recent article on alphazero ... 12/11/2017 ...

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

Moderators: hgm, Rebel, chrisw

Milos
Posts: 4190
Joined: Wed Nov 25, 2009 1:47 am

Re: recent article on alphazero ... 12/11/2017 ...

Post by Milos »

hgm wrote:So what is your point? That we should suspect a big fraud if they manage to beat a drunk Lee after they managed to beat a sober one without any possibility of cheating?
No, the point is, they are not actually capable of beating SF in fair and optimal conditions for SF, which would mean they wouldn't have generated nearly as much publicity as they did in case of dominant victory, therefore they used totally immoral approach of crippling SF in any way possible that is not immediately obvious and using absolutely unfair comparison to obtain that marketing goal.
Their result, no matter how impressive, has nothing to do with science or advancing the filed of whatever. It's all marketing and shameless self-promotion of Google as a company and their ML cloud services using all means available.
And most ppl are either afraid to say that king is naked, just following the herd, or simply too stupid to see it. I don't know in which category you are.
Last edited by Milos on Fri Dec 15, 2017 10:09 am, edited 1 time in total.
Michel
Posts: 2278
Joined: Mon Sep 29, 2008 1:50 am

Re: recent article on alphazero ... 12/11/2017 ...

Post by Michel »

Milos wrote:
hgm wrote:So what is your point? That we should suspect a big fraud if they manage to beat a drunk Lee after they managed to beat a sober one without any possibility of cheating?
No, the point is, they are not actually capable of beating SF in fair and optimal conditions for SF, which would mean they wouldn't have generated nearly as much publicity as they did in case of dominant victory, therefore they used totally immoral approach of crippling SF in any way possible that is not immediately obvious and using absolutely unfair comparison to obtain that marketing goal.
Their result, no matter how impressive, has nothing to do with science or advancing the filed of whatever. It's all marketing and shameless self-promotion of Google as a company using all mean available.
And most ppl are either afraid to say that king is naked, just following the herd, or simply too stupid to see it. I don't know in which category you are.
Please go away.
Ideas=science. Simplification=engineering.
Without ideas there is nothing to simplify.
Milos
Posts: 4190
Joined: Wed Nov 25, 2009 1:47 am

Re: recent article on alphazero ... 12/11/2017 ...

Post by Milos »

Michel wrote:
Milos wrote:No, the point is, they are not actually capable of beating SF in fair and optimal conditions for SF, which would mean they wouldn't have generated nearly as much publicity as they did in case of dominant victory, therefore they used totally immoral approach of crippling SF in any way possible that is not immediately obvious and using absolutely unfair comparison to obtain that marketing goal.
Their result, no matter how impressive, has nothing to do with science or advancing the filed of whatever. It's all marketing and shameless self-promotion of Google as a company using all mean available.
And most ppl are either afraid to say that king is naked, just following the herd, or simply too stupid to see it. I don't know in which category you are.
Please go away.
No I won't. There is something called freedom of speech. Please offer counter argument instead of totalitarian habit of sending away anyone whose opinion you don't like.
User avatar
hgm
Posts: 28134
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: recent article on alphazero ... 12/11/2017 ...

Post by hgm »

Milos wrote:No, the point is, they are not actually capable of beating SF in fair and optimal conditions for SF,
'Fair and optimal conditions' meaning that SF should not have been forced to play the moves by itself, but some other entity (namely a book) should have been allowed to play the moves instead... :lol:
which would mean they wouldn't have generated nearly as much publicity as they did in case of dominant victory, therefore they used totally immoral approach of crippling SF in any way possible that is not immediately obvious and using absolutely unfair comparison to obtain that marketing goal.
Yeah, sure. It is very crippling when you have to find your own moves, just as the opponent does. Or when you have to play at fixed time per move, just as your opponent does. In fact anything that doesn't rig the odds massively in your favor would be highly unfair. After all, Stockfish is the TCEC champion. How dare they subject it to the same conditions as the opponent! :roll:
Milos
Posts: 4190
Joined: Wed Nov 25, 2009 1:47 am

Re: recent article on alphazero ... 12/11/2017 ...

Post by Milos »

hgm wrote:
Milos wrote:No, the point is, they are not actually capable of beating SF in fair and optimal conditions for SF,
'Fair and optimal conditions' meaning that SF should not have been forced to play the moves by itself, but some other entity (namely a book) should have been allowed to play the moves instead... :lol:
which would mean they wouldn't have generated nearly as much publicity as they did in case of dominant victory, therefore they used totally immoral approach of crippling SF in any way possible that is not immediately obvious and using absolutely unfair comparison to obtain that marketing goal.
Yeah, sure. It is very crippling when you have to find your own moves, just as the opponent does. Or when you have to play at fixed time per move, just as your opponent does. In fact anything that doesn't rig the odds massively in your favor would be highly unfair. After all, Stockfish is the TCEC champion. How dare they subject it to the same conditions as the opponent! :roll:
Well that's just hypocritical coz the whole your life you are a big proponent and participant in WCCC that uses uneven hardware and guess what, opening books, and not once you said that using opening books is stupid, pointless or mocked ICGA for the format they use...
peter
Posts: 3254
Joined: Sat Feb 16, 2008 7:38 am
Full name: Peter Martan

Re: recent article on alphazero ... 12/11/2017 ...

