The transformers
Document Actions
In the largest lab of its kind in Israel, Dr. Gal Kaminka is perfecting the army's new secret weapon: electronic border policemen that are programmed to anticipate and counter the actions of terrorists - but without emotion, of course.
t's hard to believe, but the vacuum cleaner robots
that Dr. Gal Kaminka ordered from Japan are supposed to be "drafted"
into the Israel Border Police. When Elbit Systems, the defense
electronics company, completes work on it, the "patrolling robot"
program that Kaminka has developed for the Defense Ministry will be
implemented in jeeps guarding the country's borders - without human
contact. The electronic soldiers will send back images and monitor the
humans they encounter. "At this stage they are not yet shooting," says Kaminka, who is the head of Israel's largest robotics laboratory, at Bar-Ilan University in Ramat Gan, "but I am sure the army is also working on that.
The idea of building the prototype of a patrol robot came after Kaminka worked with vacuum cleaner robots over a long period of time.
"In 2005, for a year, we built a family of scientifically revolutionary algorithms for vacuum cleaner robots," he relates. "We took robots that are sold commercially and replaced their 'brain' so they would clean more efficiently. As a result, we understood the similarity between cleaning and guarding - it involves repeatedly covering a territory."
The main problem that Kaminka succeeded in overcoming was a tendency by people, as well as robots, not to cover the different parts of the area they are patrolling in the same way. Generally they visit points in the center of their sector more, and neglect those on the margins. The solution he came up with is based on cooperation between robots instead of work by individual ones: "We discovered that during a patrol in which robots operate together, the frequency of visits to different points is greater than with robots that move around individually."
Kaminka designed the plastic Border Policeman robots to move about in an unconventional way. "If there is an adversary watching and studying you, then naturally you want to switch direction every so often," he explains. "The question is how to do that intelligently. If you switch completely at random, then those watching the robots will begin to grasp the probability of their moving in a certain way. You have to change direction so that even if the observer knows within reason which way you will turn, you can still catch [that observer] with high probability."
The large size of the robots and their automatic mode of operation may unnerve people. Isn't it possible, Kaminka is asked, that some computer malfunction will make the jeep go wild and crush everything in its path? He explains reassuringly that the robots' computers undergo strict safety tests that will prevent them from malfunctioning. Even the smaller robots, he says, will not run over people or objects: "The moment a robot encounters something, it stops moving. When I brought the cleaning robots home to try them out, my kids figured out very quickly how easy it is to abuse them. If you stand in their way they stop moving."
Kaminka's previous project, Smart Entities, also involved the development of computerized brains for the Israel Defense Forces and Elbit - brains not of Israeli soldiers, but of Palestinians. The project concluded with a presentation of what are called virtual-reality figures, intended for use in simulations during IDF exercises. "[The Defense Ministry and Elbit] wanted each figure to be very realistic, but not necessarily predictable," Kaminka says. "That is, they would not necessarily escape to the nearest hiding place. The figures had to have weaknesses, too. And we had to define how well they know the area they are operating in."
What do these virtual figures know how to do?
Kaminka: "The idea was to reach capabilities that resemble the thought process of Palestinian entities in an urban environment, with the focus being on terrorist squads. You want to simulate a squad that sets out from a certain place, moves to the edge of the village and tries to reach the fence and plant a bomb. The squad has to be able to choose routes, split up into smaller groups and reorganize, react to threats, and cooperate. If they hear a helicopter, they must warn each other, hide or shoot back. This is done in a way that approximates reality. To just flip a coin and choose a mode of action randomly is something anyone can do, but it doesn't take into account what people actually do, the way they react to their environment."
Is it possible for the targets to be afraid?
