"The Intimate Machine" premiered October 22, 2002.

				Friendly Characters Animated characters can see, hear and react to computer users, making everyday interactions with computers easier, more efficient and fun.
		Select text to jump ahead in the clip: ALDA: It was six years ago that I had my first encounter with a sociable computer, or at least one that tried to be sociable. How does that feel? Now, these wires go from the gloves... down my back? Down your back and into the computer. And this eye tracker... ALDA: I was visiting MIT's Media Lab, a visit which in those days often involved getting rigged up in elaborate outfits designed to keep track of what I was doing-- and in this case, where I was looking. ALDA: Oh, my God, look at my eye on that monitor. ALDA: The gloves were to allow the machine to follow my gestures. ALDA: Oh, it speaks Italian. ALDA: The idea of all this was to allow me to gesture to my computer. ALDA: I actually gesture at my computer quite a bit, but the computer doesn't respond. In that regard, I can understand how it would be good to have gestures understood by the computer, but why are you engaged in this? Why do you want the computer to read my gestures? Because we do gesture. We gesture at computers, we gesture at each other, like I'm doing right now. We even gesture when we talk on the telephone. So yeah, there you are. You're saying to your computer, "Not that, the... I told you..." But mostly it's like this. I go like that. And it still doesn't respond. No, it doesn't-- it doesn't understand. It's a slow computer. Well, we'd like to speed it up a little bit. So the idea is that you try to talk to Gandalf... ALDA: The Media Lab is famous for its demos-- demonstrations of technology that is still far from perfected, but gives us a glimpse of the possible. The idea here was for a character I could converse with. COMPUTER VOICE: Hello, I am Gandalf, your guide to the galaxy. ( loudly ): Take me to Mars. Why not? ALDA: I love that eye action. Good-bye. No, wait, where are you going? Wait, are you going to leave me out here in space? ALDA: Gandalf had trouble recognizing my speech, let alone my gestures. ALDA: Tell me about Mars. GANDALF: Mars has two moons. Tell me more. Mars is 6,000 kilometers in diameter. ALDA: But he did have a cute personality. Mars is a cool planet. ALDA: You said it. Uh, what was the point of wearing all of this? What did that do? How did that change anything? So, we talked earlier about the interaction of verbal and nonverbal behavior. You saw an example of that. He knows, Gandalf knows when you're looking at him, and he knows when you're looking at the screen. And when you looked at the screen, he looked at the screen, too, just to be friendly. And when you looked at him, he got ready to answer a question, or to interact with you. When you looked away, he took a powder. Yeah. Did my making gestures change the way... I didn't see a change in the way he answered questions when I used my hands. In the future, he'll understand the content of your gestures. For the moment, what he understands is that you're speaking. The ways we've thought of computers up until now has sort of kept computers in the box, and the way you're thinking of computers is giving them a chance to get out of the box by thinking of them more as people. I mean, you're... you're giving... you're making them... listen and feel and hear and see like people, so then... then they can be all over the place. They don't have to be only accessed by a keyboard and a mouse. Absolutely, that's the point. They should be able to be social actors-- actors in our social world. They're not going to be people, because they're not people, and we don't want to think that they're people, ultimately. I don't. But they might be easier to use if they... if they got information from us the way other people do. Absolutely-- that's the point. ALDA: Today, Gandalf has retired, to be replaced by Rea, whose expertise is much more down to earth. I'm starting up her brains. It takes couple of minutes. She's in a deep sleep. ALDA: This is Gandalf's granddaughter? MAN: That's right, uh-huh. We've made a number of advances over the years. She does full gesture and speech generation. So when she has something to say, she actually decides whether she should put it into her speech or her hand gestures or both. So she's more natural that way in interacting with her. She recognizes everything... well, not everything you say, but everything you say about real estate. ( laughing ) She's