The 96th Installment
Good Teacher Design for Good Robots

by Hiroshi Hashimoto,
Professor, Master Program of Innovation for Design and Engineering

Robotics is a Japanese technology renowned throughout the world. Robot creation is based upon integration technology incorporating electromechanical systems, IT design, and styling. Additionally, the following must be mentioned.

Important features common to all robots is their ability to move around. They can grasp, pick up, and carry objects from one location to another. They are also capable of non-verbal communication based on human gestures. These abilities have been the subject of extensive research and development not only for industrial purposes but also for nursing care and welfare facilities as well as various customer service situations.

Here let’s consider what makes for a good robot. There are many ways to approach this question, but we now focus on “good robot behavior” based on its own good judgment. In recent years, artificial intelligence (AI) has been proven many times over to be an excellent tool and therefore is claimed to be indispensable to good robot performance.

AI has the ability to learn and therefore make prudent decisions. A good teacher (or good learning data) will produce good AI. A malicious teacher (or learning data created with malicious intent) will result in malicious AI. The same can be said for human growth. For both humans and AI, meeting good teachers will be critical to their development as well as their future life—perhaps it should be called AI service life—following the initial encounter.

Yet I seldom see arguments based on “What is a good teacher?” More specifically, media coverage of AI is limited to its successes with no attention paid to its failures. There are many failures in AI research stages. This is partially because the research is mainly focused on the development of learning algorithms. It rarely addresses the question of what makes a good teacher.

The study of “good teacher design” is indispensable. It is important that we expand this academic discipline as well as encourage relevant discussion between those involved. So what is your idea of good teacher design? There are many criteria and many approaches depending upon area and intended use. I want to consider it from the perspective of symbiosis as it relates to the Three laws of Robotics.

The term “symbiosis” as used here refers to the relationship between humans and robots. One condition for this symbiosis is that the relationship be a long one spanning 6 months, a year, or several years. The relationship itself must be preserved while the quality of human life is maintained or improved. Therefore, a robot possessed only of Japanese style cuteness will grow tiresome in a short time and the symbiosis will not be maintained. In the same way, a robot tasked solely with providing high quality services will be looked upon as a convenient machine, once again breaking the symbiosis as people are likely to direct their attention towards other new things.

Dogs have lived in harmony with humans longer than any other animal. They respond to human psychology and behavior in various ways showing both joy and anger with gestures and facial expression. They also squeal, snarl, cower in fear, snuggle, crouch, and even show empathy in response to a situation. This serves as a good example of symbiotic behavior in that it can be both positive and negative. A good teacher’s design must take into consideration behavior with both positive and negative factors.

Viewed from the perspective of Isaac Asimov’s Three Laws of Robotics that appeared in his book I, Robot (1950), we learn that these principles cause a frame problem. The three laws are listed below.

• The robot must do no harm to humans

• The robot must obey commands

• The robot must protect itself

These three laws are sometimes difficult to apply to a robot. Look at the two following examples.

The movie Terminator 2: Judgment Day (USA, 1984) ended with the hero robot destroying itself in the end. This is because of the risk that it might threaten mankind in the future. Yet this self-destruction violates the three principles of robotics.

A similar example is the seeing eye dog. The dog is capable of intelligent disobedience. When its blind master commands it to go in a certain direction, the dog looks at the situation and will disregard the command if it senses something that will put its master in jeopardy such as an open hole or an approaching car. Suppose you replace the dog with a robot. The robot must therefore be designed in such a way that ‘good behavior’ can sometimes mean not following instructions. Here I do not wish to imply that Asimov’s Laws are somehow defective nor am I talking about absolutes of good and bad behavior. I am just pointing out situations in which the principles must either be obeyed or modified. Which way does one go?

This article barely scratches the surface of ‘good’ robot behavior and how we think about teacher design necessary to realize this goal through the examples of symbiosis and the three laws of robotics. Good behavior is complicated and far from straight-forward and good teacher design is therefore even more complex. I firmly believe that this is something that deserves more attention here in Japan. The issue has been raised but concrete methodology has yet to be developed. This is what I wish to consider with my readers.