Prime Time Replay:

Rosalind Picard
author of Affective Computing, on creating
computers with emotions


MsgId: *brain_storm(1)
Date: Sun Nov 16 21:59:38 PST 1997
From: Rob_Killheffer At: 205.198.117.104

Good evening, everyone, and welcome to another edition of Brainstorms. Tonight my guest is Rosalind Picard of MIT's renowned Media Lab. In her new book, "AFFECTIVE COMPUTING," she argues that for computers to become intelligent, and to be able to perform some of the tasks we hope they'll perform, they need to be able to recognize, respond to, and even experience their own emotions. Welcome, Dr. Picard!
MsgId: *brain_storm(3)
Date: Sun Nov 16 22:01:15 PST 1997
From: Rosalind_Picard At: 18.85.12.147

Greetings, Rob. It's great to be here. (Home, and online)
MsgId: *brain_storm(4)
Date: Sun Nov 16 22:02:24 PST 1997
From: Rob_Killheffer At: 205.198.117.104

Thanks for coming. I hope the interface will hold up! For many people, the idea of an emotional computer will seem a contradiction in terms. Could you explain a little about why you think emotions are -- far from antithetical to intelligence -- actually vital for reasoning and perception?
MsgId: *brain_storm(6)
Date: Sun Nov 16 22:08:51 PST 1997
From: Rosalind_Picard At: 18.85.12.147

Sure. There is mounting evidence from neuroscience and from psychology pointing to the role of emotion in intelligence. For example, there are findings (Damasio et al) of people who essentially do not have enough emotion, due to brain damage, and they are far from the Spock image we would expect. For example, they might have trouble making a simple apt, and completing other decisions that usually come effortlessly to people. They are not merely indecisive, but they seem to lack the emotional values that healthy people associate with memories, and use for making decisions.

Apologies for my delays (I'm not one of the patients I just refered to, but my Netscape just died and I just discovered that one of my mindless emacs editing commands also clobbers the chat room.)

MsgId: *brain_storm(9)
Date: Sun Nov 16 22:17:49 PST 1997
From: Rob_Killheffer At: 205.198.117.104

You've spent a lot of time working on computer perception, especially vision, and you say that emotion plays a key role even in such a basic function as seeing. How does that work?
MsgId: *brain_storm(10)
Date: Sun Nov 16 22:19:20 PST 1997
From: Rosalind_Picard At: 18.85.12.147

If people look at neutral stimuli, say faces with no expressions, and one group is put into a bad mood, then that group will see the faces as somewhat negative in their expressions; similarly, a group in a good mood will perceive the faces as somewhat positive. There is also evidence that some aspects of vision happen in the emotional parts of the brain (amygdala, etc.) BEFORE they happen in the visual cortex. For example, you think you see a snake and you jump out of they way. Then you realize it was just a stick.
MsgId: *brain_storm(12)
Date: Sun Nov 16 22:23:44 PST 1997
From: Rob_Killheffer At: 205.198.117.104

We've all had that kind of experience. What it adds up to is an appreciation of the fact that emotion isn't some meaningless add-on in our brains, at best a luxury and at worst an obstacle to clear thinking, but actually crucial to it. And, from a "hardware" perspective, a recognition that the various areas of the brain are more interconnected and interdependent than we usually think -- things don't just happen in the visual or auditory cortex, but go through the limbic system and other areas as well. But that brings us to the more practical issue: if computers need emotions to achieve what we call intelligence, how can we give it to them? Is it possible with the kind of computers we have today, or will it require a radically new design (based, for instance, more closely on the brain itself)?
MsgId: *brain_storm(14)
Date: Sun Nov 16 22:31:05 PST 1997
From: Rosalind_Picard At: 18.85.12.147

Yes, about the limbic system (those systems in the brain concerned primarily with emotion, attention, and memory). Your question about the architecture is much more difficult. It is the same question Rod Brook's group and I discussed for nearly an hour at one of their "zoo" meetings. I used to be a computer architect, designing VLSI chips for various signal processing tasks, so I think about architecture on many levels, probably more than folks here want to know about! Of course, we know that present computers can "simulate" anything, so long as time restrictions are loose. However, emotion definitely does involve timing; for example, certain emotions that are sometimes called "primary" work in an especially rapid way, effectively "hijacking" other processes.

For these emotions, such as the fear that made you jump out of the way of a stick, the architecture has to have levels of interrupts, which most current machines have, and which might simply need to be tweaked. However, there are certain ways in which emotion influences memory --- storage and retrieval --- that are not as obviously easy to implement in present machines. For these, I would encourage some radical architectural rethinking and probable redesign.