"The clients wanted to move on to emotions - that was the second stage of the project. But Bar-Ilan does not specialize in emotions, so Elbit did it. The idea was to achieve a situation in which a frightened entity reacts differently than a braver or better trained one. In general, we are interested in the effects of pressure and tiredness. We want [the virtual terrorists] to have the same strengths and weakness as people. For example, the speed in which a robot thinks must be similar ... We want to understand how people think and simulate it, including the mistakes they make. People are sometimes not rational, and they sometimes make obvious mistakes in their choices."
Empty brains
Dr. Kaminka programmed the computerized Palestinians with the use of artificial intelligence software from Michigan-based Soar Technology, which aims to simulate the activity of the human brain. "The program keeps learning," he says. "It concentrates on just one thing at a time. It's a bit like people: When you first learn how to play the piano, you have to concentrate on what each hand is doing; it's only when you attain a state of automatic thought that you can work with both hands simultaneously."
Is there any significance to the fact that you are simulating a Palestinian rather than an Israeli brain?
"Not for me. I suppose that everyone has the same brain, the same architecture. We brought Elbit an empty brain with a small number of parameters they could play around with. If they inject knowledge of a Palestinian, it will behave like a Palestinian; if they inject knowledge of an Israeli, it will behave like an Israeli."
Isn't it a bit simplistic in terms of understanding the brain to think that the same computer commands can simulate armed popular resistance and also an organized army?
"The brain of armed popular resistance and of an army is the same brain, but not the same knowledge. You look at the box as a unit, but I separate it into two parts. My role, in terms of dealing with artificial intelligence, is to distinguish between the architecture of the processes and the architecture of the knowledge that makes them occur in this or that way. My point of departure is that we all have the same processes."
A key trait of the Palestinian entities that Kaminka developed is their ability to cooperate with one another as part of a team. "There are general principles of teamwork," he explains. "For example, if you know something and others don't, you have to tell them. If there is a mission, a joint decision has to be made on how the tasks are going to be divided. Those matters can be turned into algorithms that will be correct no matter what the mission is."
Because of the project's military nature, it apparently has to take into consideration a clear chain of command. "There is a commander and a deputy commander and a whole chain of robots that replace one another if one escapes or is hit," Kaminka says. "Our robots are exemplary team players. If I send them in a row toward a cliff, when the first one sees the cliff it will not be able to stop, because it does not have the freedom to stop before receiving the agreement of all of them."
Are there other ways robots can work in a team?
"Yes, there are robots that cast a vote. We used that in a simulation of robots moving in a certain formation. When they had to decide about moving from a triangular to a linear formation, a vote was taken and the majority opinion prevailed. A vote can be preferable in a situation in which I want to reflect more accurately the knowledge of the majority of the robots. For example, in a convoy the robot in the front is not necessarily the leader. If it sees a cliff and says, 'We have to stop,' and the commander is in the back, they will not stop, and they will all fall - apart from the commander. A democratic decision is better in a situation in which we assume that the majority of the robots have more information than one robot."
Kaminka tells a story about a group of anarchist robots that cleaned floors and were able to work with one another by mutual consent, without resorting to use of a leader or a vote. "All the robots have equal rights," he notes. "They work as a perfect team and help one another, but none has the right to tell any of the others what to do, nor is that necessary. They divide the mission among them - each knows what to do. The only time you have to make a decision about which does what, is [when you need to decide] which of them will help another. That decision is made according to which of them is closest to the event. If the distance is the same for all of them, they flip a coin."
Asked if he has not actually constructed an ideal model for human society, Kaminka says: "It rarely happens that I look at robots and think, 'If only people would work like this.' That might be the case on a patrol, where I know that the robots do the work better than people can. You can say teamwork among robots is better than teamwork among people, but the question is, of course, what you are looking for in people. I cannot reconstruct people's dynamics, their ability to adjust to what is happening around them."
Mexicans vs. robots
Kaminka, who is 37, wears shorts in his lab and spices his remarks with jokes about computer people that only he and his students understand. He relates that as a young programmer, he was already interested in the connection between computers and human behavior.
