 |
MOC: Why would we want computers to have emotions or to recognize emotions?
Picard: It's hard to imagine walking into a store and asking for a machine with more emotions, like more RAM, because we think of emotions as irrational. What scientists have been learning is that that emotions play a very important role in rational functioning, as well as the well-known roles in irrational functioning. It turns out that if a person has brain damage in such a way that they have too little emotion, then you might think as first that they would be highly rational, but in fact we find that their behavior is severely impaired. Yes, they seem unemotional and detached, but also they malfunction terribly when it comes to making basic decisions, like Ôshall we get together for tea tomorrow', and they wander off down this infinite chain of possibilities, and never really return with a reasonable decision, and feel no embarrassment for taking so long to make such a simple decision.
MOC: Can you talk about the relationship between emotions and intelligence.
Picard: Emotional intelligence has come to refer to a set of skills which some people say is more important for predicting your success in life than IQ, the traditional Intelligence Quotient. Instead of focusing on mathematical and verbal ability and things like, could HAL play chess, it focuses on social skills, like could HAL sense if people were disturbed by what HAL said. Can a person sense if another person likes or dislikes what's going on, is at ease, is troubled. Then, can the person respond is a socially adept way. Those are skills of emotional intelligence; dealing with the emotions of another person and dealing with the emotions of oneself. It's interesting, we see in 2001, HAL having the appearance of having fear and paranoia. We think of how HAL dealt with HAL's emotion, if we call that a computer's emotion, and we don't think that HAL was that emotionally intelligent because HAL started killing people when HAL was paranoid instead of perhaps going to the source of his mission goals and talking to them about this problem, or talking to the crew mates about what was going on. ÔMaybe you guys should reconsider, what does disconnection really mean? Will you power me back up when all is well?' No, there's no sort of rational treatment of HAL's emotions. Instead, HAL acts paranoid and starts killing people out of this fear of death.
MOC: How would you rate the emotional intelligence of today's computers and what is your vision for the future?
Picard: Today's computers have dreadful emotional intelligence. As an engineer I think that I can sometimes say that we engineers have largely made them in our image. Machines in our image of being ones that are maybe great at technical skills, math and science and so forth, and have put people skills a little bit further down the ladder. Now I hate to propagate the image of engineers as lousy at social skills, because many engineers are fabulous at social skills, but in terms of priorities, I think we've emphasized the technical skills of machines over the people skills. That is changing, as the typical user has gone from being an engineer to the average person on the street, we're recognizing that machines need a lot more social and emotional skills. They need to not completely ignore you when you're getting irritated at them. They should at least be able to acknowledge when they've made mistakes.
MOC: In your chapter you were writing about the possibility of computers having emotions. Do you think this is in fact what's going to happen any why would that be desirable?
Picard: We should be careful when we say computers Ôhaving' emotions because what they really have is a bunch of procedures and mechanisms that perform functions similar to what we think emotions perform in people and animals. They do not have the feelings and awareness of feelings that we have, so what we're calling emotions in a machine is dramatically different than what we would call emotions in a human being. When we talk about giving machines emotions, what we're really referring to is things like states that help regulate and modify what the machine is doing. For example, if a robot is on another planet collecting data, and if amidst that, its sensors detect some damage, it might shift quickly to a state that helps it survive. We might call it a state of fear because it has shifted its resources to self protection survival. It can no longer focus on its primary task. We might call that fear, but the robot might not really feel fear like you and I know fear. But to an outsider it might look like it feels fear. So that's an example of an affective state that we might think is vital to give that computer. There are other affective states we might give it such as a background sense as to how well the interaction with the user is going.
MOC: What's been the most successful applications of this to date, or is this it just in the realm of pure research?
Picard: There are some applications out there now of affective computing. They meet with great success in some people's opinion and not in others. This goes to show that how affect is handled is highly individual, and that what works for one person may not for another. One example application that's out there now is the Eudora chili peppers that's raiding your email for inflammatory content. If you're composing an email that appears to be something that you might regret later, it labels it with an appropriate number of chili peppers. Some people are offended and some appreciate it. If the person is showing dislike for the way the computer's behaving, the computer recognize that and back off. We've modified the Microsoft office assistant so that if you let it know you don't like it, it apologizes and removes itself from the screen.
MOC: Can you describe facial analysis and was that successful?
