Robots and Avatars as Teachers and Learners
(Lew's Review of Information in Cyberspace)
by Lewis Berman
Robots are starting to take an active role in teaching children and learning from them as well.
A robot as your child's teacher? The New York Times recently published a Science article entitled Students, Meet Your new Teacher, Mr. Robot. A three foot tall humanoid robot at the University of Southern California maintains eye contact with an autistic six year old who imitates its motions. The robot can detect periods of attention vs. inattention and adapts accordingly, reflecting the integration of sensor processing, artificial intelligence, and end-effector control. Another robot, RUBI, at the University of California, San Diego, teaches foreign languages. Unlike their human counterparts, these robot teachers are infinitely patient, and they are beginning to incorporate learning algorithms that can help customize instruction to the individual student. Another robot that can teach through social interaction is Simon at the Georgia Institute of Technology's Socially Intelligent Machine Lab.
The LIFE NSF Science of Learning Center is acting upon the hypothesis that "humans, including children, learn best from entities that are socially responsive and interactive." They are applying this "biologically plausible model - child development" to robotics, having assembled an interdisciplinary team of engineers, computer scientists, and child development researchers to design "social robots," which can perform such learning by observing others. Their probabilistic framework for robotic learning by imitation has been demonstrated to work in noisy (real-world) environments.
The same LIFE NSF Science of Learning Center has performed a study in which they found that students are more motivated to do homework when they do it by having an avatar in a virtual environment. They learn by teaching their avatar, which is actually part of a computer program called a Teachable Agent. Then their agent competes against other students' agents in a virtual game show, answering questions posed by the show's automated host. As a sidelight, a separate study, not especially computing-related, a few years back by Sandra Okita of Stanford, has shown that teaching someone to solve a problem, then observing them solve the problem, leads to better learning than teaching and then doing the same problems yourself.
The computing-related examples above illustrate the notion that a modern-day application-oriented computer researcher or software engineer must gain mastery of multiple areas of interest. It was once sufficient to specialize in artificial intelligence, sensor fusion, robotic motion control, distributed systems, databases, or educational software. Truly there is still a need for deep knowledge of each of those areas, but someone with cross-knowledge has to pull the project or team together. As these systems interact more and more with humans, software safety and reliability will also come more and more out of the avionics world into the everyday world, as will all the major precepts of Software Engineering.
It's time to re-read Asimov!
(Pictured: Simon, Georgia Tech Socially Intelligent Machine Lab)
Lewis Berman is the Director of Program Operations for the Computer Science and Software Engineering graduate programs.