got a one-track mind. BICKMORE: That's right, that's right, uh-huh. Her grandfather needed me to get all duded up in an outfit-- gloves on my hands and a helmet on. She doesn't need that. She doesn't need that, because she has two cameras up here which recognize the position of your head and hands in space, and from that she can determine where you're standing, so she can gaze as you move around and look at you. She can recognize... REA: Hello. If you're gesturing... Hello, hi. How are you doing? I'm pretty well. Well... Pardon me? Have you been in the Media Lab before? Yes, I have. Cool. Cool. They are doing some crazy things in here. ALDA: Yeah, I see... ALDA: Tim and Justine have found that people tend to trust computer characters more if they chit-chat before getting down to business. ( laughter ) What did she say? ALDA: But Rea's patience with me quickly runs out. ALDA: She said good-bye? Yeah, I have never seen her do that before. Look, I... I... listen. I swear to God, I really think it's cool. ( Cassell laughing ) ALDA: So Tim resets Rea's brain and we try again. Hello. ALDA: Hello. How do you like Boston? Oh, I like Boston. Well... What? Don't go. No, no! ( laughter ) I want to buy a house. What? What? How about this weather? ( Cassell laughing ) Yeah, it's cool. ALDA: Uh, do you sell houses or anything? Rea, real estate. ( Cassell laughing ) Does that strike a chord? ALDA: This is a best example of soft sell I've ever seen. CASSELL: I know. She figures if you don't want to buy, she doesn't want to push you. I think she just sees in me that I'm not really serious. I think so. Look at her, she looks sad. ALDA: She really does. ALDA: But finally, Rea gets down to business. REA: It has one bedroom, one bathroom, one living room, one kitchen. Could I see the kitchen? ALDA: In theory, she's listening to what I'm saying with both my voice and my hands. ALDA: Where is the living room? ALDA: And then figuring out how to respond-- again with both voice and gestures. The living room is downstairs across from the kitchen. Uh-huh. Could I see the living room? REA: Were you talking to me? Yes, I was. I'm sorry, I did not quite catch that. That's okay. ALDA: Like Gandalf before her, Rea's dependent on an off-the-shelf speech recognition system, which is having trouble understanding me. ( slowly ): Show me the living room. Ah, here we go. ( Cassell and Bickmore laughing ) Ah, nice. ALDA: Nice TV set there. Tell me about the living room. I am sorry, I did not quite catch that. I bet your granddad was Gandalf. ( Cassell laughing ) You have this remarkable effect on her, and I think it's actually because you have a low- pitched voice. I would suggest that you watch Tim interacting with her. Sure, okay, sure. Let me see what I have available. ALDA: Rea has had months of practice recognizing Tim's speech, so their interaction goes much more smoothly. One living room, one kitchen. What can I tell you about those rooms? Tell me about the house. Once we get recognized speech from this commercial recognition system, we're understanding what the person is saying, and then Rea, in real time, as you're having this conversation, is using her intelligence to decide what an appropriate response would be, to formulate that response in English, and also to know how to use her body. And that had never been done before. BICKMORE: Where is the bedroom? The bedroom is upstairs. ALDA: Justine hopes that Rea will evolve into virtual characters that could make all sorts of interactions with the computer easier and more natural. Yes, there is a beautiful garden with a gazebo. ALDA: She's already developing a fitness instructor that could keep an eye on you as you work out at home, a tutor for long-distance learning and Sam, a virtual playmate for children. CASSELL: When Geneva starts to play, Sam is all about eliciting from her the best of her language, the best of her storytelling, and then giving her back what she needs to be encouraged, to be sophisticated in her language use and to be actually focused on the kinds of language use that are going to help her as she goes through school. So Sam hears her speaking, not only understands... the story part of it, but understands what she means and picks out the best parts of her storytelling and encourages that? That sounds like an awful lot for Sam to know. Right, and Sam doesn't do any of that. ( both laughing ) Who does that? Tune back in in a couple of years when we can do that. ( laughing ) ALDA: What Sam can do now is share the Magic Tower with