Robert Jasiek <jas...@snafu.de> writes: > So there is a superstrong neural net. > > 1) Where is the semantic translation of the neural net to human theory > knowledge? > > 2) Where is the analysis of the neural net's errors in decision-making? > > 3) Where is the world-wide discussion preventing a combination of AI > and (nano-)robots, which self-replicate or permanently ensure energy > access, from causing extinction of mankind?
As far as (1), if we could do it, it would mean we could relate the structures embedded in the net's weight patterns to some other domain -- if nothing else, the domain of the meanings of words in some natural language. We cannot, and most certainly the net cannot. So far neural nets have been trained and applied within single domains, and any "generalization" means within that domain. A net may learn to recognize and act similarly with respect to a certain eye pattern on different parts of the board. No one, as far as I know, has presented a net that would be able to use a guideline like, "two eyes alive, one eye dead" to help it speed learning of how to act on the board. But a human can apply "one"/"two", and "alive"/"dead" once it has been made clear that "eye" in this context is standing for a recognizable structure of same-color-surrounding-space, and thereby learn in one step what the net learns in thousands of incremental iterations. And (2) presupposes (1), since to understand why a situation was mis-perceived or mis-acted upon requires some understanding of what exactly the perception and judgment process was in the first place. It may be that the recent successes in brute-force learning powered by improved hardware together with improved crafting of the architecture eventually play some role in understanding and recreating "intelligence". But so far, using the term "AI" in connection with 99% of this kind of work is just hype. Useful in accomplishing engineering goals, yes. But not so much to do with intelligence. _______________________________________________ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
