[extropy-chat] Bluff and the Darwin award

[Eliezer S. Yudkowsky]

Russell Wallace wrote:
> On 5/17/06, *Eliezer S. Yudkowsky* <sentience at pobox.com 
> <mailto:sentience at pobox.com>> wrote:
> 
>     Russell, your argument also proves that Deep Blue's programmers can't
>     make changes to Deep Blue that improve its playing ability without
>     actually testing it in many games against Kasparov.  One game can't
>     possibly be enough to distinguish between potential changes because the
>     win or loss only provides one bit of information.
> 
> Actually, I've seen people who work on grand master level chess software 
> comment that not only do they have to test putative improvements in 
> actual chess games, but that it helps a great deal to have a human 
> expert on hand to point out the strengths and weaknesses in the 
> program's play.

This is why the ability of the world's most powerful chess players has 
been decreasing slowly over time; you can't learn chess except by 
playing against a better player and getting advice from better players. 
  Only the environment can teach you; there's no way to learn 
internally.  Thus, over time, more and more of the ancient chess 
techniques are lost, degraded from signal to noise.  Kasparov is 
probably barely as powerful as the average aristocrat of the seventeenth 
century.  And Deep Blue, which beat Kasparov, was programmed by aliens.

-- 
Eliezer S. Yudkowsky                          http://singinst.org/
Research Fellow, Singularity Institute for Artificial Intelligence