Your desk lamp doesn’t understand you. You wish it could be more sensitive to your needs. You wish it could move freely like an animated lamp from the PIXAR cartoon company. Well, perhaps one of these days you will put your old lamp in the trash and replace it with one that communicates with you.

Crazy? Maybe not. Imagine a lamp that would “feel” when you want more light. Sensing your needs, it would slowly bend its graceful neck, bringing its illumination closer to your page to better suit your eyes, because it “understands” your movements.

At the Media Lab of the Massachusetts Institute of Technology, in Cambridge, scientists are at work, “socializing” robots. On the fourth floor of a cube of a building on Ames Street, behind a large glass window, the Robotic Life Laboratory run by Cynthia Breazeal hosts scientists who work on “social” robots. Guy Hoffman, one of Breazeal’s PhD students, is trying to build a lamp that could interact with you. You would not need to give any verbal instruction to make yourself understood; Hoffman’s lamp would follow your body language. You won’t even need to think about the lighting anymore. It would be all under the lamp’s control.

“Here, we build robots anybody could work with,” Hoffman explains, rolling the silver stud in his pierced right ear between his thumb and forefinger. Hoffman arrived in 2003 at the Media Lab to start working on his doctorate. He got his Master’s degree in computer science from the University of Tel Aviv in Israel, where he studied everything from film and mathematics to computer science, biology and psychology. Hoffman even took animation classes at the Parsons School of Design in New York. Today, amid the jumble of teddy bears, fluorescent plastic flowers and puppets that populates the lab, Hoffman’s new lamp is under construction. He thinks that if you want to produce robots that could be sold at Wal-Mart or Best Buy, and that anyone could take home to live with them, the robots should be very simple.

“We don’t think that robots should be thought of as people,” Hoffman insists, passing his right hand through his platinum blond hair. “But they should be built in a way that human beings could understand them, to work with them.”

The researchers at the Robotic Life Laboratory all agree that robots should behave naturally. In other words, what they call ”social robots” are robots that won’t necessary look like humans, but that would be easy for a human to interact with. “We know how to interact with dogs. They are not human, but still, dogs act in a way that is understandable to us,” Hoffman says. The idea he wants to stress is that a machine can be a machine, something that is not human and that doesn’t behave like a human, but still behaves in a social way.

When the first robots were created in the early 20th century, an age of political and social upheaval, they were made to look human. Two legs, two arms, a steel body and a head were enough to characterize a robot. The term “robot” was taken from the Slavic “robota” which meant forced labor. Eighty-five years ago, the robot made its first on-stage appearance in the play R.U.R. (Rossum’s Universal Robots), written by Czech playwright Karel Capek. Indeed, in R.U.R., robots were mass-produced to work in place of people. Whether or not they understood people’s needs and feelings wasn’t of any concern. In Japan, the creation of the robots inspired some visionary thinkers to see robots as a solution to predicted labor shortages. Research was funded to build robots to help avert this perceived crisis. Much work in robotics, though, aims to create not mindless worker drones but human-like companions for people. Japanese scientists speak positively about a future in which people will live side by side with animate machines. The Honda Asimo robot is a good example. Asimo looks sort of human. He can run, walk and even climb stairs. He can also reach for and grasp objects. But what makes Asimo a potential companion is the fact that he can comprehend and respond to simple voice commands, recognize faces and even interpret the natural movements of human beings. All of this raises the question of a meaningful human relationship with a machine. Will the new robots be capable of substituting for human beings not just as workers, but as companions too?

Victoria Groom studies humans-robot interactions at Stanford University in California. She assumes that as long as robots don’t go too far in acting like humans, people will appreciate them. The line the machines shouldn’t cross hasn’t been determined yet. “We’re still observing how people react and interact with machines; the issue is so subjective,” she explains over the phone. There are already clues to humans’ reactions to synthetic voices. “Some people would find it inappropriate for a robot to say ‘I’,” Groom says. “A machine would cross the line in affirming its identity.”

Today in the United States, the field of “sociable robotics” is emerging, and Guy Hoffman is not the only one to argue that we are seeing the first steps in the creation of a companion species of robots. But as Hoffman’s work shows us, it is not just the field of humanoid robotics that is expanding; there is a general shift to create alert or conscious machines.