Picard: There's been a lot of research on recognizing affective states from facial expressions. There's, depending on how you count them, 6 to 8 basic facial expressions that can be recognized by the muscle movements to interpret if the person is showing joy, anger, sadness, fear, surprise, disgust, things like that. We have found that those facial expressions, although they're very important across cultures for social communication don't necessarily occur as frequently as some other facial expressions that happen between users and machines. Although with a limited number of subjects who are deliberately making those say 6 facial expressions; we can recognize them with say 80-95% recognition accuracy. That's still a far cry from recognizing the natural communication from the face during ordinary interactions.
MOC: Would you like to say anything about Kismet?
Picard: Kismet is a wonderful example of trying to insert the ability of the machine to both recognize and express emotions into a human Ð machine interactions. Kismet is starting at a level that is a little more workable right now presuming Kismet is an infant as opposed to a sophisticated adult. Infants we think of as having very few emotional states and probably not able to recognize a whole lot from other people. Some coarse signs of excitement and maybe pleasing and displeasing. So that's a nice level to start with because it's a little more doable right now. Kismet also exciting research in machine learning. Traditionally machine learning has worked by the researcher deciding what the machine should learn and giving the machine a fixed space of things to learn, and hoping that the machine might be able to generalize from that space. With Kismet, the idea is that humans learn continuously by interacting with one another, and if another human in there to scaffold the learning, and say Ôoh no that's not right' or Ôthat's very good' or Ôlook at this', calling attention to things, helping limit the space of input then that's a much more natural way that people are accustomed to learning. It's more natural for us to interact with a machine that way. So Kismet is really the first computer system to really learn in a way that takes the affective scaffold from its teacher.
MOC: Looking ahead, do you expect complex AI entities that we interact with naturally, and seem to have emotional intelligence, and why would that be useful?
Picard: I think there's going to be a big push towards developing AI entities and smart systems that have emotional if only so that it will be easier for us to interact with them. In fact, in Ô2001' that was one of the points Dave made. That HAL gives the appearance of having emotions. He's programmed that way because it makes it easier for us to interact with him. I believe that we'll find it will just be easier to interact with systems when the systems interact on out grounds. On what is natural to us rather than forcing us to learn some set of behavior and principle for interacting that aren't natural for us. That doesn't mean that the computer is going to be as emotionally smart as a person. It doesn't mean that the computer's going to have feelings like we do. It doesn't mean that the computer's going to be as smart as we are. But what it does mean is that the computer's going to try to operate on our ground. Now, how well it succeeds at that is a big open question. One of the things that concerns me is that giving computers the skills of emotional intelligence, I think is in many ways similar to trying to give people with certain disabilities or impairments the skills of emotional intelligence. For example, autistic people who can be very smart in certain ways, are often very impaired in their social emotional skills. People who work with autistic people, to help them develop their social or emotional skills, find it very difficult to teach them. It's very much like teaching a computer, they say. They can learn a very specifically taught lesson, but generalizing that lesson to natural situations is very hard. Computers are very good at learning the specific pattern that you teach them, but generalizing that in a sensitive way to new situations is quite difficult for them.
MOC: Could you talk a little bit about how our brains are built?
Picard: Our brains still remain a massive mystery. We're starting to learn that the emotional centers of the brain are activated quicker by certain inputs then the higher cognitive centers of the brain. Certain messages go straight to the emotional part of the brain and only very short moments later reach other parts of the brain. That has implications for how emotions can influence our thinking even before we're aware that they're influencing our thinking. Emotions are playing Ôbehind the scenes' roles in intelligent processing that we hadn't previously acknowledged. For example, a lot of your day to day decisions are not purely logical outcomes, but they're weighing the trade-offs of your time, of other people's feelings, and those priorities change constantly. The emotional system seems to be the way we regulate and adapt priorities on the fly in what appears to be an intelligent way. If these emotions are removed we become very brittle and slow at responding to situations in what appears to be an intelligent way.
