Seed Software
A short while ago, I commented on Kevin Kelly debunking Kurzweil’s singularity, citing the fact that an all-powerful AI can’t simply infer solutions to all our problems through Cartesian means. Interestingly enough, another kk post, The Forever Book, hypothesizes writing a basic text from which a newcomer to a subject could begin the process of recreating all knowledge in it from scratch. In this way, the book would become a replicating meme, a knowledge seed from which a whole body of knowledge could grow.
 Mustard Seed Credit: echoesofstars |
I keep thinking of these knowledge seeds in terms of computer programs. Books and DVDs can present nice algorithms to start human beings and civilizations off on a lifetime of reconstructing knowledge about the world, but a computer program can perform the reconstruction far more quickly. The only drawback is that the computer could only do this for subjects discoverable through Cartesian rather than Newtonian methods.
Assuming someone had a computer at their disposal and a Seed Software for the subject of Mathematics, they could set the program running to discover all mathematical knowledge and beyond. The program would come up with its own names, in the form of identification numbers, for every mathematical statement(?) (ie. Pythagoras’ Theorem would be “91254987”), and would not value one statement over another (ie. Pythagoras Theorem would be as important as “2 + 2 = 4”).
It would be fascinating to see how the knowledge would unfold. Unlike humans, computers would have no bias toward two or three dimensional thinking, and could run away discovering mathematical concepts of little use to the human user.
Somebody needs to get Stephen Wolfram working on this. : )
Question: Wont a proper implementation of Quantum Computing greatly change the playing field?
Comment by ClintJCL — January 15, 2009 @ 2:43 pm
I have a problem with this:
“The program would come up with its own names, in the form of identification numbers, for every mathematical statement(?) (ie. Pythagoras’ Theorem would be “91254987?), and would not value one statement over another (ie. Pythagoras Theorem would be as important as “2 + 2 = 4?).”
It assumes an infinite amount of storage. Items are only named if they need to be recalled and if storage isn’t infinite then it would need to be prioritized. Else how would your AI know what to keep and what to discard when it hit its storage limits?
-BMF
Comment by BMF — January 15, 2009 @ 6:10 pm
I actually think it would value one statement over another, too. The basis of this knowledge is that knowledge is built on other knowledge. To do that, you need to recall certain elements. Obviously some theories will use more prior elements than others; for example physics will use calculus, calculus will use algebra, algebra will use arithmetic, etc. I would think priority would be assigned to this bits of knowledge that are most often referenced, and to which the most things depend on. Whatever creates more dependencies would be more priorities.
This would also speed up computation — when considering new input with respect to everything else that is currently known, it would make sense to think about those things that most often come into play first, and then to consider the more obscure facets of knowledge later. (Of course, we don’t want to miss super-obscure discoveries, so I propose an additional element of randomness.)
Comment by ClintJCL — January 15, 2009 @ 10:06 pm