“I want to separate being socially intelligent, and being human. That’s why I’m building a lamp,” Hoffman explains, with a dimpled smile, his eyes twinkling behind brown-rimmed glasses. He doesn’t seek to provoke. But behind his casual jeans-and-T-shirt appearance, he is serious about his work.

“The social lamp will recognize human activity in the workplace, and react by illuminating the right area at the right time,” Hoffman wrote later in an email. It will also use its light as a gesture, to point to different objects. For that, the lamp will use its articulated steel body to direct your attention, and will interact with the user, anticipating the lighting he or she needs.

Credits: Guy Hoffman

“In the case of the Robotic Lamp, people will follow the light, the same way they follow a human’s gaze,” Groom explains. In Hoffman’s concept, the eye of the lamp should work like a human iris, with a metallic diaphragm that will open and close to adjust the light intensity. Its head, a white plastic cone for the moment, will be supported by an articulated steel arm that was borrowed from another project in the M.I.T. laboratory.

“The lamp will use some of the same hardware infrastructure as the Robotic Computer, and add some new software components, especially for the collaborative part,” Hoffman says, enthusiasm shining in his brown eyes. For his lamp, he is working on a software program to enable robots to work fluently as partners in human-robot teams.

RoCo is the Robotic Computer created in collaboration with the Affective Computing Group, also based at the Media Lab. “Computers today live in autism. They can’t express or detect any emotions,” Hyung-il Ahn explains, tilting his laptop to check if any reaction occurs in the computer. Ahn is a PhD student waiting for his paper about RoCo to be published in a “review with a capital R,” as he says. His project is aimed at building a robotic computer with no explicit face that would interact with its users, moving its monitor head and neck. “RoCo has been designed to interact with users in a natural way for applications such as interactive teaching and posture improvement,” Ahn says.

Toward this goal Ahn and his collaborators are giving the system the ability to recognize the user’s physical states and make subtle responses. For instance, if you were bending too low over your desk, RoCo would stretch its head and neck to get you to sit up straight. How can a machine see that your posture is bad? Ahn explains that the “eye” of the computer is a simple camera that catches images of the user and responds to the user’s physical movements. Other sensors are being tested that would help RoCo detect the user’s emotions through facial expressions. But that’s another story. Hoffman’s lamp won’t have that option. Ahn explains that their main challenge was to make RoCo move and react in a natural way. “For that, we taught the robot how humans behave.”

Teaching a robot? Well, there is another intriguing issue. To understand what that means, you first need to know what a robot is. “A robot is similar to a computer that has motors,” Hoffman explains. “And the software we build is its brain,” he adds, spinning in his chair to sketch his idea on a dry-erasable white board. Robots can’t know everything before being placed in a person’s home – some things would have to be learned.

The term “machine learning” relates to the development of algorithms and techniques that allow computers to “learn.” Those methods create computer programs by extracting rules and patterns out of massive data sets, focusing on statistical methods. “Until now, most people have thought about robotics only on the logic level,” Hoffman explains, drawing a schematic with a red marker. “But we think that intelligence is more than logic. It also includes social behavior.” He gives a simple example of a conversation where the contextual dimension of a question shows a human’s social attitudes. “I would ask a friend, ‘Did John finally marry his girlfriend?’ And my friend would answer, ‘He realized she is a real bitch.’ My friend doesn’t answer the question directly, but I understand that John didn’t get married,” Hoffman says.

In fact, if we looked at people only on the information-based level, we couldn’t understand each other. Scientists at the Robotic Life laboratory try to look at human communication in a more psychological context. Then they try to build robots that will use these ideas in a way that makes sense to human beings. For that, they combine logic with social attitude. Inspired from the psychological litterature, Hoffman observed how humans teach each other: we ask questions and learn through dialogue. So in order to be taught something by humans, the robots must be receptive to dialogue.

