Re:AlphaGo Is Not AI
by chejazi on Sunday March 19, 2017 @04:47PM
(#13909539)
OK, so we have AIs that are capable of solving a very specific problem. My questions is, are AI's capable of being composed? In the sense that if I have 1000 AIs that are each experts at performing their domain specific tasks, can one or more AIs be built on top of them that specialize almost exclusively in delegation? i.e. they delegate intelligently depending on the task at hand. If that's possible, then we could in theory build a generalized AI, right?
[ Parent
| on HN ]
A simple decision tree could suffice. Is it a speech problem, is it an image problem? Is it a classification problem? Failing to find an acceptable answer, it does what seems closest or makes a random but plausible decision. This doesn't seem too far from what humans would be doing. Part of the problem is the meta-recognition of the problem and how to organize that within an actual system.
It's not good to abuse of the analogies, but the human brain have to spend more time in task with conflicting information, for example to read the word "yellow" when it's written with blue ink, ... More details: https://en.wikipedia.org/wiki/Stroop_effect