Geneva... SAM: Cool! Okay, my turn. ALDA: Sam can even share toys across the boundary that divides the real and virtual worlds. ALDA: How does Sam know what she just did? CASSELL: There are sensors in the bottoms of the toys and sensors in the Magic Tower, and that's what lets Sam know whether one toy is in there or two toys. SAM: Long ago there was a fairy, an old man and a wolf... ALDA: Sam's background is generated by a video camera in Geneva's world, so even the room is shared between them. SAM: ...little voice crying from outside. "Help, I'm a fairy who has lost her wings." ALDA: Sam is happy to tell his own stories, or to listen when it's Geneva's turn, prompting her if she pauses. But... SAM: What happens next? She kept falling off the branch... Hmm! So she took the watering can and she poured it on each of her seeds. She's just... totally inspired by this. Would she be that... that active in her storytelling if she wasn't prompted by Sam? That's exactly what we looked at. So we ran a big long study in a local school, and we had four conditions: one kid playing alone without Sam, two kids playing with each other but without Sam, one kid with Sam and two kids with Sam. And what we find is that without Sam, they stall. They hum, they look up at the ceiling, and that's it, they don't tell. And one kid with Sam, even with this minimal feedback, and the fact that she knows that when she's finished, Sam is going to take a turn, seems to inspire children to be really creative. GENEVA: She gave the boy two, a radish and a lemon, and carefully... SAM: Then what happens? Picked them up... Hmm! And they put them down. ALDA: It's very interesting. She's moving these figures around the house, and every time she puts it in a new room, she looks up at Sam and gets a reaction. CASSELL: Right. One of the reasons that I launched this project was in response to what I saw as a series of games that talked at children-- computer games or other games that just bombarded the kid with images and noise. And I wanted something that listened and that evoked real physical play in the physical world, that got kids to move around. ALDA: Geneva, how much fun is Sam as a playmate? Do you like playing with Sam? Yeah. What do you like about Sam? Because he listens, and he says what happens next and stuff. Then he goes upstairs. ALDA: While Sam encourages his playmates to make up stories, his own are all prerecorded. SAM: Okay, my turn. Two kids walked up the stairs... ALDA: But another character in Justine's lab-- a lot less wholesome-looking than Sam-- is able to put together sentences on the fly. If you want it to speak, you just have to type something. MAN: In a way, Beat actually is sort of like text to speech, except that it's not only producing the speech; it's producing the gesture that goes along with the speech. So it takes as input text and outputs an animation script that a character can execute. This is both good news and bad news. Okay, well, that's really interesting. He goes, "This is both good news and bad news." Which is, by the way, exactly what you did. What I did-- I know. But how did he... Wait a minute. He got it from "both," huh? No, he... How did he know that there was this and this? Right. Because one of the things that it knows about the world is that there are sets of things in which one stands in contrast to the other. So we'll let you type in a bunch of other sentences that have contrasts in them, and you'll see that it'll always understand, if it has that knowledge in its database. Sometimes I like blue things, and sometimes I like red things. ( both laugh ) What was that? Why did he point towards the truck? I mean, he knows it's blue. Oh wait, wait-- the truck is blue, but he pointed to that when he said red. Let me just hear that again. I think it was reminding you that he actually likes blue better. Sometimes I like blue things, and sometimes I like red things. Now, why did he do it differently the second time? Don't ask me. I didn't invent it. Now I'm going to tell you. It's the Socratic method. ( both laugh ) I'm seeing if that's in you somewhere, but since it isn't, I'll tell you. So, in fact, not only does this system have a model of the world; it also has something else that we have, which is a model of what we've said previously, and he knows that he already said "blue thing," so there's no need to gesture to it again. You share it now. ALDA: Beat also knows a little about where he's standing. VILJHALMSSON: He actually knows what the scene looks like. I