How do modern chess engines work?

Speaker: Daylen Yang, University of California at Berkeley
Deep Blue was the best chess computer in 1997. Since then, modern chess engines have made significant improvements and are now far stronger than any human grandmaster. This talk gives an overview of how chess engines work, covering search/evaluation and interesting high-level and low-level optimizations. It will also discuss the Fishtest distributed testing framework, a method to measure strength improvement during chess engine development.
Daylen Yang joined Stockfish open source project in 2010. He is the developer of the Stockfish for Mac app and has played various roles for the project team. He is currently studying electrical engineering and computer science at the University of California, Berkeley and is interning at Facebook this summer.
Recorded at the Big Techday 8 / www.bigtechday.com of TNG Technology Consulting / www.tngtech.com on June 12th, 2015 in Munich / Germany

[@styleisaweapon]

Why is the audio so terrible on everything chess related? Seriously... it isnt hard to set a volume level AND THEN NOT TOUCH IT.

[@klaarnou]

Check your speakers or get your hearing tested moron.

[@diabl2master]

@@klaarnou did you even watch the video?

[@diabl2master]

I don't think the problem is caused by someone turning the volume down to 2% every now and then

[@JackIsNotInTheBox]

They deliberately lower the volume on confidential informations.

[@diabl2master]

@@JackIsNotInTheBox like what...?

[@stephenkamenar]

guys. the audio doesn't cut off. it's actually there just VERY QUIET. if you blast your headphones you can hear it

[@Orotachi]

12:57 Then tell me what it is.😑

[@modolief]

Very poor production qualities on this video. Daylen Yang very knowledgeable though.

[@peter_castle]

12:56 not audio. Can someone bring captions, probably by mailing the author?

[@QuasiELVIS]

I was really interested in this but the audio is so amateur it's basically unwatchable.

[@stormixgaming8389]

Amazing how much progress has been made in chess engines since this video was published. The rise of using neural networks in particular.

[@brookestephen]

I'm only sorry that audio keeps cutting out! Very illuminating talk!!!

[@dannygjk]

The sound problem could simply be that the microphone has a loose wire, (often due to people abusing the wires by yanking on them).

[@richarddeese1991]

I gave this video a thumbs up, because it answered my question as to what a chess engine is. It also seems quite interesting in and of itself, for those interested in chess and/or the workings of chess engines. However, as most commenters have noted, the audio leaves much to be desired. Applying my own logic tree to this situation, the possibilities are: 1) You are unaware of the problem. [I deem this highly unlikely.] 2) You don't care. [I also deem this highly unlikely.] 3) You're working on it. [I deem this very unlikely, as four years is a long time to be working on such a thing without tangible results.] 4) The audio problems are inherent to the recording - rather than being an artifact of translating the video to suitable format, or to the uploading process. [I deem this nearly certain, both by default *_and_* as the only reasonable assumption.] Most unfortunate. But thank you, nonetheless! Rikki Tikki.

[@TheBilly]

"TNG Technology Consulting: We Can't Even Operate A Camera" Great advertisement, guys.

[@michaelbauers8800]

Audio drops out. Otherwise enjoyed this

[@JackIsNotInTheBox]

They deliberately lower the volume on confidential informations.

[@user-im7km8tq7j]

@@JackIsNotInTheBox it doesn't make sense. they interrupt audio when he explains OPEN-SOURCE algorithm, what are you talking about. Also it doesn't make sense twice as in some parts sound is lowered not enough to make it impossible to recognize. I accept your claim as suggestion, but rather a conspiracy theory, sorry

[@JackIsNotInTheBox]

@@user-im7km8tq7j Okay, so you're telling me the camera magically reduced its microphone volume without any human intervention?

[@maxjohnson0703]

why does the volume change so much?

[@doorhandlesarecircle6450]

I love how in this comment section people are talking about the chess engines and not about audio volume..

[@spacefarers6960]

I cannot understand some parts AT ALL due to the audio quality, but the topic and the speech was awesome!

[@yesil1026]

I am wondering if some neural network based "evaluation function learner" could be written for chess like Alphago for Go. And how would it compare to Stockfish 8 if all other things are the same.?

[@sikfaka1]

google alphazero is a neural network, there are games between it and stockfish and it demolished stockfish

[@privatprivat7814]

wtf audio drop outs

[@diabl2master]

I figured there must be something akin to this 'quiescence' concept at play in Stockfish

[@philj9594]

So, is the way that stockfish decides that a move isn't worth exploring as deeply the reason why sometimes stockfish will miss the best move at a certain depth even though the move is obviously best if you only look ahead to something shallow like 4 ply? Is it because even though it's set to a depth above that, the way stockfish optimizes it reduces the depth on "weak" moves and only goes up to the set depth on a few of the "best" moves? Therefore, raising the depth setting will result in also raising the reduced depth of the search of the "weak" move, thus resulting in a higher depth making Stockfish figure out that the "weak" move was actually best? I want to make sure I'm getting this because this helps me understand how to use Stockfish better. I usually set Stockfish to very low depths because "no one can see that far ahead anyway", but now I'm considering raising it to something like depth=30 if this is the case. God that was difficult for me to write properly sorry.

[@joachimdietl6737]

what is up with the audio?

[@LouigiVerona]

Unfortunately, the video has serious problems with audio. Whole sections have dropped out. Could have been a great video.

[@dannygjk]

When he said Stockfish reaches depths of 40+ he should have said explicitly that it does not examine every variation to that depth unless it is a position with few pieces. So many people have argued with me that the engine sees every variation to the depth it displays. That drives me nuts. lol

[@dannygjk]

***** just do the math. Calculate how many nodes it would have to create then divide that by the nodes per second. If you get a huge number of seconds it isn't doing it.