Post by peter »

hgm wrote:
Milos wrote:No, the point is, they are not actually capable of beating SF in fair and optimal conditions for SF,
'Fair and optimal conditions' meaning that SF should not have been forced to play the moves by itself, but some other entity (namely a book) should have been allowed to play the moves instead... :lol:
which would mean they wouldn't have generated nearly as much publicity as they did in case of dominant victory, therefore they used totally immoral approach of crippling SF in any way possible that is not immediately obvious and using absolutely unfair comparison to obtain that marketing goal.
Yeah, sure. It is very crippling when you have to find your own moves, just as the opponent does. Or when you have to play at fixed time per move, just as your opponent does. In fact anything that doesn't rig the odds massively in your favor would be highly unfair. After all, Stockfish is the TCEC champion. How dare they subject it to the same conditions as the opponent! :roll:
Bias (and I'm sure not because of lack of knowledge) is to let a learning machine let play 100 games with fixed 1'/move against an engine programmed for using the TCs on its own time management and programmed for using books to play reasonable chess as for what chess players would call reasonable.
At least all the rating lists compare Celo always using books for the matches or selected opening position sets , don't they? So where come the Elo given after this "match" from, if SF without book isn't rated at all neither?)

Main bias is to rate the "tested" learning machine against maybe only 3 or 4 at all different opening lines in all 100 games (who shows there were any more at all than in the 10 most "beautiful" ones shown?) and come to the conclusion, A0 would have reinvented opening theory already yet on its own?
To make believe that, showing some graphs about probality of normally in theory played opening lines in selfplay of A0 only?
There's the plan to hope for misunderstanding of the reader with this kind of "results" in a "paper".

A0 in the ten shown games had its good performance having learned to beat SF in a few opening lines, repeated 100 times.
Period.
If it wouldn't have managed to learn playing against SF 1'/move fixed 100 times repeating these few lines, what kind of a learning machine would be that?
Any hash- and or book- learning SF would have performed quite well too against a bookless SF repeating these few lines on and on not being able to learn at all.
So what
:?:
Last edited by peter on Fri Dec 15, 2017 11:30 am, edited 3 times in total.
Peter.
User avatar
hgm
Posts: 28134
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: recent article on alphazero ... 12/11/2017 ...

Post by hgm »

This doesn't seem to relate in any way to 'fairness'. Book building is one aspect of Chess, and taking it out of the equation by forcing participants to play without book, (or EGT) or from a common book specified by the organizers cannot be said to be a 'Chess championship'. It would be a 'search championship'. That doesn't mean it is in any way unfair.

The only purpose of the AlphaZero experiment was to demonstrate that it was possible to create a top-class searcher based on NN-guided MCTS through self-training from nothing but the rules. And that is exactly what they did. They used Stockfish' search as a yardstick.

Pointing out that Stockfish could play better at other TC is beside the point. Do you really doubt that AlphaZero could also play better, if they had taken the trouble to also train a NN for time management? But that wasn't their goal, so they did not do it, and played a match where both players had to use fixed time per move.
Milos
Posts: 4190
Joined: Wed Nov 25, 2009 1:47 am

Re: recent article on alphazero ... 12/11/2017 ...

Post by Milos »

hgm wrote:The only purpose of the AlphaZero experiment was to demonstrate that it was possible to create a top-class searcher based on NN-guided MCTS through self-training from nothing but the rules. And that is exactly what they did. They used Stockfish' search as a yardstick.
Yea sure, that is why 500 billion$ worth company, that does exclusively advertising as a the main business, uses 50 million$ worth equipment for training. They use it once, and then they close the project (at which there were a few thousands of man-hours spent in the previous 3 years). Just for pure love of science and demonstration.
Give me a break, you are either extremely naive or just playing stupid.
User avatar
hgm
Posts: 28134
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: recent article on alphazero ... 12/11/2017 ...

Post by hgm »

M$50 at an interest rate of 2%/yr, for 8 hours ~ 1milli-year is $1000. So yes, I would say that for a company like Google this is just a joke. The big thing was the Go victory against Lee Sedol, in a game that was supposed to be unfathomable to 'computer intelligence'. In a sense the equipment was just a left-over from that, and it is natural that they would try other things with it that could be easily done and could raise some interest.

Of course the hope is that this form of AI could be used in the future towards solving real-life problems, rather than just silly games. But not many real-life problems are equivalent to two-player zero-sum games with complete information, so this is rather a long shot.
Milos
Posts: 4190
Joined: Wed Nov 25, 2009 1:47 am

Re: recent article on alphazero ... 12/11/2017 ...

Post by Milos »

hgm wrote:M$50 at an interest rate of 2%/yr, for 8 hours ~ 1milli-year is $1000. So yes, I would say that for a company like Google this is just a joke.
Again you are comparing apples and orangutans. What the hell does interest rate have to do with it???
Google will not rent TPU in cloud services for less than a 1$/h.
5'000TPUs x (9h+12h+34h) = 275000k$ of lost revenue just for self-play games in the paper. So you are at least 2 and a half orders of magnitude wrong.
I wonder if you make such "great" estimations also in your main field of work?