Awful paper)

[Dann Corbit]

BTW, one deep learning evangelist in Russia (Anatoly Levenchuk) was so excited after reading news about Giraffe that said in one of his public lectures (literally):

...all teams that for 5-6 years with the help of grandmasters worked on game tricks and positions analysis... that teams are now fired, they are resting now bacause the new type of computer programming is now appeared!

http://22century.ru/video/levenchuk-deep-learning [from 16:56] (sorry, only Russian)
A lot of news sites in Russia wrote a lot of sensational and merely idiotic relations about this project.

http://medialeaks.ru/news/1509dalex_chess
https://geektimes.ru/post/262612/
http://www.interfax.ru/world/466755

etc.

It is natural that you get annoyed with over-enthusiastic reactions to the project (including mine). But keep in mind that it is not the author's fault that people over-react in that way.

[Sergei S. Markoff]

It is natural that you get annoyed with over-enthusiastic reactions to the project (including mine). But keep in mind that it is not the author's fault that people over-react in that way.

Yes and no. Of course, his work is well done and interesting. Of course, over-enthusiastic reactions are sometimes annoying. But anyway the author made some wrong statements himself. For example:

Their power relies on brute force, the process of searching through all possible future moves to find the best next one.

We all know that the past progress in computer chess has nothing common with brute force.

[Dann Corbit]

It is natural that you get annoyed with over-enthusiastic reactions to the project (including mine). But keep in mind that it is not the author's fault that people over-react in that way.

Yes and no. Of course, his work is well done and interesting. Of course, over-enthusiastic reactions are sometimes annoying. But anyway the author made some wrong statements himself. For example:

Their power relies on brute force, the process of searching through all possible future moves to find the best next one.

We all know that the past progress in computer chess has nothing common with brute force.

Perhaps this is a mistake of notation.

There were two schools of thought in early chess programs. There were those which were extremely selective searchers and did not examine the whole tree. And there were brute force searchers. One can easily argue that alpha-beta is also still completely brute force. Now, some pruning algorithms are clearly not sound {in that they do not provably still always arrive at the correct answer like alpha-beta alone} and so all programs that prune are no longer brute force, as it were.

But the early nomenclature was selective search verses brute force.
It is also a possible attempt at being somewhat dramatic (which is questionable in a scientific paper -- e.g. Ben Franklin and the kite).

I think that the reaction to the paper is probably overblown (I myself was utterly geeked out by it) but it accomplishes several useful things:
1. Spawns interest in computer chess amongst the unwashed masses.
2. Publishes AI techniques that are now well known (I read Tridgell's TD-Lambda paper, but this looks different to me). There are others, perhaps, doing the same thing, but they did not publish something that I read.
3. Raises new thoughts (e.g. using neural net in SEARCH decisions is utterly novel, so far as I know).

All in all, I thoroughly enjoyed reading the paper. I think it undoubtably has some defects. But I still give it an A. That's for me. I felt educated and refreshed. Probably just because I was back behind everyone else.

[matthewlai]

Is it only me who's being driven mad by reading this paper?

https://www.technologyreview.com/s/5412 ... al-master/

Most enraging quotations:

"...while computers have become faster, the way chess engines work has not changed. Their power relies on brute force, the process of searching through all possible future moves to find the best next one"

"In a world first, a machine plays chess by evaluating the board rather than using brute force to work out every possible move"

And so on.

Human brain contains about 86 000 000 000 neurons and several hundreds of trillions of synapses. Comparing to our best elecronic machines it is much bigger and stronger in a terms of logical elements number. That's why computer algorithms for chess looks to be much more sophisticated than human ones because they works on descent hardware but provides a better play)

When someone says: "Let we limit computer speed to only 5 nps and will see...", I want to reply: "Let we limit your brain size to only 1000 neurons and then will see!"

