AI Chess Bots Defeat Humans Because 'No Human Being Would Play Chess Like This'

Levy Rozman
American chess International Master, The 'Internet's Chess Teacher'

Back in the late 1997, Garry Kasparov, a Russian chess grandmaster, and a World Chess Champion at the time, made news around the world when he was defeated by IBM's Deep Blue.

In a highly-publicized match, he was defeated by the IBM supercomputer managed in game six, marking the first time a reigning world champion lost a match to a computer under standard chess tournament conditions.

While the match stirred quite a controversy and speculation about whether there were any external interventions or unusual moves made by Deep Blue, Kasparov later revealed that he underestimated the machine's abilities and made strategic errors.

Fast forward, AIs in computers have become so smart and powerful that a bunch of chess bots have surpassed human-level performance in chess, with some of the world's top chess engines are not only powerful, but also considered unbeatable by any human player.

While there have been some chess players who have defeated chess engines in certain circumstances, in general, even the grandmasters could barely survive after 20 to 30 moves.

Levy Rozman, known online as GothamChess.
Levy Rozman, known online as GothamChess.

There are reasons why chess bots are far superior to humans

Not only that chess bots don't make human-based mistakes, but also because chess bots can predict moves multiple times ahead of humans, in fraction of the time.

Levy Rozman isn't a tech person, but he has seen some of the world's most powerful chess bots, and barely survived when pitted against Stockfish, an open source AI chessbot engine. Speaking to Gary Linscott, a software engineer who has been working on computer chess for more than 20 years, it's revealed that Stockfish makes use of super-powerful computers can can calculate more than 10 million nodes, or positions, per second.

In Linscott's example, give it 30 seconds, and it would calculate more moves than any human chess players in their entire lifetime.

While humans can only predict what would happen in a few moves time, Stockfish can already access the possible end of the game move, right after the opening move.

In response, Rozman said that he could only predict a possible 10 moves before it happens, and comparing that to Stockfish, he could literally live a hundred lifetimes and can still not beat Stockfish.

In the middlegame, the time when all the game pieces are out, Rozman found that Stockfish does moves that no human would.

"That move violates most human chess principles that we've brought up with," said Rozman.

It does this because the AI thinks that it's best move for that current time, even if the move is not something a normal human would do.

Like for example, it has long been a general rule of thumb that the King should be protected at all cost, and no player should weaken any position the King has, in order to ensure its longevity if the King has to escape.

But Stockfish can literally create a huge opening for an opponent to enter.

Stockfish would do this, after accessing a game tree, which is uniquely created after the first move. That game tree would have all of the legal moves in the game. Using each turn, Stockfish rank all of those moves, and choose the best one.

While powerful chess bots can process millions of move per second, it focuses only on two or three moves in the future, which Linscott refer to as the "Alpha-Beta technique."

And this becomes more apparent, when there are only seven pieces remaining on the chess board.

In this end game situation, Stockfish can literally solve the game "perfectly."

What this means, Stockfish can effectively calculate all possible moves, and know which is best to take.

According to Linscott, knowing all possible moves for 7 pieces remaining only requires around 10-20 terabytes of database of knowledge, "which is a lot, but manageable."

Read: The Most-Watched Streaming Chess Game, Following The Loss Of A Master Against An Amateur

Garry Kasparov, in one of his matches against IBM's Deep Blue.
Garry Kasparov, in one of his matches against IBM's Deep Blue.

Chess has an Elo rating system, which calculates the relative skill levels of players.

The highest Elo rating achieved by a human chess player is held by Magnus Carlsen, who reached a peak rating of 2,882 in 2014, surpassing the previous record held by Garry Kasparov.

Rozman himself is a grandmaster with an Elo of over 2,300. And this pale if compared to Stockfish, which according to Linscott, exceeds the Elo ranking of 3,500.

Stockfish was able to foresee a checkmate in 35 moves that neither Magnus Carlsen nor Fabiano Caruana could see during the 2018 World Chess Championship.

Despite humans do make victories over chess bots, these circumstances only happened occasionally.

With improved software and hardware, and the advancements of technologies, the overall trend in chess is that AI is becoming increasingly dominant.

It's pretty much inevitable that a computer will become increasingly effective at playing chess.

The reason why AI can't beat everyone in chess is because chess is an extremely complex game with an incredibly large number of possible moves and outcomes.

Furthermore, humans have the ability to think creatively, take risks, and adapt to changing circumstances. This is something AIs aren't generally trained to do.

In the end, while AI may be able to process more information faster and make more accurate calculations than humans, there is still a level of intuition and creativity that is difficult to replicate in machines.