Historic Milestone: AlphaZero

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

Moderators: hgm, Harvey Williamson, bob

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Historic Milestone: AlphaZero

Post by Lyudmil Tsvetkov » Sat Dec 09, 2017 3:03 pm

Jouni wrote:Did this published paper give any estimate for perfect chess? One figure has 3500 limit or +100 from SF8. Are we near? Where is Kai, when we need him :) ?
They are not in a position to give that estimate, as too many factors are unknown.

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Much weaker than Stockfish

Post by Lyudmil Tsvetkov » Sat Dec 09, 2017 3:05 pm

fern wrote:" just that outcalculates Stockfish consistently...."

Well, that is enough, pal.

Fern
Sufficient to beat SF, but not to prove the match was fair.

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Historic Milestone: AlphaZero

Post by Lyudmil Tsvetkov » Sat Dec 09, 2017 3:07 pm

Laskos wrote:
Jouni wrote:Did this published paper give any estimate for perfect chess? One figure has 3500 limit or +100 from SF8. Are we near? Where is Kai, when we need him :) ?
I started following yesterday this development, I have no much time recently.

About your question. First, to define perfect chess. Say, perfect chess in the sense of non-swindling or fooling around, 32-men tablebases.

I had a wide range of estimates, from 3700-4500 CCRL 40/40 rating points, and with the latest data of Andreas Strangmüller tipping to the lower bound, say 3800.

With this result, +28 =72 -0 (is it correct, or I missed some more results?), the tip is again towards a low bound in ELO. a +36 =56 -8 result would tip towards a higher ELO bound, although the ELO advantage of AlphaZero would have been the same over SF8.

It seems both these monsters play very often perfect moves, and sometimes maybe even perfect games. So, it seems the ELO bound cannot be very far away.
I would not agree here, perfect chess is over 5000 elo, that is completely certain.

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Historic Milestone: AlphaZero

Post by Lyudmil Tsvetkov » Sat Dec 09, 2017 3:08 pm

That poor thing, called Alpha, still achieves whatever rating it has achieved, playing 1.d4.
Of what perfect chess can we talk then?

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Historic Milestone: AlphaZero

Post by Lyudmil Tsvetkov » Sat Dec 09, 2017 3:10 pm

Leo wrote:
Laskos wrote:
Jouni wrote:Did this published paper give any estimate for perfect chess? One figure has 3500 limit or +100 from SF8. Are we near? Where is Kai, when we need him :) ?
I started following yesterday this development, I have no much time recently.

About your question. First, to define perfect chess. Say, perfect chess in the sense of non-swindling or fooling around, 32-men tablebases.

I had a wide range of estimates, from 3700-4500 CCRL 40/40 rating points, and with the latest data of Andreas Strangmüller tipping to the lower bound, say 3800.

Again, this result, +28 =72 -0 (is it correct, or I missed some more results?), the tip is again towards a low bound in ELO. a +36 =56 -8 result would tip towards a higher ELO bound, although the ELO advantage of AlphaZero would have been the same over SF8.

It seems both these monsters play very often perfect moves, and sometimes maybe even perfect games. So, it seems the ELO bound cannot be very far away.
Alpha Zero won only 3 games with black.
Because they were so close in rating, while the computing power was enormous, so relatively few mistakes.

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Historic Milestone: AlphaZero

Post by Lyudmil Tsvetkov » Sat Dec 09, 2017 3:12 pm

IanO wrote:
corres wrote:2, Alpha Zero did not start from zero knowledge about chess
because it was [fed] a lot of human games at start up.
Where did you see that? My impression is the opposite, they explicitly did not use any prior human chess knowledge besides the basic rules, hence the "Zero" in AlphaZero.
They did, that is why Alpha almost always played Bg2.
You can not learn that in any other conceivable way, unless you follow human theory.

Lyudmil Tsvetkov
Posts: 6033
Joined: Tue Jun 12, 2012 10:41 am

Re: Historic Milestone: AlphaZero

Post by Lyudmil Tsvetkov » Sat Dec 09, 2017 3:16 pm

shrapnel wrote:
Jhoravi wrote:I wonder if neural chess for PC is competitive against the best engines using this latest hardware

https://www.anandtech.com/show/12135/nv ... 00-dollars
A computer chess program which can harness the Power of the GPU !
My dream come true !
Hmm... a bit on the expensive side, but I might just be able to afford it.
Now all I need is Google to start selling the Alpha-Zero Program.
We have good offers, starting from $250 000. :)

