers a complete model of human emotion -- the events that cause it, the ways we interpret those events, and what we do in response. Such a model could make them truly empathetic -- and better able to moderate their decisions with the same emotional intelligence that we enjoy as human beings.
00:45 [VOICEOVER] Join me at Northeastern University in Boston as I talk with Stacy Marsella about Computationally Modeling Human Emotion.
00:55 [Intro graphics/music]
01:05 [VOICEOVER] Computers excel at calculating probabilities. But they aren't nearly as good at making decisions. Dr. Marsella, working with Jonathan Gratch, believes that imbuing them with emotional sense will help.
01:21 [DR. MARSELLA] People are essentially serial information processors, they have multiple goals, and events come in that force people to deal with some immediate, unexpected event. And how do you actually deal with that? Well, you need some sort of interruption mechanism. And emotion serves as this interruption mechanism. So people in AI and robotics have looked at these emotions, the function emotion serves in humans and said, well, does that tell us something about how we can then incorporate similar kind of functionality in our systems to make them more robust.
01:56 [VOICEOVER]An unexpected event provided insight into how these lightning-fast processes work.
02:02 [DR. MARSELLA][CLIP4.MOV-00:04]We were shooting some scenes of some actors, and all of a sudden a bird flew in the window. And so we got a very interesting moment-by-moment video of how people were reacting to this bird flying in the window: Looking at first the surprise, and then fear, then anger, and these aggressive stances where they're trying to whack the bird with an umbrella, and then finally all the people in the room organizing around saving the bird. It took about two and a half seconds, and in the paper we sort of take people frame by frame, showing the emotional reactions and how that's actually driving behavior.
02:40 [VOICEOVER] But what actually happened in those two and a half seconds? There have been many theories throughout the years. But to teach computers about human emotions, Dr. Marsella prefers something called Appraisal theory. Appraisal theory considers not only the events that elicit emotion, but also one's inner life.
03:00 [DR. MARSELLA] Is this event desirable? Was it expected? Do I feel I have control over it? Who's blameworthy? The pattern of assessments leads to certain kinds of emotional states.
03:14 [VOICEOVER] Many computer models have used appraisal theory to try to understand human emotions. Dr. Marsella and his colleague Jon Gratch created one called "EMA", based on the work of psychologist Richard Lazarus. EMA assigns values to each of the relevant features. What comes out is a coping mechanism, such distancing, wishful thinking, or resignation.
03:38 [DR. MARSELLA] The process of coping basically involves thinking about either, "How do I change the world so I achieve a better ecological fit with the world?", or "How do I change myself to achieve a better ecological fit with the world?"
03:51 [VOICEOVER] The resolution of that process affects the situation... and the cycle begins anew.
03:58 [VOICEOVER] Which brings us back to the bird. The actor communicated surprise, fear, anger, aggression, and cooperation. Those emotions affected others around her and helped them reach a resolution.
04:15 [VOICEOVER] If computers can understand the processes behind these emotions, they might make decisions that make more sense in the real world. And they may even teach us to understand ourselves better.
04:28 [VOICEOVER] Find out more in this month's Communications of the ACM, in the contributed article, "Computationally Modeling Human Emotion".
04:36 [Outro and credits]