Artificial Intelligence: The Human Machine

My favorite American entrepreneur, Elon Musk, who is very much the Thomas Edison of our time, has helped start a new enterprise aimed at advancing the field of artificial intelligence. It's called "OpenAI". On their website they state:

OpenAI is a non-profit artificial intelligence research company. Our goal is to advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return.

Since our research is free from financial obligations, we can better focus on a positive human impact. We believe AI should be an extension of individual human wills and, in the spirit of liberty, as broadly and evenly distributed as possible.

An Ivory Tower?

The phrase "unconstrained by a need to generate financial return" is troubling. I'm all for non-profit. I really am. But the question arises, what is going to motivate this organization to produce anything? There's a very real risk of ivory-towerism.

Which would be a shame. So much talent and resources wasted in that case. It seems that their first priority needs to be to set measurable goals that they can evaluate themselves against. A couple of things come to mind:

• Number and quality of peer-reviewed papers.
• Number and quality of users of their planned open source software.

What is the question?

Before you can come up with an answer, you need to know the question. Before Einstein could come up with special relativity, he needed a clear statement of the problem that he was trying to solve. In retrospect, coming up with the question was perhaps the hardest part and took many people working many years. Without the Michelson-Morley experiment in 1887, for example, Einstein would not have known what the problem was he was trying to solve in 1905.

So instead of trying to come up with answers before knowing the questions, perhaps the focus needs to be more on coming up with the questions. In the context of AI, this paper from the Proceedings of Aslib Informatics provides some insight.

The human machine.

One possible way to generate the questions as well as to test progress would be to build what may be called a "human machine". This would be a platform consisting of networked software that lets a group of humans work together to interact with a human interrogator. Someone who wants to chat would go to a web site, and the other side of the conversation would be handled by the platform.

This is a difficult problem in itself, even with no AI plugged in. In one possible architecture, a single person would manage the platform side of the conversation, and would create sub-problems to be distributed to other humans to solve. How you do that, and what is the architecture, even with no AI involved, is a hard problem. Once you had such a platform, you could begin adding AI features to make it faster and better, with the goal of ultimately replacing all the humans.

The point of such an experiment would be to focus on the appropriate questions as well as to serve as a measure of progress, in terms of how satisfying the interaction is to users. In the absence of financial or military motivations, some method is needed to map the space being researched and to know if you're going anywhere, whether it's a "human machine" or something else.