was just visiting the Media Lab. So he knows where the Media Lab is and is able to point at it. ALDA: Let's see how he says a new sentence mentioning the lab. Soon the Media Lab will have another building attached to it. There's just a little half gesture toward it. Exactly, and that half gesture, he's gesturing when he's saying the new thing-- another building attached to it-- because that catches your interest and pulls you to the contentful part of the sentence. Let me see what he does if I let him say it again. He probably won't make that same gesture, will he? Soon the Media Lab will have another building attached to it. You see? It's almost in response to the unspoken question, "What was that you said?" because he's just telling you the words. Exactly, yep. If I do it again, will he give me a look like "I told you already"? He'll start rolling his eyes. I really like the "automated actor." ALDA: Giving a cartoon character the ability to interpret typed speech as if he understands it may be useful to animators, but it isn't yet a threat to the acting profession. ...a lot easier than dealing with a real actor. CASSELL: That's you he's pointing at. And what we did to do this was, he already knows that in his world is this building called the Media Lab, he's got a blue truck. He knows that he's a so-called actor. Hannes just went in and stuck another object into his world-- an object called Alan Alda equals actor. And a real one. And a real one. ALDA: Well, if I'm going to be just another object in Beat's little world, at least I want better billing than his truck. Alan Alda is a real actor, and a great one. CASSELL: Ooh... I think this is a wonderful program. Yeah, yeah, you want to take it home. I think this gives you confidence.

				Alpha Wolf Researchers build artificial intelligence software modeled on the canine mind.
		Select text to jump ahead in the clip: Good boy, Sydney. ALDA: Bruce Blumberg loves dogs, especially his silky terrier, Sydney. And dogs have long inspired Bruce's efforts to make computers more sociable. Hello, Silas. ALDA: Silas, for instance, another of Bruce's dogs, once lived happily in a room at the Media Lab. Where's he going? Come here. ALDA: Although only a virtual dog, he'd been trained to respond to the gestures of the people who entered his life. Sit. Sit. Oh, wow, look at that. Sometimes he'll respond to your gestures, and sometimes he won't. Will he beg? Yeah, put your both hands over your head. Come on. Whoa, whoa. BLUMBERG: Now, sometimes Silas will do what he wants. So right now, I think he's heading off to get a ball, because he figures... Here he is, here he is. Give me the ball, give me the ball. Okay, okay, I'll throw the ball for him. ALDA: Several years and several thousands of lines of computer code later, Silas left his cramped little room and grew up to become Duncan, a sheep dog, with an entire Scottish estate to roam around on and a flock of virtual, and rather flighty, sheep to look after. ( virtual sheep baaing ) Away. ALDA: Bruce is trying to train Duncan much as he would his real dog. BLUMBERG: When I tell him "away," for example, he understands that command, but then it's up to his own intelligence in order to figure out how far away from the sheep to go, how to avoid the obstacles that he sees, very much as if you had a puppy, and the puppy comes with some basic abilities, and then you're training it to use this action to get particular... ALDA: Now, he's got them in kind of a lump over there. BLUMBERG: Yeah, that's good. Maybe I shouldn't do anything. He's doing a great job of herding if I don't do anything. Would you like to try it? Yeah, sure. All right. Away. ALDA: Things start off promisingly. ALDA: Steady. I've lost a sheep here. ALDA: But then Duncan and I seem to go our different ways. Bye. Pay attention. Bye. No. He's just separated the whole herd of sheep. Someone separated the whole herd of sheep. Well, if you don't mind, I'll blame him. Now, this really calls to mind an interesting question. Let's say I got really good at this. ( laughs ) Why? Yeah, why? Why have you gone to all this trouble? Well, the fundamental thing that I'm interested in is sort of the nature of intelligence. What is it that sort of is going on in our brains that allows us to do... sort of have the everyday kind of intelligence that gets us through our day? And that for me, dogs are a perfect model of that, because dogs really have dinky little brains-- their brains are the size of lemons if they're a big dog-- and yet they have