Hoffman also bases his research on the cognitive approach to the use of body language in human communication. A lot of the theory was established in the 1960s by the British social psychologist Michael Argyle. He was a pioneer in the scientific study of social behavior, and his research showed that non-verbal, rather than verbal, communication dictated the impression people make on others. He observed the functions and interplay of gestures, posture, touch, proximity and facial expressions to better understand human communication.

Credits: Guy Hoffman

Basically, Hoffman can decide how he wants the information to flow inside the robots. “What you do also influences how you perceive things, so I think information should flow in both directions,” he says. Sometimes, you can perceive information and react before you have the time to make a decision. Hoffman gives the example of a very simple situation where somebody offers you a pen. At first, if the person making the offer doesn’t warn you, you need few seconds to decide whether to take the pen. After this is done several times, you don’t think about whether or not you should take the pen. You don’t need to make any decisions or analyze what’s happening. Your brain says, “Okay, this is a pen, I have to extend my arm and take it.” Hoffman calls that a “routine” reaction – some action occurs in your perception field that triggers your reaction. It is also a sort of anticipation. As soon as the person starts the movement of offering you the pen, you’ve already extended your arm to receive it. You’ve assimilated your interlocutor’s movements.

“I try to make the software resemble the human behavior,” Hoffman explains. A software program is constructed in different layers that constitute its architecture. But how does the architecture represent perception and anticipation? For the social lamp to learn how to anticipate your needs, it must learn to take the signals from your body language. Once Hoffman has understood how socialization works, he builds the different layers of the software so that it combines logic with social attitude. A typical logical chain follows that sequence: Symbol→perception→decision-making→action. A symbol is perceived and triggers an action after a decision is taken. “In our systems, all of those steps are connected to each other. So the machine will have the ability to perceive what’s going on,” Hoffman says. The fact that the software doesn’t follow the chain strictly in one way gives the software the ability to develop itself.

Then, to learn the different human postures, the software is fed a lot of possibilities. “That software is what we call a classifier,” Ahn says. That means it can interpret what it “sees” and relate it to a database. Ahn explains that the process is divided in two steps: First, they teach the software the possible postures. Then they test it. Through its camera, RoCo catches pictures and video from different human postures: straight, bending, standing up and so on. A posture becomes a feature, and numbers in the multi-dimensional area of the software represent each feature. Just imagine that for the software, each posture is a sequence of numbers.

Once scientists have taught the software the different features, the robot detects what best matches the user’s state. Through the image it catches with its eye, it recognizes your posture and interprets your body language. “Everything relies on the architecture of the software,” Hoffman explains. With those kinds of classifier software, your objects, from your lamp to your computer, become social and can understand you better.

Some consumers might get excited about those new machines; others might be leery. The lamp and the computer both become animated companions. They might be useful, but interacting with objects may strike many people as weird, despite the conditioning we’ve gone through watching movies like “Star Wars”.

Many contemporary thinkers believe that we are living in a post-human age, due to the relationships and dependencies we have already developed with machines. Rodney Brooks, the former head of CSAIL at MIT, believes that humans are just machines and have special qualities only to the extent that they are special kinds of machines—flesh and machines. Hoffman pretty much agrees with Brooks. He assumes that we already are half-human and half-machine. “Take the example of people who have a pacemaker implanted, or those who wear contact lenses,” Hoffman says, cracking a smile.

The idea of human uniqueness is being challenged by a number of robot-builders, artificial-intelligence scientists and social theorists. Hoffman firmly believes that robots can be made that people would love more than they love their neighbors, “because robots will actually cooperate with you.” He is not afraid to think about living with machines. He thinks that robots will help humans, and especially older people, to have a better and easier life. But those times are still far off. Robotic technology is still in its infancy.

“We are still waiting for the 21st century big revolution we expected in robotics,” Hoffman says. His prediction for a next technology gizmo is an implanted thermometer. “You won’t need to put your hand outside because you’ll feel the temprature through the implant, which gets its data from some server.” For the moment, he continues to work on social machines. The social lamp is still “a work in progress,” Hoffman explains, showing a visitor several plastic pieces, which may or may not be eventually used as components. He estimates it could be on the market in five years. “I hope that by the end, [people] will feel this lamp is a social being,” Hoffman says.