Any strong modern chess program is a masterpiece of machine learning. We have at avarage 35 possible moves in average chess position. So if chess programs are so dumb and using exhaustive search, Deep Blue with it's 300 mln nps will search about 11 days to solve average "mate in 6", but at practice most programs can solve 99% of these puzzles in a few seconds using conventional desktop.

Best chess program in 2016 has more then 300 Elo advantage over the best 2006 program *at the same hardware*.

But sometimes some people are going to jump out like the Jack in the box to say: "You're all idiots! I'm just invented the selective search! And I want to publish the epic folio about it in MIT journal!"

Author of Giraffe here.

Awful paper? Or awful article?

Note that I didn't say most of the things in that article, which does have quite a few technical inaccuracies.

But anyway the author made some wrong statements himself. For example:

Their power relies on brute force, the process of searching through all possible future moves to find the best next one.

Hmm... when did I say that? Is it in my paper (not the article)?

The biggest difference between Giraffe and existing programs is automatic feature extraction. There are many engines that have auto-tuning for eval parameters now (using CLOP, genetic algorithm, etc), but they are still tuning parameters of an evaluation function designed by humans. Giraffe is about being as "hands off" as possible, and having the training "design" features as well, not just tuning their weights. It is fundamentally very similar to techniques used in AlphaGo (though I didn't know about AlphaGo at that time), but with minimax instead of MCTS.

Is it impressive? That's for other people to decide. Is it novel? Yes, I do think so. I don't think anyone has achieved that before, with chess or other games of similar complexity.

Yes, there has been quite a bit of over-reaction from the media, but it's the media. They over-react on everything, and there's not a lot I can do about that. Of the 20-30+ articles written about Giraffe, I was only contacted by about 2-3 of them, and had absolutely no input in the rest.

[matthewlai]

It is natural that you get annoyed with over-enthusiastic reactions to the project (including mine). But keep in mind that it is not the author's fault that people over-react in that way.

Yes and no. Of course, his work is well done and interesting. Of course, over-enthusiastic reactions are sometimes annoying. But anyway the author made some wrong statements himself. For example:

Their power relies on brute force, the process of searching through all possible future moves to find the best next one.

We all know that the past progress in computer chess has nothing common with brute force.

Perhaps this is a mistake of notation.

There were two schools of thought in early chess programs. There were those which were extremely selective searchers and did not examine the whole tree. And there were brute force searchers. One can easily argue that alpha-beta is also still completely brute force. Now, some pruning algorithms are clearly not sound {in that they do not provably still always arrive at the correct answer like alpha-beta alone} and so all programs that prune are no longer brute force, as it were.

But the early nomenclature was selective search verses brute force.
It is also a possible attempt at being somewhat dramatic (which is questionable in a scientific paper -- e.g. Ben Franklin and the kite).

I think that the reaction to the paper is probably overblown (I myself was utterly geeked out by it) but it accomplishes several useful things:
1. Spawns interest in computer chess amongst the unwashed masses.
2. Publishes AI techniques that are now well known (I read Tridgell's TD-Lambda paper, but this looks different to me). There are others, perhaps, doing the same thing, but they did not publish something that I read.
3. Raises new thoughts (e.g. using neural net in SEARCH decisions is utterly novel, so far as I know).

All in all, I thoroughly enjoyed reading the paper. I think it undoubtably has some defects. But I still give it an A. That's for me. I felt educated and refreshed. Probably just because I was back behind everyone else.

Thanks Dann! Means a lot to hear that from you. Glad you enjoyed reading it!

[Sergei S. Markoff]

Perhaps this is a mistake of notation.

There were two schools of thought in early chess programs. There were those which were extremely selective searchers and did not examine the whole tree. And there were brute force searchers. One can easily argue that alpha-beta is also still completely brute force. Now, some pruning algorithms are clearly not sound {in that they do not provably still always arrive at the correct answer like alpha-beta alone} and so all programs that prune are no longer brute force, as it were.

At least since Kaissa Stockholm version (1974) we have null-move pruning, so even if we will not account alpha/beta, programs are selective at least since 1974.

But the early nomenclature was selective search verses brute force.

