NRC discontinued its weekly chess column because much about chess could already be found online. That decision presents opportunities for Folia. After all, chess is immensely popular among high-educated people. And the Bridge on the Roeterseilandcampus would be ideally suited to host the world’s finest chess tournament, Han van der Maas argues.
The NRC has stopped Hans Ree’s chess column after 34 years (in Dutch) because there is already so much about chess online. Indeed, there is some truth to that. The NRC could, as per their reasoning, restrict itself to publishing obituaries. That presents opportunities for the online Folia. Do not worry, this chess column is especially for non-chess players, as there is a staggering amount you need to know about chess.
For example, chess offers hope in frightened AI days. AI has already struck chess hard twice. The first time was when Deep Blue defeated world champion Kasparov. Unforgettable was Kasparov's theatrics after he gave up in the last game. But Deep Blue was a calculating monster and had access to all the book knowledge about chess openings and endgames. Foul play, you could still think, along with Kasparov.
Another blow followed in 2017. A revolutionary chess program, AlphaZero, rediscovered in a few hours the chess opening theory that humans had developed in a collective effort of several hundred years, played surprisingly creatively and pulverized not only human opponents but also the best chess computers.
Many had predicted the end of chess when computers would beat humans (which, by the way, was long thought to be impossible). But nothing could be further from the truth. Chess has become immensely popular in the past five years or so. On the world's largest chess site, Chess.com, people play chess against random other earthlings. At the beginning of this year, the Chess.com website was down for weeks. The cause was a spectacular growth since the fall, long after the Covid crisis and the popular Netflix series The Queen’s Gambit. In January, the website was registering hundreds of thousands of new members a day and in the month of February, 1 billion games of chess. The causes are anyone’s guess but social media undoubtedly plays a role. The most popular social media post in 2022 was a fake chess match between Messi and Ronaldo. Some believe that the much-discussed, alleged fraud of a top American chess player plays a role, but that seems as likely as Dutch teenagers started studying Psychology thanks to Diederik Stapel.
Anyway, every phone beats the world champion with ease and yet chess is more popular than ever. That offers hope for the arts!
You should also know something about AlphaZero. This AI system taught itself chess, using neural networks and a special form of operant conditioning, a bit like how we learn. Where for years the progress was in brute computation, something humans are very bad at, AlphaZero works on principles that are also essential in human intelligence. In a recent paper in PNAS, the authors (with former world champion Kramnik as co-author) compare AlphaZero's learning process to that of humans and see many similarities.
AlphaZero was recreated by a team of volunteers under the name Leela Chess Zero. Leela dominates the competitions between chess computers with “human” chess. This is evident not only in Leela's creativity but also in the blunders. As befits a chess column, I have two diagrams. In the first, Leela, with a staggeringly high chess rating of 3233, chooses Dd8xd4. Incomprehensible and yet also reassuring! By the way, this was an UltraBullet game against a human opponent. I doubt you want to know what UltraBullet chess is. The second position is an example of a position that all chess computers still fail to judge properly. People quickly see through that this position is a draw, but even Leela does not understand this position. This “understanding” is at the heart of the discussion of the differences between natural and artificial intelligence.
And then ChatGPT. If you want to win from the computer as a complete amateur, I recommend ChatGPT. But chess is also a good one for those looking for an example of the danger of blind faith in Large Language Models. I asked version 4 this: “This is a chess position: white has Kd5, Ra7, e6; black has Kd8, Tg6. Black to move. There are no other pieces on the board. What would you play?”
The answer is remarkable. First, ChatGPT comes up with all sorts of wise words with nothing wrong with them: “The white pawn on e6 is close to promotion and the white king is in a strong central position. Meanwhile, the black rook is not well positioned to prevent promotion.” But then ChatGPT suggests Tg5+. A grave move, as we say in good Dutch!
By the way, nothing but positive things about ChatGPT. To the vast majority of my questions, the answers are staggeringly clear and good. Just type in some conspiracy ideas: Neat refutations. And once he's wrong, ChatGPT is never shy about apologizing. If there is somewhere to sign a manifesto for the latest AI developments, I will gladly do so. Long live the democratization of knowledge, what a blessing for humanity.
Finally, the finest chess tournament in the world is held in Wijk aan Zee. Each year a round is played at an outside venue, such as the Rijksmuseum or the Arena. How wonderful it would be if De Brug of the Roeterseilandcampus is that venue next year: we need to lure all those chess-playing teenagers to the UvA!