[ricraz]

As an AI researcher who did a degree in philosophy, I think most of the things you mentioned are pretty irrelevant. How to build an AI with "feelings, sensations, etc" is indeed a mystery, but we don't need to aim for that, we just need to aim for intelligence, which can be defined without reference to consciousness or qualia. Similarly, if we can invent a working, fully automated Chinese room that passes the Turing Test, then whether or not it fits Searle's definition of "understanding" is a moot point (especially since his understanding of "understanding" is pretty weird).

More generally, although we should be heavily inspired by human intelligence when designing machine intelligence, it's a mistake to use the way humans think to define intelligence. Kuhn's account of scientific revolutions, for example, is primarily descriptive, not prescriptive. We can certainly imagine possible setups where science doesn't proceed like that, which may well be superior. Science isn't defined by revolutions, but by experimentally searching for the truth. In the same way, knowledge isn't defined by having a body, but by having beliefs which correspond with the state of the world.