Designing Sociable Robots

by Cynthia L. Breazeal.
Bradford Books, MIT Press, Cambridge MA, 2002.
263 pp. Illus. and cd-rom. Trade, 49.95US$.
ISBN 0-262-02510-8.

Reviewed by Stefaan Van Ryssen,
Jan Delvinlaan 115,
9000 Gent,
Belgium,

stefaan.vanryssen@pandora.be

The field of robotics has come to age in the last decade of the Twentieth Century with the creation of highly specialised machines that perform all kinds of tasks in very different and sometimes hazardous environments: from roving the surface of a nearby planet to cleaning the endless corridors of hospitals, from replacing the hand of the surgeon to keeping lonely children happy. In fiction, movies, sci-fi and comics, robots have out-shined human protagonists: who doesn't remember HAL 2000 in Stanley Kubrics masterpiece A Space Odissey, Commander Data in 'Star Trek, The Next Generation', Asimov's helpful and sometimes troubled robots or R2D2 and C-3PO? Yet, there is a big gap between our imagination and what the machines that are actually built. It's not hard to see where the difficulty is. The robots around us are slow learners, bad communicators, only responsive in a limited way and by no means empathic. They can be smart and skilful, but they're just no good to go out with or to chat with about the weather, soccer or the pains of growing up.

Cynthia L. Breazeal, Assistent Professor of Media Arts and Sciences at the MIT Media Lab, has taken a first step towards building a robot that understands what we say and mean, communicates and interacts with us, learns and grows along the way and might eventually become more of an assistant and a companion. It is a first step, but a very important one. In this very well written and clearly structured book, she defines the key components of social intelligence for these machines. Drawing from a very wide range of sciences - from psychology to linguistics, from engineering to artificial intelligence - she actually built a robot that acts and communicates at levels of complexity of an infant. "Call me Kismet" it might say - and you would be delighted at seeing it smile, turn its head away if you come too close or angrily snarl at you when you offer it a toy instead of your face to look at.

Breazeal originally wanted to build a learning robot as much as a social robot, but she soon realised that the machine should have an interface that allows humans to monitor its learning processes. To make encouraging actions meaningful, the human tutor should get an almost instantaneous response expressing the positive result of the action. Learning, it seems, is a regulatory process for both participants.

Building on the results of developmental psychology and human-computer interaction, Breazeal chose to tackle the communication and interaction levels first, before fully concentrating on the learning capabilities of the robot. You actually have to see it to get an idea of the power and the lifelike quality of Kismet: www.ai.mit.edu/projects/humanoid-robotics-group/kismet/kismet.html.

The book is much more than an endearing description of a toy. The author gives detailed descriptions of the physical robot, the vision system, auditory, motivation and behaviour systems, the facial animation and expression space and the expressive vocalization system. Each chapter clearly states the problem and underlying theoretical principles and explains how the features are implemented, down to the level of the actual equations and parameters used.

In a final chapter, Breazeal sums up the "Grand Challenges of Building Sociable Robots". Among them are the problems of how to endow sociable robots with a rich personality, how to give them the ability to reflect upon themselves, how to design them so they can learn in very different and unpredictable social situations and many more.

Having once designed an interactive improvising computer and a 'moody' software program myself, I can fully appreciate the complexity of the challenge to build a machine that shows the full range of emotions and interactions of an infant. Kismet and its siblings have still a long way to go, certainly as far as learning and understanding are concerned. But one thing is clear: if we really want to understand what it means to be a social being, a conscious being and an active being, building sociable and interactive robots is one of the most promising paths.