"We had a computer at home from the time I was 11, and I always wanted to build a smart computer and understand what intelligence is," he explains, adding that he started computer studies at the age of 17 in the United States, where he lived with his parents. He took courses at a university in America in Talmud, philosophy and economics, and when he returned to Israel he was drafted into Unit 8200, a secret intelligence unit. He completed his bachelor's degree in computer science at the Open University.
After his discharge he worked for companies engaged in virtual reality, but quickly tired of the field and did a Ph.D. at the University of Southern California. "Virtual reality is largely about using graphics and trying to make them work faster," he says. "I was far more interested in programming things easily, an area that is becoming more difficult today. The truth is that I want things to work in the world."
His greatest ambition is to design robots that will replace people in boring and repetitive work. "I was familiar with the automatic car wash places you see everywhere in Israel, and when I got to Los Angeles, in 1995, I was amazed to find that they didn't exist there. I was told that it is simpler to employ Mexicans. The machine that washes cars is a very non-independent robot: It has to be told to start and it hardly has to decide anything. So think how absurd it is that it's more expensive than a human worker. There are jobs that are a waste of the human brain: patrolling, cleaning, assembly-line tasks - whatever is monotonous, repetitive, boring and eats away at the human soul. There is no reason that robots should not do these things."
According to Kaminka, not only are robots resembling human beings more and more as technology advances, but also the opposite is true: The "robotization" of people is manifested in a mechanism that he says he heard about, which operates in a fast-food chain. "It turns out that one of the chain's problems is the need to supervise work that has to be done within given timeframes. Every so often the place has to be cleaned, the cooking oil changed, all kinds of products ordered. They developed a system that tells each worker what he is supposed to do at a given moment. I find it terrible that the computer tells people what to do - that is precisely human robotization. The computer issues a command, 'Go and clean the bathroom,' and knows that no one will do it better than a human being. But I can think of a situation in which people think that this is excellent, because they don't have to bother to think about it."
In addition to developing robots for practical uses, Kaminka also tries to implement psychological theories by means of computerized models and robots. An example is the "crowd behavior" project he is preparing for the Defense Ministry. "From psychology we took what is known as 'social comparison theory' and tried to transform it into an algorithm. The idea was to create virtual characters that would approximate the behavior of a human crowd in its collective actions."
How does a crowd behave?
"One characteristic of a crowd is that it has a tendency to work as though it has a central brain. The people in a crowd do not talk to one another, and there is no one who decides things, but they act in coordination. We want to understand how that happens. The theory I worked on assumes that people look at one another and adjust themselves when there is no objective gauge, when they are unable to rely on themselves."
The example of the crowd that Kaminka uses is pedestrians: "That is a crowd that operates in the form of currents. People walk behind each other in the same direction in order to allow smooth passage in both directions. Moreover, groups or couples walk together and bypass obstacles [while going] in the same direction."
Kaminka says the model he developed shows that people adjust themselves in relation to one another when they possess the tools to decide independently how to act: "If we take what the theory says and run appropriate algorithms, we get an incorrect result. The situation created when people adjusted themselves to one another only when they did not know where to go, does not resemble the true behavior of pedestrians. But if I program the characters in the crowd to compare themselves to the others all the time, without any connection to the knowledge they have, we get a model that resembles reality."
In the program generated in the lab, increasingly large groups of people form within the crowd. Even though the groups have no leader, many people tend to look at one person who is aware of the goal and in practice leads the others. The reason the Defense Ministry supports this project, Kaminka explains, is because of its potential to show how a crowd behaves when it gets a warning about a suspicious object or during a terrorist attack. "There is little documentation of true events, and it is impossible to organize anything like that. On the other hand, they want to adapt the way buildings are constructed and the way a crowd is evacuated to real crowd behavior in such eventualities."
Another project he is involved in deals with a different aspect of crowd behavior. For the Industry and Trade Ministry, he has developed a robotic eye that can identify nonnormative behavior of individuals in a large group of people. The system is based on a computer that for three months "learned" how people move normally, such as in a corridor. It identifies people who try to pass others, who walk against the current or zigzag, and who generally behave differently from the others.