[@jinxblaze]

probably not, i bet it stops that variation which keeps getting worse

[@turtle207]

The engine removes all other variations with the lower score at the begining, therefore it virtually calculates all the variation till depth of 40 directly and indirectly. If engine removes the wrong one and the oponent can find it, this will be a blunder.

[@conradnal8934]

Could you explain to me which positions reachable within *depth* moves are removed from the computer analysis and why? Thanks

[@dannygjk]

@@turtle207 I know it's been about three! years but devs stopped using forward pruning, (it sounds like you are talking about forward pruning), many years ago.

[@dr.ibbdebach7399]

It's a pity that there are these sound issues - in particular, regarding the very interesting intermediate questions from the audience (around 29:00 - 31:00) and in the Q&A part at the end. It is these very parts that are of particular interes to anyone with experience in chess engine development. Maybe a transcript could be added here? The rest is a very nice (brief but concise!) introduction to the field of modern chess programming. I really can't understand why there are these complaints here about the content qualiuty of this video.

[@TheBilly]

I agree. Nice overview if we could hear all of it. It pretty clearly gives introductions to the high level topics relevant to chess engines (although it assumes we all have CompSci degrees and know Minmax by heart)

[@AkshayAradhya]

Can build an advanced Chess Engine, but cant have stable audio.

[@jimfrien9040]

In stark opposition to the other comments, I actually find this video informative and very well made.

[@SuperYtc1]

Jim Frien Spiffing.

[@MusikPiratCH]

Very funny how fast after this Q & A AlphaZero showed what AI is capable of! :D

[@Lord_Volkner]

The sound keeps dropping out in this vid.

[@ThePadfam]

THIS is a comment about something other than audio. He seems a little nervous.

[@dikransarko7486]

Very poor audio.

[@birnumara1212]

At the end of it all , I just wanna say , Love you Stockfish Chess Love you Komodo Chess Love you Leela Chess Zero & all your counterparts.

[@birnumara1212]

Because , in this video , whatever has been explained from start to finish is done by the engines in just a matter of a second or two !! Jeez !! Absolutely Insane !!

[@diabl2master]

Surely +0.50 is 50 centipawns not half

[@MRbeqa100]

I have stockfish on my iphone 6s plus and i also have it on Galaxy S7 (Exynos) i tested and compared them both with the same strength, on android i tryed both smallfish and stockfish uci as well but thay cant even compare to stockfish on iphone, i dont know why but ios version simply destroys android stockfish versions, any idea?...

[@jkadoodle]

I'm 5 minutes in and the demo of "that's how it works" was pretty weak

[@peter_castle]

But the rest of the talk it's great (the parts in wich Tord talks and the audience asks questions no, because the audio is hard to listen).

[@diabl2master]

I agree. Fortunately I'd picked up the basic ideas around how stockfish works by using it many times. Someone who had no idea would be pretty lost.

[@diabl2master]

Of course, he had to be selective about what he says because he needs to explain the basics quickly

[@diabl2master]

He could have explained so much by just opening chess.com/analysis and doing a few moves (including some awful blunders)

[@DCth2]

Unwatchable, audio level constantly changing

[@caligulajodorowsky]

Can't hear anything

[@valentinewiggin9705]

alpha zero has joined the server

[@prohz9129]

6-8depth vs 18 Thats not very deep of blue.

[@1.4142]

modern

[@pellythirteen5654]

Fantastically interesting subject ruined by terrible audio. I am halfway but decide to not watch the end.

[@simeyD]

Now ask them about Alpha Zero.!!!!!!! lol

[@diabl2master]

Describing chess engines as a function taking a board and returning the best move is needlessly simplified and does not explain that what they really return is evaluations of board positions (and best move comes from that)

[@w00dyblack]

sound is fucked.

[@diabl2master]

It's very interesting but I wasn't really impressed by the quality of the lecturing, let alone the audio. I think he had to fit a lot into just one hour so he couldn't explain every detail to make the most coherent story.

[@Dude31463]

"Chess Engines"=The best and quickest way to ruin the brilliant game of Chess for everybody... That's all you had to say.

[@ling6701]

This talk assumes you know what minimax is and don't take the time to explain it for beginners, not the talk I was looking for.

[@SehrGutPeter]

"This talk assumes you know what algorithms are and don't take the time to explain it for beginners, not the talk I was looking for." One has to draw the line somewhere. And seriously, if you don't understand the concept of minimax, you should maybe start at a lower level than programming a chess engine anyway.

[@MichalMarsalek]

Not really, he explains it and shows pseudocode which is quite self explanatory.

[@JansherAquib]

It's simple. Pause, google search read , watch sources , get an intuitive feel, don't have to understand everything about everything, just an intuitive feel is enough and resume. Good day ✌️🏼

[@philj9594]

Okay? Well now you know that you have gaps in your knowledge and can work from there. So, in a way, this talk was actually useful for you. Also, I don't know what most of the concepts discussed in this video actually are, but it was still possible to for me to develop a basic understanding of how a chess engine works from using this talk. You just need to know that minimax is essentially just an algorithm (or something lmao) designed to find the best possible move for two opposing sides and is just the system being used to determine the move that stockfish decides is best. Minimax is being combined with an evaluation function (algorithm that decides value of resulting position of the move) and quiescent search (algorithm for searching through the tree of possible moves). My point is, you don't need to fully understand how these things work in order to follow along... just what they are and what their purpose is.

[@geonerd]

Terrible audio, awkward delivery, and not a lot of specific details. :(