We are doomed - AlphaGo Zero, learning only from basic rules

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
IanO
Posts: 481
Joined: Wed Mar 08, 2006 8:45 pm
Location: Portland, OR
Contact:

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by IanO » Fri Oct 27, 2017 2:34 pm

A thing I saw fly by my news feed is that Gian-Carlo Pascutto (author of Sjeng) has extended his own Go program Leela with the results of the paper, releasing it on GitHub as an open source program Leela Zero.

https://github.com/gcp/leela-zero

Of course, it isn't as strong as AlphaGo Zero because it doesn't have Google's training resources, but they have plans:
One reason for publishing this program is that we are setting up a public, distributed effort to repeat the work. Working together, and especially when starting on a smaller scale, it will take less than 1700 years to get a good network (which you can feed into this program, suddenly making it strong). Further details about this will be announced soon.

yolin
Posts: 30
Joined: Thu Mar 30, 2006 4:12 pm

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by yolin » Thu Nov 23, 2017 1:58 pm

Gian-Carlo's project is generating a lot of interest:
http://zero.sjeng.org/

Also this adaptation to chess just got posted:
https://github.com/gcp/leela-zero/issues/155
https://github.com/Zeta36/chess-alpha-zero

MikeGL
Posts: 943
Joined: Thu Sep 01, 2011 12:49 pm

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by MikeGL » Thu Nov 23, 2017 3:19 pm

duncan wrote:
...the resulting breakthroughs have the potential to positively impact society.
History already taught us that the latest tech are always used first by military for espionage, reconnaissance and other things related to destruction. Developing the society was not the top priority.
Rockets, computers, encryption, submarines, internet, airplanes, satellites, radar, and many others, were developed by and for military.

But I am optimistic, maybe times have changed,
maybe society will be the first to benefit from latest tech on this day and age.

.

corres
Posts: 3065
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by corres » Thu Nov 23, 2017 10:42 pm

The rules of Go are very simple so these are very advantageous for a supercomputer using similarly simple logic. Only its speed is the factor what
promises victory for the machine. But the world is more complicated and it can not be transcribed with formal logic and two bits mathematics.
So I think it is very pessimistic to believe we are doomed now.
In the far future maybe this will be the case.

corres
Posts: 3065
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by corres » Fri Nov 24, 2017 6:40 pm

[quote="Michel"]

Note that the aim should be to have an eval which is _much_ better than the one of Stockfish, to offset the fact that NN are inherently slower.

[/quote]

Why Stockfish?
Komodo has a better and more complex evaluation than Stockfish.
The power of Stockfish arises from its higher speed and not from the better evaluation.

User avatar
Laskos
Posts: 10314
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by Laskos » Fri Nov 24, 2017 7:29 pm

To reply to the thread title:
We are doomed - AlphaGo Zero, learning only from basic rules

Sure, humans are doomed, at Chess and Go, and many other things. Observing carefully my cats, and remembering my dogs from when I was a teenager, I think their doom is so far in the future, and even asymptotically to infinity prone to a logical definition, that I am not that concerned thinking of this "doom".

We were doomed since we have learned that tertium non datur. Cats and dogs hardly follow any of these human arrogances, which prove to be no more than imitations of an ideal diode.

User avatar
reflectionofpower
Posts: 1461
Joined: Fri Mar 01, 2013 4:28 pm
Location: USA

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by reflectionofpower » Fri Nov 24, 2017 8:12 pm

Laskos wrote:To reply to the thread title:
We are doomed - AlphaGo Zero, learning only from basic rules

Sure, humans are doomed, at Chess and Go, and many other things. Observing carefully my cats, and remembering my dogs from when I was a teenager, I think their doom is so far in the future, and even asymptotically to infinity prone to a logical definition, that I am not that concerned thinking of this "doom".

We were doomed since we have learned that tertium non datur. Cats and dogs hardly follow any of these human arrogances, which prove to be no more than imitations of an ideal diode.
Nicely said.
"Without change, something sleeps inside us, and seldom awakens. The sleeper must awaken." (Dune - 1984)

Lonnie

Uri Blass
Posts: 8771
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by Uri Blass » Fri Nov 24, 2017 10:28 pm

corres wrote:
Michel wrote:
Note that the aim should be to have an eval which is _much_ better than the one of Stockfish, to offset the fact that NN are inherently slower.
Why Stockfish?
Komodo has a better and more complex evaluation than Stockfish.
The power of Stockfish arises from its higher speed and not from the better evaluation.
We do not have source code of komodo so no way to test your theory.

corres
Posts: 3065
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by corres » Sat Nov 25, 2017 7:59 am

O.K.
But the evaluation of a chess engine depends not only on the static parameters but dynamic behaviors such as move generation, prunings etc.
too.

Dirt
Posts: 2851
Joined: Wed Mar 08, 2006 9:01 pm
Location: Irvine, CA, USA

Re: We are doomed - AlphaGo Zero, learning only from basic r

Post by Dirt » Sat Nov 25, 2017 9:10 am

reflectionofpower wrote:
Laskos wrote:To reply to the thread title:
We are doomed - AlphaGo Zero, learning only from basic rules

Sure, humans are doomed, at Chess and Go, and many other things. Observing carefully my cats, and remembering my dogs from when I was a teenager, I think their doom is so far in the future, and even asymptotically to infinity prone to a logical definition, that I am not that concerned thinking of this "doom".

We were doomed since we have learned that tertium non datur. Cats and dogs hardly follow any of these human arrogances, which prove to be no more than imitations of an ideal diode.
Nicely said.
Really? What did he just say? Something about dogs and cats and rectifiers...I didn't understand any of it.
Deasil is the right way to go.

Post Reply