Author: Dann Corbit
Date: 12:21:56 10/27/98
Go up one level in this thread
On October 27, 1998 at 14:52:17, Ed Schröder wrote: [snip] Fernando: >>And what about some kind of changeable tables with RAM memory? The exe is kept, >>but some variable data to exceute could change... >>Fernando Ed: >Theoretical this is possible. I never tried because I never came up with a good >plan. Some have tried. Alex van Tiggelen is one. His program is commercial. I >forgot the URL of his home page but maybe others know. To run new instructions in memory simply means that either you write self modifying code, or you write an interpreter. Neither of these is very well done at the moment. Self-modifying code is impossible to maintain, and nearly impossible to understand. I once had a Fortran program with computed labels. What a nightmare. I recoded the darn thing in PL/1 and it was orders of magnitude faster (change of algorithm). Interpreters are slow. Even JIT compilers simply cannot be made as fast as precompiled code. None of this answers the big question: "How will I write code that becomes smarter as it runs?" If we knew how to make it smarter, why didn't we just make it smarter in the first place? Neural nets are one possibility, but quite frankly are in their infancy. Besides which, we have to emulate them in software. If we had NN chips with one million neurons and used a million of them, they would be competitive. Neural approaches require phenomenal power. Did you know the brain of a honeybee does ten billion calculations per second? (Not bad for a few micrograms of nectar as energy input). Eventually the more esoteric apporoaches of AI [such as neural nets] *will* supplant traditional techniques (and be used in conjunction with them.) I predict that this is several years off, probably at least a decade. Hans Morovic wrote a very interesting book on AI. Anyone interested in such topics should get a copy or at least check it out from the local library.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.