My response will be to the talk given by Professor Alan Bierman. I found the system to assist a human fix a circuit board most interesting. The system asks a question, and the user is expected to answer. The user has the choice of either answering or possibly changing the subject. The algorithm is supposed to determine whether it should follow the subject of the speaker, or not follow. If it decides not to follow, it might repeat the question in a louder and more firm tone. This made me think of assertiveness training. Like humans, maybe some are naturally assertive enough, and others need to be made more assertive? How assertive should the computer be? Perhaps the computer could be programmed to override the commands of humans. If the computer knows what is best, then maybe it should override the human?
This scenario calls to mind a scene from a classic science fiction movie: 2001 - A Space Odyssey. In the movie, there is a space ship that has a very "intelligent" computer - called HAL 2000. There is one human, named Dave, on board that wants to take a walk outside. The interests of the computer and the human conflict, and so HAL does not want Dave to go outside, and refuses to open the doors. I wonder if this is where computer assertiveness could lead?
Are computers just tools, or are they more? What do we want computers to do? I would suggest that we want them to help us do things that are tedious or difficult for humans to do. I would argue that computers are just tools to help humans, and not to think for us. Artificial intelligence is only useful to us to the extent that it furthers our interests.
If I am talking to a computer, and I change the subject, I think it would annoy me if the computer continued to ask the same question. If the objective is to train a monkey to fix a circuit board, then that is another matter - a monkey would need to be constrained. A human, if he understands the basics about the system, would not change the subject without a reason. I think the only reason that the computer in this case does not follow is because it can't. They don't know how to make the computer understand the human when the subject changes, and there is no context information. It is mereley a limitation of the system that it is unable to understand a statement without context.
This made me think about the limitations of computer understanding, and how far we have to go to get to the point that a human cannot distinguish between interacting with a computer or a human (the Turing Test). There is little doubt that artificial intelligence is a viable field that has made many contributions to society. It remains to be seen whether a computer will be able to approach (or exceed) human intelligence. My expectation is that computers will be able to mimik human behavior very closely in many ways. I don't know what the limits to computer behavior will be. If the brain is thought of as purely a neurological/chemical computer, with a finite capacity, then it would be possible for a computer to duplicate human intelligence.