Why AI Chess Bots Are Virtually Unbeatable (ft. GothamChess) | WIRED

I'm about to play stockfish 16, the world's strongest

chess

computer, and let's see how long I can survive. What's that? Yes, I'm fine, I'm not fine. The queen checkmates the bishop, that was horrible. I never want to do that again. They checkedmated me in 34 moves. The most important thing is that I had a lot of fun. No, I did not do it. My name is Levy Rosman. I want to know a little more about how Stockfish thinks so I can use it. his tricks to beat other humans today we will talk with Gary Lincot, a software engineer who has worked on computer

chess

for more than 20 years.

Gary, you worked with dried fish, what exactly makes it so good? Dry fish really takes advantage of super powerful computers, it searches tens of millions of doctors per second, so if you give it 30 seconds it will search over a billion positions and that's more positions than a human will see in their entire life. He could live a hundred lives. I couldn't win this game. Not only is the evaluation function you use to evaluate the board alone probably stronger than almost every chess player in the world, your chess ELO is basically a number that quantifies your skill level, the highest ELO ever. reached by a human is by Magnus Carlson, my ELO right now is 2322, what exactly is the ELO of stock fish right now?

It's probably higher than 3500. I never had a chance to understand how stockfish actually thinks we should start from the beginning of the game. Most chess openings are subjective to humans. humans will choose an opening because they have more fun with it. You may like something because it gives you a great center or very active pieces. This is how most players choose their openings in my game against dry fish. I played Pawn to D4, let's see what opening the dry fish is. plays and Knight answered F6 Gary, so in the same way that humans learn chess openings, we read books, create chess files and try to memorize them because we can't access them during tournaments, which would be like cheating in a school exam in a book of openings. in chess it is a gigantic chess database does the stock fish have an opinion on the chess opening already on the second or even the first move? position what he is evaluating are the resulting structures that could evolve from that opening, but way down the game tree he is already evaluating the endgame from GetGo, so on his second move, stockfish played D5, which is a very popular move and here I made a decision to capture this Knight humans can see double AP and feel unpleasant they can see a weakness in his position does the dry fish have these little buzzwords like weakness or inactive piece?

What does it really mean when the evaluation function is trained from a huge library of chess games? that have been played and then it absorbs that knowledge, it just all goes to a giant neural network and then you can train your evaluation function to learn what is good and what is bad and then that network runs when stockfish is looking for the position and uses It To navigate the opening to succeed in our quest to understand dried fish a little better, let's delve into the middle games, most middle games begin between moves 10 and 15, it is the stage just after the opening, when you have taken out all your pieces.

Think of the middlegame and chess as a gigantic ocean, this huge tree of possibilities, so in my middlegame against stockfish Pawn to C4 is what I played, stockfish made a move that I don't think any human being would play here against me . The Absurd Pawn to G5, which made me audibly G. What is that, my God, is that actually the best move in the position? That move violates most of the human chess principles we have been raised with. You shouldn't push pawns in front of your king. not weakening an area of ​​the board where no advantage can be claimed, how does stockfish break these ancient principles?

In this case, it could simply be that the movement is too far from the human boundary to be reasonable. to evaluate every possible move and rank them, it all comes down to a single number, which is what is the probability of you winning from this position. Dried fish has no emotions or opinions, it will simply make the best move it can. think it exists, yeah, stockfish definitely doesn't look at it emotionally, it's evaluating with granny level quality, but then it looks 50 moves, 60 moves into the future to evaluate what the best possible move is. I'm curious, as an incredibly strong human being, how many? movements Are you looking towards the future?