the kind of common sense that probably underlies 90% of our behavior. ALDA: Bruce uses a clicker to give Duncan a virtual reward when he does the right thing. Now I'm going to see if I can teach him something. ALDA: Jump. ( clicker clicks ) ALDA: I first wait till he jumps, then say the word "jump" as I reward him with a click. Jump. ( clicker clicks ) ALDA: After several training runs... Jump. ALDA: Will Duncan treat the word as a command? ( clicker clicks ) Cool! He did it, huh? Do it again, do it again. Jump. ( clicker clicks ) Awesome! Nice, huh? You know, you are the... This is a pretty fresh demo, meaning you are the first person outside of the group that's actually trained this dog. Oh, so that must make you feel great that an amateur can do it. That's great, yeah. ( barks ) ALDA: As a graduate sheep dog trainer, my next assignment is tougher-- raising a wolf pup. MAN: All right, so now he's woken up. I'm sorry, I'm laughing like a wolf. Do they laugh? Now what? Try howling for him. ( howls life wolf ) ( virtual wolf howls back ) So he'll howl along with you. And I can wake my pup up, too. ALDA: This cute little black wolf pup is mine to mold, to shape his social interactions with two other pups-- one being raised by Bill Tomlinson, the other by Marc Downie. We can move our pups with a mouse, but the only way to control how they behave is through the sounds that wolves themselves make. ( Alda barking and growling ) TOMLINSON: Barking means play, so you're giving him mixed signals. I'm giving him mixed signals. You are. But just a nice... ( growls ) Oh, okay, I see. ( continues growling ) You know, you're threatening me when you do that. No, seriously, I'm a little frightened of you. ( whimpers ) That's better. ALDA: Just as Duncan the sheep dog models animal intelligence, these wolf pups are intended to model social behavior. BLUMBERG: Wolves are really pretty amazing creatures. A lot of a wolf's day is taken up sort of sorting out the social relationship between it and the other members of the pack. And what we're trying to do here is to understand what sort of things might be going on in the brain to be able to do that. ALDA: Bruce and Bill got their introduction to real wolf behavior through a visit to Wolf Park in Battle Ground Indiana, where several packs of gray wolves live in a seminatural environment. Each pack has a strict social hierarchy, with an alpha wolf at the top. The Media Lab researchers are now trying to model the way wolf pups find their place in that hierarchy. If you have your pup keep coming at me and growling a lot, my pup will learn that you tend to be a more dominant individual and that... it will, in turn, form a submissive relationship with your pup. ALDA: How is your pup going to learn that if you keep giving him sounds to make that go against what he's learning? Well, it exhibits that relationship by the emotional style in which it performs the actions that you ask it to take. So if you tell it to growl at somebody who is always dominant to it, it'll do so, but it'll growl kind of like this, knowing that it's a bad move and that no good can come from growling at this individual. Whereas if you try to growl at somebody who you're traditionally dominant to, it'll run over and do it very confidently. Well, let me see what happens if I don't give up and you don't give up. Okay. I mean, I want to see, can you get to a kind of standoff here? Let me see what happens. ( barks ) There, wake up. ( both howling ) I'm going to go visit you. ( virtual wolf howls ) There, that's a boy. ( both growling ) ALDA: Get up, get up! I was all over you on that one. All right, well, let's try that again. ( both growling ) There, he's getting wimpy again! What is this? Is it me? No, I got a head start, because I beat you down the last time, and so it's just easier this time. One of the problems is, too, that human speech gets recognized as whining, which is an interesting thing. Human speech gets recognized as whining? You know, this has been my problem in life. ( all laughing ) So, um... So I should just keep growling. If you want a dominant wolf, you have to make sure not to talk near the microphone. ( gruffly ): I'll talk like this. Hey, turn around. ( growling fiercely ) I think I won that one. I think I finally got the idea. Now I know how to dominate these guys. Uh-oh, is that... that's the adult. ( whimpering ) ( growling ) Now, what if I tried to get the adult to back down? TOMLINSON: Try growling at her. ( both growling ) It doesn't do any