HAL was a system that was all over the spaceship. You couldn't go anywhere where HAL couldn't see you. Even in the pod we know that HAL read the lips of Dave and Frank. This raises a question that is becoming a greater and greater concern to all of us who are recognizing that technology is seeping into the furniture, into the walls. It's not just a video camera surveilling you as you go to make a transaction at the ATM, but also the technology around you can be sensing who's there, how long they're there. Now, not just what they're doing but how they're doing it. Now, that can be good for some people in certain situations where, say, we've been building systems that to help monitor your stress. If you've got too much stress in your life, you want to understand what events in your day elicit stress, then we've built some things that can help you begin to try to assess some of that. We built a system in the car that can keep track of your behavior and look for signs of stress which can be useful for example for the navigation system and other smart systems that they're putting in the car to know things like not to interrupt you now because there's three pedestrians crossing the street and you're sort of distracted.
We have been looking at moving the sensors from the environment onto the person primarily for control by the person. It's hard to control a system that's behind a wall. You might see that the lens is there and when you can cover it or not, if it's a camera. But it's hard for the average person to understand what that system is really doing. We've been building things that are much simpler like this little wearable sensor that maps a signal measured from the surface of the skin to a little simple readout that you can either just look at or show other people. In this case, it maps the skin conductivity across the palm to an LED, and when the skin conductivity is low you can baseline it so the LED's very dim, and when the skin conductivity is very high, the LED is bright. Mine will glow brightly shortly here. We also have a little pin that you can read this out into your computer or your wearable system that can help you keep track of this information over the course of the day. When the level is low or high, and what may have caused it. Was it the jog you took when you perspired a lot and your skin conductivity went up or was it the talk you gave that made you really anxious and your skin conductivity shot up. It gives you control over it because you can take it off, put it on, and know exactly what it's doing, whereas if this was sensed by some electromagnetic field in your office you'd have a hard time knowing when it was on or off and what it was sensing and what your own signals were doing.
They gave the gloves out to a movie audience of 1200 people. When they were bored the lights went dim, and when people laughed or were involved in give and take, the audience tended to glow brightly. So we recorded the brightness level of the audience over the period of the show. This was both feedback to the people on stage as well as an amusing and interesting way to look at the course of the day's events. These have been interesting to students in classrooms, who have noticed that when the teacher asks them to just read from something they're not interested in their light goes dim. When asked to write in their journals or devise their own experiments their lights glow very brightly. This particular symbol is a good readout of your engagement level.
We've been trying to build sensors that are not only wearable but washable and you'll see new lines of clothing coming out in the not-too-distant future that involve heart rate sensing and sensing some muscle tension in ways that are meant to help you understand more about what your body is doing. Again, our emphasis is on gathering information in a way that is presented to you, that's private and under your control, as opposed to gathering it by some big machine in some big space that is taking it away and who knows what's being done with it.
MOC: How would you rate HAL's Emotional Intelligence?
Picard: I wouldn't rate HAL's EQ very highly because you have to wonder why was he so afraid of death, if death for a machine is simply being turned off and quite possibly later being turned right back on. Why couldn't HAL understand that death for HAL would be a temporary thing? What we see is HAL acting quite irrationally, out of paranoia of his own death, disconnecting the life support of human beings, who have no chance in this life to pop right back up.
MOC: Can you describe HAL's affective state?
Picard: Dave and Frank have no access to HAL's emotional state. They did not know that HAL was starting to be paranoid. All they could see was that it appeared that HAL might have made some mistakes. What we don't see until the end or until HAL starts disconnecting people that there are strong behavioral indications that HAL is really disturbed. Only when HAL makes these moves to kill people do we really see that HAL is acting irrational and acting paranoid. We don't hear it in HAL's voice we see it in HAL's actions. It's interesting that HAL wasn't given a face like so many systems today are being given animated agents, or characters, or faces. All HAL had was a voice and that voice was really pretty unemotional throughout the film until that famous last scene.
There is the line where HAL says to Dave "without your space helmet, Dave, I think you'll find that rather difficult" HAL's remarks are rather sinister at that point. Even Dave, who HAL previously talks about having a wonderful relationship with. He seems to be easily discarding that friendship with Dave and is out to get him.
MOC: Are emotions separate from sense of self?
Picard: Emotions are very complex and so is consciousness. Some emotions, like fear, sometimes don't require self-consciousness. Other emotions such as shame and guilt, we refer to as self-conscious emotions. Until you have a sense of self, these emotions don't really even make sense. So very rudimentary emotions, we can build in mechanisms of without having to worry about notions of self. As we get to some higher ones, we do need to think about what it would mean to give the machine a sense of self.
To learn more about Steven Pinker visit:
Or
|
