[ExI] alpha zero

[spike]

-----Original Message-----
From: extropy-chat [mailto:extropy-chat-bounces at lists.extropy.org] On Behalf
Of Alejandro Dubrovsky

>>... I have a theory which exonerates DeepMind of any attempt at deception,

> a theory which I think is most likely the right one.
...
>>... The DeepMind guys aren't really specifically chess guys, they are 
> programmers.  So they aren't specifically trying to necessarily create 
> the top chess program, but rather demonstrate a paradigm where 
> software can teach itself, given a clear end goal... spike

>...I looked them up: that Dharshan Kumaran that appears in the paper's list
of authors appears to be a brittish GM (photos match up) and one of the
founders of the company is a 2200. Relatively sure this is real.
_______________________________________________


OK, well then, a grandmaster and a 2200 would know how large a handicap it
would be for top level software to be programmed to eschew draws if that is
what was done.

The way we will know if this is real is if Deep Mind lets AlphaZero compete
soon in public under controlled third-party observed conditions.  If I
understand it correctly, the machine learning part of the experiment created
the software, so they could allow the software to compete without revealing
the code.  If it really is playing in the same league with the big guys (and
beating StockFish where someone outside the company controls the SF
settings) then I too join the ranks of true believers.

I am very cautious now, for a truly desperately want it to be true.  I can't
quite shake the suspicion that all isn't as it appears with that oddball
result, 100 games with no losses.  Even in the drawn games, Alpha didn't
appear to be playing into the well-known drawish lines much.

Even if there is something amiss, I will declare that this collection of
games includes some really exciting chess.  There were plenty of beauties in
there.  In a way that also kinda raises suspicion: why aren't there several
boring old Ruy Lopez snoozefests?  Humanity discovered that opening back as
soon as the rules were solidified in the early 1500s, but this learning
software didn't seem to do much with it.

spike