I didn't realize this was Levy Rosman's roast. Sometimes I can make 10 moves if it's an ending position and I can identify the forced moves for both sides. Sometimes I'm stuck in a middle. game and I'm already undecided it's like being in a restaurant you have three good options they are all very difficult to evaluate and I will flip a coin but that doesn't work against the dried fish I counted in this position the black has 41 legal moves, how does stockfish know which of Eliminate them and finally decide to choose one if three or four look really good? build a game tree and that game tree will have all the legal moves, it will sort all those moves that are necessary one step forward at a time it will look for two moves forward, then three moves forward, then four moves forward, three or four moves speak to me in about 5 seconds and that's probably why I'm a good speed chest player, but I decide which move is best.

Sometimes I need to spend 5 minutes and 10 minutes finding out the truth about one of those movements. Chess computers actually do a similar process. Stockfish actually only looks at two moves per position and that is what is called the alpha beta search technique which is what Stockfish and most other engines are built on. Alpha Beta allows the engine to remove many moves from a position. because it knows they're worse than the best move it's found so far and that allows Alpha Beta to prune a lot of the search. trees, for example, if it were White's move, one of the legal moves is Knight to E4, which is really stupid because the pawn would take the Knight and White would be in a catastrophically worse position, some computers will be able to rule this out. immediate. and humans will too Alpha Beta alone can reduce those 35 moves you have to look at on average to about 15, which is a big reduction the last stage of chess is the end of the game is there a difference in the way What does dry fish move?

It approaches the end of the game versus the middle game or the opening, yes, once the number of pieces drops below a certain amount of seven, then dried fish can solve the game perfectly, chess is solved if seven pieces remain on a chess board or less exactly, yes, that's what they call endgame table bases and by resolved we literally mean every combination of possible moves if the bishop goes here and the knight goes here or if the bishop goes to that square originally and the knight goes there or there or there or any of these combinations.

This is solved, literally, dried fish just has to look up the position in its database and it knows exactly what the answer is, even for seven pieces, it's only 10 to 20 terabs of data, which is a lot, but you know that is manageable, there is nothing more. evaluating a best move is the best move in a position is winning a draw or losing and that's it, it's completely non-negotiable exactly now chess computers are also capable of playing without table bases and will continue to play incredibly strong chess there but they might make a mistake Gary, I lost to stockfish.

They checkedmated me in 34 moves that we didn't make. We got to the end of the game, but that's good. Stockfish always knows how to close out a game in the fewest number of moves. when the best players in the world do not have a game here Magnus Carlson against Fabiano Carana from the 2018 World Chess Championship Magnus was in an endgame in which he was losing a piece but neither of them found the correct technique Fabiano has a bishop and a knight and Magnus only has a bishop. Fabiano couldn't convert despite his extra material because it looked like Magnus had a defensive Fortress, the game ended in a draw as the players played and the whole world watched Stockfish yelling at them from the Digital Cyberspace Becomes Idiots, Forces Checkmate in 35 moves, that's ridiculous, like how could I have known that if both sides made a move for 35 moves, Black would win?

What dried fish can do is look at the bases of the table and each path to the table. bases, the best you can do is that M 35, if stockfish was playing black here instead of Fabiano, he would have played Bishop to H four, then brought his Knight here and then trapped his own Knight on the edge of the board so that the Knight now You can't escape anywhere because White's bishop would accept it. No human being would play chess this way. You can't trap your knight on the edge of the board, but apparently White runs out of moves first and slowly but surely Black would have won that ending. from a distance no human being on the planet saw that, so stockfish is a very specialized AI, how does it compare to other self driving cars AIS or something else? stockfish is only good at one thing and it is super good at that one thing which is playing.

Chess engines have now borrowed the latest generation research in artificial intelligence which is deep learning; however, they are still fundamentally limited to the domain of chess. I think chess has opened up in a way that AI can take advantage of. Chess engines to improve their own skills and that's the ideal use case for technologies is that they help humans do things better. I agree with Gary. I think AI has been a net benefit for chess. I think it's helping people improve at a rate we've never experienced, obviously we have to address the issue of potential cheating, but I'm still going to stay positive and optimistic, now that I know a little more about dried fish, I'm going to have my revenge.