Kaminka: "This system identifies sequences of activities and can warn about a nonstandard series of actions."
However, the system goes further: "Not every unreasonable behavior can hurt you. The idea is to create an algorithm that 'sets a price' for abnormal sequences and warns when the price is highest. For example, if our system identified someone leaving an object in a public place, the probability is that the person forgot it. On the other hand, if it's a bomb the price is quite high, and you have to find the way for the system to warn about it."
Electronic caregivers
One of the most powerful motives for developing the robotics industry has been the need for more caregivers for Japan's steadily aging population. The manufacturers of robots in Japan, some of which are at the cutting edge of the industry worldwide, are preparing for this situation and eyeing the vast economic potential of producing electronic caregivers. Kaminka emphasizes that this is not feasible with current technological capability, however, and that it will take many years. But he does find this development extremely positive.
Would you like a robot to look after you when you are old?
"That is not the question. The question is what the alternative is. What do you prefer: a robot or a stranger you cannot rely on? The robot is 100 percent reliable: It can take your blood pressure, it can communicate with the person it is looking after - which caregivers do not necessarily know how to do - and it will be strong enough to pick up the person. I can give you a whole list of very positive things about robot caregivers. If I go to a hospital and have a certain face and a certain level of education and I call myself Dr. Kaminka, I get treatment that someone else will not get. With robots that would not happen, because everyone would be treated in the same way.
"Of course, there are also negative things. I don't see robots emulating human warmth and empathy - though I don't see modern medicine doing that, either. If this service will save the nurses time, maybe they will have time to empathize."
Kaminka is astounded to hear about strong emotional bonds that can arise between people and their robots: "I heard that there are people who take their robot vacuum cleaner on family outings, rather than leave it home alone, so it will not feel hurt. That sounds about as idiotic as taking your iron or your washing machine on an outing, but on the other hand, if it makes people feel good - why not?"
Of a Japanese researcher who developed a robot that looks like a seal and makes seal sounds for residents of old-age homes, he says: "I suppose they know it is not a real animal, but it moves around a little. If you talk about it, it turns its head to you and makes noises. But experiments will show that it makes them feel good, so what difference does it make?"
In the coming months, Kaminka will receive two humanoids in his laboratory. What makes these robots special is that their manufacturers try to copy not only the human brain, but the human anatomy as well. They look like short people, with flexible knees and elbows, and they also have a neck that can move up and down or sideways. They even have a face: a camera in place of eyes, and a mouth drawn on their head.
Kaminka defines the humanoids as a "concentration of problems," which illustrate the complexity of trying to control the human body. "For this computer we will write a program that will resurrect the dead body," he says. "When they arrive, all they will know how to say is that their battery is drained, and after we recharge it they will lie motionless. We have to do everything, from making them walk to enabling them to see and to communicate with people."
Programming humanoids reveals the close connection between different problems related to the activation of robots (and also to activating the human body). "All the problems are interconnected. The camera is unstable because the robot moves, so it is impossible to distinguish between movements of the robot and changes in the surroundings; it is impossible to deal with the camera without thinking about motion. If the robot walks, for example, or if the camera operates only when it is stationary, the problem becomes easier. Because of the robot's limited conceptual power, each of its functions takes place at the expense of others."
Despite the successes in creating robots that resemble human beings, Kaminka says that only very rarely have robots surprised him with their cleverness.
"If anything, they usually surprise us with their stupidity," he notes. "So whenever people talk to me about the theory that robots will take over the world, I laugh. Because I know how hard I have to work just to make a robot do something quite silly. Robots are not smart enough to take over the world. If only they were a bit more independent - that would save me a lot of time."
Source : http://www.haaretz.com/hasen/spages/1000981.html
Crime Mapping Website - Your Own Batcomputer