zenpawn
Posts: 260
Joined: Sat Aug 06, 2016 6:31 pm
Location: United States

Re: Historic Milestone: AlphaZero

Post by zenpawn » Sat Dec 09, 2017 3:37 pm

Lyudmil Tsvetkov wrote:
IanO wrote:
corres wrote:2, Alpha Zero did not start from zero knowledge about chess
because it was [fed] a lot of human games at start up.
Where did you see that? My impression is the opposite, they explicitly did not use any prior human chess knowledge besides the basic rules, hence the "Zero" in AlphaZero.
They did, that is why Alpha almost always played Bg2.
You can not learn that in any other conceivable way, unless you follow human theory.
From the paper, "Starting from random play, and given no domain knowledge except the game rules,...". And: "The AlphaZero algorithm is a more generic version of the AlphaGo Zero algorithm that was first introduced in the context of Go. It replaces the handcrafted knowledge and domain-specific augmentations used in traditional game-playing programs with deep neural networks and a tabula rasa reinforcement learning algorithm."

shrapnel
Posts: 1107
Joined: Fri Nov 02, 2012 8:43 am
Location: New Delhi, India

Re: Historic Milestone: AlphaZero

Post by shrapnel » Mon Dec 11, 2017 8:25 am

zenpawn wrote:From the paper, "Starting from random play, and given no domain knowledge except the game rules,...". And: "The AlphaZero algorithm is a more generic version of the AlphaGo Zero algorithm that was first introduced in the context of Go. It replaces the handcrafted knowledge and domain-specific augmentations used in traditional game-playing programs with deep neural networks and a tabula rasa reinforcement learning algorithm."
Tsvetkov and Szabo have Proof that DeepMind was lying.
They will soon show us.... :roll:
i7 5960X @ 4.1 Ghz, 64 GB G.Skill RipJaws RAM, Asus ROG Strix 11 GB Geforce 1080 Ti and AMD Ryzen 7 1800X @4.0 GHz, 32 GB DDR4-2400 G.Skill RAM, ASUS Prime x370-PRO, Noctua NH-D15 SE-AM4 Cooler.

User avatar
hgm
Posts: 22184
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Contact:

Re: Much weaker than Stockfish

Post by hgm » Mon Dec 11, 2017 11:42 am

Lyudmil Tsvetkov wrote:You wanna bet that they don't change the code?
No human interference at all?
Why do they need all that big team then?

There is no concept like artificial intelligence, there is simply no such concept, someone conceived it wrong and it spread across the world.
The machine is just executing the code written by humans, and it will always be so. It does not matter at all if it is linear or non-linear, it is still executing the code.
And that code changes. Basically, a huge autotuner, with what are CLOP and other tuners different, essentially?

For example, it should have started from somewhere, they should have at least the basic piece values and some psqt values. Then, when it wins a game and an e5 pawn is featured or an advanced knight outpost, the autotuner will increase their values, so what? How advanced is that?
What is the difference with normal tuning, I just can't see it?

No machine can start without a code, and that code will guide all its operations, EVER, regardless of whether it is a multilayer algorithm or not.

It is you who are mixing up things, not me.

You also believe they improved so much in 4 hours?
Want to bet they will not have a 5000 elo engine in another week, do you?
If they don't, then why are you arguing?

I have been there mentally and I know what is necessary to construct a much stronger chess playing entity. They don't have the necessary preconditions and will never succeed. Still 1850 currently.
You really have no clue whatsoever what you are talking about, eh? You'r just shooting off your mouth on issues and events you are utterly ignorant about. This of course shouldn't much surprise us, as your doing the same about Chess here all the time. It is a safe bet that when it would ever be reported astronomers detected radio-messages from an extraterrestrial intelligence, you would appear here to tell us exactly what these aliens eat and how they dress, and how what they told about it in the actually received message is all lies, and how much better you know what goes on on their planet...

In the reality outside your delusional Universe no piece values or PST were given to AlphaZero by the programmers, not even the hint that such concepts should be used. No human example games of what constitutes good play were fed to the machine. The only information it got in advance was how to calculate legal moves in a given position, and how a game ends. The machine could not have cared less whether this info described Chess, Checkers, Go, Shogi, Hex... It just took the rules, and after 4 hours of thinking about them it reached a conclusion of how to best play the game decribed by those rules, that was good enough to perform at 3000+ Elo level. And PST or piece values did not make part of its considerations.

The only code that was programmed and executed was the code that describes how to learn from its experience. This is no doubt a very complex and challenging task, which is why the team is so large. Nothing was programmed about Chess or how to play it. The only thig that was programmed is how to learn from matches of whatever game you could play. And then restrict the games it could play to Chess games, by imposing FIDE rules on the moves in the test games. And from that it learned itself how to play strong Chess, without any human interference.

Post Reply