AlphaZero - is this what 4000 rated chess looks like?

[Paul Bonham]

"Nigel Short @nigelshortchess Dec 14

"I have been watching Matthew Sadler's instructional videos on AlphaZero, on @YouTube. On seeing the engine's games, I feel like I am in the presence of God. There is an extraordinary beauty to these long-term positional sacrifices that it is so fond of making."

My sentiments exactly. Looking at the way this thing plays chess is truly something other worldly to behold. Material aspirations are now replaced by a library of patterns to aspire towards. It is like seeing an old for humans but radical new approach to chess for computers (that now do it far better then humans) that we have not seen since perhaps Tal beat Botvinnik in their 1960 World championship.

Very good insight Sid that you realize what AZ is doing is "old for humans but radical new approach to chess for computers (that now do it far better then (sic) humans".

It is old for humans to play many complete games of chess and learn from each one of them. Until AZ came along, no chess program ever did this. But now AZ does do exactly this, because from a given position, it plays complete games with semi-random variation of moves and collects scores and uses that data to evaluate what best move to make. Totally outperforming humans who simply don't have the perfect digital memory and score feedback mechanism to match.

All we are learning from this is that humans are not good at perfect information games because we have such imperfect memories, 99.999% of us can't for example store all values we calculate or measure in various memory locations and access them all perfectly later. It isn't "other worldly' as Sid calls it, it is simply technological. Computers have perfect memory, humans don't. It would be interesting if someone with photographic memory could learn to play chess by playing game after game after game and memorizing everything, but even then, that person is not able to match AZ on speed. AZ is orders of magnitude faster.

So this approach is the way forward for computers to take over from humans in evaluation and "learning"... but only when the learning involves quantitative measurements that lead to distinct and unequivocal conclusions. Chess is such a world: complete games always end 1-0, 0-1 or draw. There are no "gray areas" in chess when it comes to results. Thus AZ can play millions of complete games and measure each result.

To apply this to human economics or politics or social policy is still an impossible goal because there are so few quantitative measurements available and so many of them are subjective. Some are implying that AZ is heralding a new age where computers and AI will take over everything, and that is balderdash. Humans still have intuition and instinct that AZ can never have. And of course, we have some humans (Trump) who have intuition and instinct which goes against everything sensible for the human race. There is no coding morality or common sense into some future Alpha Zero. Morality and common sense are very human traits and only some of us have them. That has been the central theme of many SciFi stories / novels / movies / TV shows over many years and AZ doesn't change anything there.

I still say at this point we need to see AZ (with opening book) versus AZ (with opening book) in a match of many hundreds of games to get a clear picture of where chess is now at. I think the result is pretty predictable, chess is at a point where this new best play leads to something like 99.9% draws.

[Sid Belzberg]

"Nigel Short @nigelshortchess Dec 14

"I have been watching Matthew Sadler's instructional videos on AlphaZero, on @YouTube. On seeing the engine's games, I feel like I am in the presence of God. There is an extraordinary beauty to these long-term positional sacrifices that it is so fond of making."

My sentiments exactly. Looking at the way this thing plays chess is truly something other worldly to behold. Material aspirations are now replaced by a library of patterns to aspire towards. It is like seeing an old for humans but radical new approach to chess for computers (that now do it far better then humans) that we have not seen since perhaps Tal beat Botvinnik in their 1960 World championship.

[Vlad Dobrich]

Vlad, you overestimate the wealth of a regular 2780-2800 player (unless he has a wealthy parents).

Hi Victor,
I based my estimate on previous postings of the top earners. For example, here are the 2012 estimates >>>>>>>https://thechessworld.com/articles/n...-in-the-world/
Also, now there is the Grand Chess Tour to augment the winnings.
Granted the players have expenses such as travel but that is usually paid for by the tournament they are invited to.

[Sid Belzberg]

The hardware cost for a single AlphaGo Zero system, including custom components, has been quoted as around $25 million. AlphaGo Zero - Wikipedia


https://en.wikipedia.org/wiki/AlphaGo_Zero

An open source chess engine using the same methodology as Alpha Zero and trained on a volunteer mesh network that after a year is approaching Alpha Zero's strength is available for download and can be run uisng various GPU cards that cost between $1000-$3000 here is the link with the info.
http://blog.lczero.org/2018/09/guide...chess-gui.html

[Vlad Dobrich]

I guess we'll have to wait about 5 years for the price to drop under $1 million. I'm still working on my first million. :D

[Kerry Liles]

I expect that each of the top 10 Super GM's have at least a million Euros in the bank and will spend whatever it takes ASAP. Not to mention the principal sponsor of the St Louis Chess Club who would probably advance the price for the engine.

The hardware cost for a single AlphaGo Zero system, including custom components, has been quoted as around $25 million. AlphaGo Zero - Wikipedia

https://en.wikipedia.org/wiki/AlphaGo_Zero

[Victor Plotkin]

I expect that each of the top 10 Super GM's have at least a million Euros in the bank and will spend whatever it takes ASAP. Not to mention the principal sponsor of the St Louis Chess Club who would probably advance the price for the engine.

Vlad, you overestimate the wealth of a regular 2780-2800 player (unless he has a wealthy parents).

[Vlad Dobrich]

I expect that each of the top 10 Super GM's have at least a million Euros in the bank and will spend whatever it takes ASAP. Not to mention the principal sponsor of the St Louis Chess Club who would probably advance the price for the engine.

[Egidijus Zeromskis]

No doubt the Super GM's and others will be checking their lines against Alphazero over the holidays - assuming the engine can be downloaded for a price only they can afford.
This will be whole new ball game.

As I recall it runs on a special hardware that is not that easy to purchase.

[Vlad Dobrich]

No doubt the Super GM's and others will be checking their lines against Alphazero over the holidays - assuming the engine can be downloaded for a price only they can afford.
This will be whole new ball game.

[Vlad Dobrich]

And now to get Alphazero's opinion of the Caruana - Carlsen Match check this out >>>>> https://www.youtube.com/watch?v=VeqbDMzi_IY

[Paul Bonham]

Extremely impressive that Alpha Zero could be so dominant. Daniel King said that in some ways Alpha Zero played like Shirov, in that it went for active pieces and long term compensation for sacrifices. Some great games and proof that chess is still far from being perfected.

Unless AZ is right now playing near-perfect chess, which an 84% drawing rate and only 6 losses to SF8 could indicate. Someone calculated that the result amounts to a +52 ELO rating difference in this match, whereas SF10 beats SF8 with a +55 ELO difference. Can anyone verify the +52 number for this match?

What we need now is a similar matchup between:
(1) AZ and SF10
(2) AZ and AZ

where in all cases full opening book is used.

A match between AZ and AZ might result in a 95%+ draw rate.

[Ken MacDonald]

Did I miss something, OR, did Alpha make a couple of howlers very late in the last 149 move game? I used a machine
and it looked odd how the game changed so suddenly.

[Sid Belzberg]

Wow! That was an impressive game! Thank's for sharing Vlad!

[Ian Findlay]

Extremely impressive that Alpha Zero could be so dominant. Daniel King said that in some ways Alpha Zero played like Shirov, in that it went for active pieces and long term compensation for sacrifices. Some great games and proof that chess is still far from being perfected.