Nine Things You Should Know About AI

I've been involved in things originally called "AI" for most of my "career", such as it was. I've worked in a lot of, as they were called at the time, Artificial Intelligence projects. Of course, once they moved into the implementation phase they were no longer AI. For example, I worked on natural language understanding, robots, machine vision, neural networks, expert systems, speech recognition, touch sensing, self-reproducing systems, path planning, assembly planning, malware detection, computer algebra, knowledge representation, and human-robot interaction.

Russell states: "I want to be clear that we are a long way from achieving general purpose AI".

I want to be clear that anyone who talks about "general purpose AI" is failing to define their terms clearly. In fact, I suspect that any definition would exclude most, if not all, humans.

Rod Brooks is probably closest when he discusses the idea of learning at the 1-year, 2-year, 5-year, etc. levels of self-learning.

When you hear the phrases "Artificial General Intelligence" or "General Purpose AI" you should insist on a clear, exact definition. Otherwise the discussion is about nothing, basically pop-philosophy.

Russell defines "intelligence" as "machines that achieve their objectives". My toaster certainly achieves its objective of making toast. It even uses feedback from a sensor to know when it has achieved the goal.

It is not the achieving of objectives that motivates me. It is the defining of objectives, otherwise known as choosing research goals. Usually I find that the original research goals change, grow more complex, and spawn yet more goals. It is the "side-effect" of learning that motivates my efforts, not the achievement of goals.

It is, of course, open to debate whether I'm a "general purpose intelligence". I would be the first to argue that I am not. There are areas, like writing good music, which despite my years-long best efforts are a complete failure. I can't understand why I fail.