May be it's from pre-practical era of first Shannon works or from very early time of ITEF-M20 vs Stanford program. But it was already completely incorrect at the late 1970s.

It is also a possible attempt at being somewhat dramatic (which is questionable in a scientific paper -- e.g. Ben Franklin and the kite).

I think so. Please understand me correct — I'm just a little bit ironic about all this story. Just a little bit upset because I see how much work was done by Stockfish team, Komodo team and other leaders, they spent a lot of time mastering their brilliant learning/tuning stuff, but now there is a spread of claims that their work worth nothing.

http://www.smbc-comics.com/index.php?db=comics&id=1623

I think that the reaction to the paper is probably overblown (I myself was utterly geeked out by it) but it accomplishes several useful things:
1. Spawns interest in computer chess amongst the unwashed masses.
2. Publishes AI techniques that are now well known (I read Tridgell's TD-Lambda paper, but this looks different to me). There are others, perhaps, doing the same thing, but they did not publish something that I read.
3. Raises new thoughts (e.g. using neural net in SEARCH decisions is utterly novel, so far as I know).

Of course, there are positive sides of this story. Unfortunately there is a spoonful of creosote in the barrel of honey(

[Sergei S. Markoff]

Awful paper? Or awful article?

Note that I didn't say most of the things in that article, which does have quite a few technical inaccuracies.

Article, sorry for my language inacurracy.
It's really bad that MIT article contains such a loud statements that was not a part of original paper. Also the article was not signed that's why I've supposed that article and paper was written by the same person.

The biggest difference between Giraffe and existing programs is automatic feature extraction.

A lot of people played with it. Just a few links:

https://chessprogramming.wikispaces.com ... ecognition
https://chessprogramming.wikispaces.com/Morph
https://chessprogramming.wikispaces.com/KnightCap
http://www.ismll.uni-hildesheim.de/lehr ... /chess.pdf
http://erikbern.com/2014/11/29/deep-learning-for-chess/
https://chessprogramming.wikispaces.com/Octavius
https://chessprogramming.wikispaces.com/Meep

etc.

Is it impressive? That's for other people to decide. Is it novel? Yes, I do think so. I don't think anyone has achieved that before, with chess or other games of similar complexity.

Impressive: yes. I like your work, thank you for it.
Novel? Generally: no. In some specific thinks — yes, of course.

[Sergei S. Markoff]

One more important paper:

https://users.soe.ucsc.edu/~levinson/Papers/CNFCRL.pdf

[matthewlai]

A lot of people played with it. Just a few links:

https://chessprogramming.wikispaces.com ... ecognition
https://chessprogramming.wikispaces.com/Morph
https://chessprogramming.wikispaces.com/KnightCap
http://www.ismll.uni-hildesheim.de/lehr ... /chess.pdf
http://erikbern.com/2014/11/29/deep-learning-for-chess/
https://chessprogramming.wikispaces.com/Octavius
https://chessprogramming.wikispaces.com/Meep

From that list, only Erik's project attempts automatic feature extraction, and he only did it with supervised learning, and didn't achieve very good results. Morph is very interesting, but is a totally different type of learning. All the others were based on weight-tuning of handwritten eval, even if their tuning techniques are more advanced than what we are seeing now.

[Sergei S. Markoff]

From that list, only Erik's project attempts automatic feature extraction, and he only did it with supervised learning, and didn't achieve very good results. Morph is very interesting, but is a totally different type of learning. All the others were based on weight-tuning of handwritten eval, even if their tuning techniques are more advanced than what we are seeing now.

Some more links:

http://eprints.utas.edu.au/62/1/Final_Thesis.pdf [from section 5.6.1]
http://www.top-5000.nl/ps/Automatic%20G ... dgames.pdf

At practice only a small part of chess AI techs are going into research papers. At least there was some academic projects before Giraffe and I suppose there was a lot of things done not in public.
Anyway it's no doubt that you have done good job. The only visible problem is media coverage.