Rosalind Picard

I want to understand things that interest me, and build things that understand and interest people.

I'd like to figure out how people perceive affect: What is it that tells you when somebody is pleased or displeased with what you did? Then, once we understand this, how can we encode this ability in a machine, so it too can see what pleases or displeases?

If the computer, robot, PDA, or future earring senses that it annoyed you, what should it do? When should it be silent or when should it interrupt you, perhaps even to apologize? How can it make such complex emotion-savvy decisions given limited inputs, limited intelligence, and limited processing resources?

Neuroscientists have provided increasing evidence that the human brain regularly makes complex intelligent decisions quickly without logically considering all the possibilities (which would take too long). The brain apparently does this by using the emotion system to regulate information—to shift attention, to bias what is retrieved and to influence the memory search and decision making process in largely beneficial ways.

This is surprising because we usually think of emotion as making people irrational. But that is only half the story: emotion also appears to be essential for flexible, intelligent, rational thinking.

I want to build systems that are truly flexible and intelligent when it comes to interacting with you. These systems should be judged not just by what they can help you accomplish and how fast, but also by how much you enjoy working with them.

These goals drive my desire to figure out how to quantify states like frustration, stress, and pleasure. I develop tools of engineering to measure affective information from the human body. Lately I have enjoyed learning about fascinating biological processes such as the way small positive emotions lead to more creative thinking, and the ways molecules of emotion carry signals between many organs—showing body-wide information processing. I have been learning about the brain in your gut, with its 100 million neurons and ability to learn, about bodily homeostasis, and quantifying how your heart's beating patterns behave when you are under stress.

An affectively intelligent system has many uses: for instance, as a companion in helping a child better deal with frustration when trying to learn math. The system could see when the child is frustrated, bored, interested, or pleased, help the child to better handle these emotions that naturally occur in learning, and help the child to be a better learner. This involves a host of technical problems, where I benefit from collaboration Lab-wide: from multi-modal affect recognition, to causal reasoning, to synthesizing conversational or other helpful responses. It also involves real-time learning, decision-making, and action selection by the machine from a complex set of inputs, discerning the success or failure of what it just tried, and regulating and adapting its own resources to improve its future attempts.

Or, a wearable affective system might help a smoker who wants to quit identify the stresses that provoke her to relapse (stress has been shown to be the biggest factor), a step toward regaining control. Alternatively, an affective computer might help artists develop new ways to utilize emotional information in musical expression.

Think about it: what could you do with a machine that, literally, reads your heart?