good. Look at this. I'm totally cowed by this. So what they've learned as puppies, that personality will carry on. For instance, if I've been dominant as a puppy, will I be the dominant adult? Exactly-- your adult and my adult will continue to manifest the same relationships that we trained into them. This is fascinating and fun. What do you think it could possibly lead to? Well, I think there are a couple of different ways that I would love to see this work out. Number one, as a tool ultimately for understanding more about social behavior in wolves. You know, I would really... All of the work that we do-- I really want it to inform our understanding of the natural world. Number two is that one of the things that computational systems don't do is social behavior. And social behavior is an important aspect of natural systems, so I'm very interested in what would be the benefits of incorporating social behavior into computational systems. ( howling ) ALDA: These wolf pups are, of course, nothing more than complex software programs. But they're programs that exhibit socially appropriate behavior, emotionally charged memories and the ability to learn from their experiences-- all features that would be desirable in the computer now sitting on your desk, just as long as you don't have to growl at it and have it growl back. TOMLINSON: People are often worried about making computers that are sentient and social, in that, you know, like the Terminator, they're all going to come back and kill us all and it's all going to be this horrible thing. But people all around the world are making entities that are stronger and smarter and more interesting and more capable than us. We have children, and if the process of making computer systems can be more like making children than like building bombs, hopefully it'll make computer systems that don't turn out to be bombs, that turn out to be more like children in the way that they interact with us. And just as I don't grow up to want to kill my parents, hopefully the computer systems will have wonderful interactions with us over the course of our lives. But we're going to have to deal with adolescent computers. ( chuckles ) I mean, that is one of the challenges-- that if you have a system that learns, at some point, it hasn't learned yet, and that is a problem. ( howling ) ALDA: Once our pups grew up, they sat around and howled. ALDA: What are they all howling about? BLUMBERG: Well, actually, real wolves do, um, do this. It's called a rally howl... ( computer wolves howling ) and particularly in the early morning and late afternoon. ALDA: And since my wolf is one of them, it seems appropriate to join in. There they go. ( computer wolves howling ) ( howling ) ( chuckles ) The fade-out. TOMLINSON: And so that's the end of the generation.

				Getting to Know Us Computers monitor users' emotions by measuring heart rate, jaw-tension and other physical cues.
		Select text to jump ahead in the clip: COMPUTER VOICE: Welcome to Ted Selker's office. What is your name? Ted Selker. Hi, Ted Selker. Welcome to your office. I'm going to go on in my office here. You see me going in there. Why don't you try to be Ted Selker? Okay, I'll try to be Ted Selker. ALDA: Ted Selker invented that little button that acts like a mouse in the middle of the keyboard on IBM laptops. Now he's here at the Media Lab, where his mission is to make computers aware of where we are and what we're doing. The door knows there's someone here, and it's not about to be fooled. COMPUTER: Hi, Ted Selker. Wait, Ted Selker is already in his office. Let's see, I guess I'll let you in. Ted Selker is ready to see you. Please come in. SELKER: Walk on in. Thanks, my heels are really cool by now. Thank you. Now, if somebody else comes up, it will do something a little different. Why don't you try to go in again? I'll be... I'll be somebody else. I get to play two parts in this-- that's great. Welcome to Ted Selker's office. What is your name? Ralph. Sorry, I didn't understand you. Please type your name here. ALDA: As usual, I'm having trouble getting the voice recognition system to understand me. You don't have an appointment with Ted Selker, and he is busy right now. He has just started a meeting. If this is important, please knock. Otherwise, please arrange an appointment or leave a message. This is the knocker? Yeah. I will inform him that you are here. Please wait. ( whistling tune ) Sorry, it seems that Ted Selker is very busy. Oh. Hmm. Does he know it's about insurance? ( chuckling ) ALDA: