Siri and AI

Being involved with the AI class and being a career programmer I am very well aware that the machinery behind Siri is nothing but a computer system. However, when one interacts with the computer system using a voice interface that understands so well as Siri does, one is almost obliged to be polite. I found myself automatically saying please or thank-you in the same way as I would if a human being were performing a service on my behalf. Interestingly Siri responds with things like "I'm just doing my job" or "I exist to serve you."

I meet a lot of people in my work and I have to interact with folks from many cultures. Some have different rules for what is polite, what is considered as necessary and what is completely wasteful of the effort of speaking. English children of my generation were taught that "manners maketh man" and were most often brought up with the idea that one should go through the motions of being polite even if you loathed the person with whom you were dealing so the habit of being polite for me has extended to a machine intelligence that seems to be more than it really is.

I spent a while watching a presentation given by Dr Sebastian Thrun who teaches part of the AI class. In the presentation he showed off the Google driverless car and was explaining some of the aspects of the AI system. The car stopped several times for people who had crossed late at a pedestrian crossing and on a couple of occasions, the person gave a little thank-you wave to the car. Obviously the car didn't care but this begs the question of whether it should or not.

Reinforcement learning in AI could very well take notice of the user’s approval as a cue for better understanding the intent of the user or the people with which it interacts. It seems obvious to me that this would work out well but when faced with someone who was mentally ill, aberrant or just plain obtuse a machine system that takes cues from its users could potentially become very weird indeed.

Isaac Asimov invented three clear and precise laws of robotics which many people make the mistake of considering for real robotic systems. They are at first glance precise and unambiguous with seemingly no nuances but we should also remember that Asimov's genius was creating such a set of strong rules with the express intention of finding ways in which the robots in his stories could misinterpret them. I believe that as machine intelligence becomes more competent and more capable of interacting with the human race, the auditing systems that will be needed to ensure that the system doesn't become twisted will be more difficult to create than the intelligent agents themselves.