MsgId: *brain_storm(17)
Date: Sun Nov 16 22:39:50 PST 1997
From: Rob_Killheffer At: 205.198.117.104

Assuming we figure out how to develop "affective" computers, what do you think will be the benefits? What tasks will affective computers be good for, and how will they change the way we incorporate computers into our lives?
MsgId: *brain_storm(18)
Date: Sun Nov 16 22:45:37 PST 1997
From: Rosalind_Picard At: 18.85.12.147

One question at a time . . . benefits. The first that comes to mind is systems that become aware of when they irritate a lot of people, and use this information to either improve their own functioning directly, or use it to inform their designers so that their designers can improve the system. People who get along well with others do this all the time; they notice when they've upset someone, and modify their behavior in accord. And, they might repeat behavior that is pleasing. Imagine if people didn't do this, (those who interact with autistics may be most understanding.) It is "very frustrating" to interact with someone or something that takes no note of its effect on you.

One of my RA's, Jonathan Klein, is investigating how systems might respond to user frustration in a way that helps reduce that frustration. I'm not sure people realize how much stress computers cause today, and how much loss that translates into . . . not just in productivity, but in creativity, health, and opportunity costs. Shall I go on with the other questions yet? Feel free to interrupt before.

MsgId: *brain_storm(20)
Date: Sun Nov 16 22:49:53 PST 1997
From: Rob_Killheffer At: 205.198.117.104

No no, go on to the other Qs. What tasks will affective computers be especially good at? And how will they change the ways we incorporate computers into our lives?
MsgId: *brain_storm(21)
Date: Sun Nov 16 22:53:31 PST 1997
From: Rosalind_Picard At: 18.85.12.147

They will (eventually, after the usually juvenile development errors) be much better at interacting with people and adapting to our preferences. Currently, the burden is on us to figure out how to adapt the computers to do what we want (features, features, features, how do we set them all?) Learning algorithms, on the part of the computer, are beginning to help them adapt to us, but they miss most of our communication, which comes not from what we type so much as from how we type, and how we furrow our brow in confusion, frown, or smile, and so forth. Often "how" something is said is much more important than what is said, and this is what they should be able to pick up on.
MsgId: *brain_storm(22)
Date: Sun Nov 16 23:05:46 PST 1997
From: Rob_Killheffer At: 205.198.117.104

Roz, are you there? . . . If you're having trouble sending a message, it may be too long. Try sending less at once, break it into parts.
MsgId: *brain_storm(24)
Date: Sun Nov 16 23:09:05 PST 1997
From: Rosalind_Picard At: 18.85.12.147

Hi . . . sent you email; last Netscape crash removed my chat box and I had to totally restart again to get it back. I see we're over time.
MsgId: *brain_storm(25)
Date: Sun Nov 16 23:10:33 PST 1997
From: Rob_Killheffer At: 205.198.117.104

The e-mail didn't come in yet. We're over time, but if you want to finish up what you were saying, that's fine (there's no show following, so we can run over if necessary).
MsgId: *brain_storm(26)
Date: Sun Nov 16 23:16:23 PST 1997
From: Rosalind_Picard At: 18.85.12.147

Some closing lines, since I have a 13-month old alarm clock that will get me up early tomorrow. Affective computers will enable much more subtle communication via computers, allowing better transmission of vocal intonation and other forms of emotional expression; they will also enable new applications, such as computers (and software agents) that naturally pick up on your preferences and help sort through information, etc. For example, we've prototyped eyeglasses that sense when you're confused, and a wearable camera that doesn't just save oodles of video, but saves the seconds that preceded the moment you were most startled, and so forth. In other words, they pay attention to what gets your attention. There's much much more (Chap 3, several future apps.)
MsgId: *brain_storm(27)
Date: Sun Nov 16 23:18:07 PST 1997
From: Rob_Killheffer At: 205.198.117.104

Thanks, Roz. Sorry it's been so troublesome. It was a nice talk, anyway, and of course, interested viewers can check out your book to learn more about the subject. Thanks for coming on! Now go get some sleep.
MsgId: *brain_storm(28)
Date: Sun Nov 16 23:19:40 PST 1997
From: Rosalind_Picard At: 18.85.12.147

Thanks for the chat; I hope people enjoy! Cheers, Roz
MsgId: *brain_storm(29)
Date: Sun Nov 16 23:21:27 PST 1997
From: Rob_Killheffer At: 205.198.117.104

Good night. Viewers, tune in next week when my guest will be Stanford philosophy professor Fred Dretske, author of "NATURALIZING THE MIND." See you then!


Home || Prime Time || Live Science || Machine Dreams || Project Open Book || SF-Fantasy-Horror
Continuum || Antimatter || Mind-Brain Lab || Interactive IQ || Gallery || OMNI Toons

Questions, comments and suggestions can be mailed to the webmaster.


Copyright © 1998 by Omni Publications International, Ltd. All Rights Reserved.