Re: A possible limitation for a SI

Anders Sandberg (
24 Aug 1998 00:06:41 +0200

"maxm" <> writes:

> I think that any SI will be limited mostly by this factor. How to do a thing
> is pretty straightfoward in most cases but choosing WHAT to do (that big
> bussines idea), so that you really stand out from the rest of the pack. That
> is hard!

Good point!

Where real intelligence shows itself is in this area. True intelligence consists of noticing the patterns that allow them to partially pierce the "probability cloud". A chess master will not just know about the possible moves, he will also judge the opponent and select the gambit most likely to make the opponent make a mistake.

So the problem isn't the combinatorial explosion of possibilities, it is discovering the patterns. This is a hard problem, and most likely is linked to the bias-variance dilemma in learning theory (it is too late for me to think clearly, but I think they are linked). It may be that an SI will need so extremely large amounts of data to predict the important stuff that in reality it will have to settle for educated guessing not that much more advanced than our own.

Anders Sandberg                                      Towards Ascension!                  
GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y