Quote:
Originally Posted by peter12
To me, right now, it is akin to IQ testing. We know that you can practice the heck out of IQ problems, but when you take an actual test comprised of secret answers, you fare slightly or no better than someone who didn't practice at all. We aren't creating machines that are able to spread their intelligence across multiple tasks, and this, of course, is the essence of real artificial intelligence.
The question remains, does knowledge about intelligence fall into the realm of diminishing returns? Is it like so many things where the more we know, the less we are able to know.
|
This is an interesting take, and I think maybe the real issue is whether AI will be able to make connections between different systems and apply techniques from one to another. Another way of looking at it is if an AI can generalize a problem - see one problem as a class of a known solved problem, then apply a solution.
I don't know what research has been done in this area, but if you think about how people have cleverly mapped one problem onto another, I think it's a doable proposition. Think about how VW Beetles used the tire pressure of the spare tire to power the windshield wiper fluid mechanism. Could an AI have deduced that this system requires a pump, that a pump is a system that produces pressure, and that a tire could be considered a pump? If anything, maybe an AI comes to that solution faster than a human because it is not restricted by the bias that a tire can only be used to roll things?
Another way to look at it is that intuition is simply one (often efficient) choice among all possible choices, and the question is whether an AI can efficiently brute force the right solution. Is it unreasonable to think that those eureka moments where you wake up in the middle of the night with a solution to a problem is really your brain subconsciously brute forcing that solution. Because it is subconscious, you don't feel that you've been actively mulling it over.