Appearance
6.5 — Chess and the Mind Sports
Games with no physical component at all, which is precisely why they became the standard test of whether a machine can think — and why the answer turned out to be more complicated than either side expected.
Chess
Where did chess come from?
From India, around the sixth century, as chaturanga — a Sanskrit word meaning four limbs, referring to the four divisions of an army: infantry, cavalry, elephants and chariots, which became pawns, knights, bishops and rooks.
It travelled to Persia as shatranj, and the Persian words survive in the English game. "Checkmate" is from shah mat, the king is helpless. The word rook comes from Persian rukh, a chariot. From Persia it reached the Arab world and then Europe, arriving via Spain and Italy by about the tenth century.
The modern rules were settled in Europe around 1475, and the biggest change was the queen. In the older game the piece beside the king was a counsellor that moved one square diagonally, making it nearly the weakest piece on the board. In fifteenth-century Europe it acquired the combined power of rook and bishop, becoming by far the strongest — a change so dramatic the new game was called "mad queen chess". The bishop was similarly upgraded.
The reason usually offered is the presence of powerful female monarchs in Europe at the time, particularly Isabella of Castile. That is suggestive rather than proven; what is certain is that the change made the game far faster and more tactical, and it is the reason chess displaced the older version everywhere.
What is an Elo rating and what does the number mean?
A system for rating relative strength, devised by Arpad Elo, a Hungarian-American physics professor, and adopted by the world chess federation in 1970.
The mechanism is simple and worth understanding, because the same system now rates footballers, video game players and language models. Each player has a number. The difference between two players' numbers predicts the expected result. A 200-point gap predicts about a 76 per cent score for the stronger player; 400 points predicts about 91 per cent.
After a game, points transfer from the loser to the winner — and the amount depends on the surprise. Beating someone far below you gains almost nothing; beating someone far above you gains a lot. A draw against a much stronger player still gains points.
The scale is arbitrary in absolute terms and meaningful in differences. Roughly: a strong club player is around 2000, a candidate master 2200, an international master 2400, a grandmaster 2500 and above. Magnus Carlsen's peak of 2882 in 2014 is the highest ever achieved by a human.
There is a known problem called rating inflation: the average rating of titled players has drifted upward over decades, so a 2500 today is not exactly a 2500 in 1975. Comparisons across eras are therefore rough.
How does a grandmaster actually think?
Not by calculating further than everybody else, which is the popular assumption and largely wrong.
The classic experiment was done by Adriaan de Groot in the 1940s and refined by Chase and Simon in the 1970s. Show a player a position from a real game for five seconds and ask them to reconstruct it. Masters reproduce it almost perfectly; weaker players get a handful of pieces.
Then show them a board with the pieces placed at random. The master's advantage almost completely disappears.
The conclusion is that expertise is stored as chunks — recognised patterns of pieces that carry meaning — rather than as raw memory or raw calculation. A master does not see thirty-two pieces; they see six or seven familiar structures. Estimates put the number of such patterns a grandmaster holds in the tens or hundreds of thousands, built over roughly a decade of study.
That is why calculation is not the differentiator. A master's pattern recognition proposes a small number of candidate moves worth examining, and the calculation is spent on those. A weaker player calculates just as hard on moves that are not worth the effort. The finding generalises well beyond chess, and it is the empirical basis of most work on expertise.
Why did Deep Blue beating Kasparov matter less than AlphaGo?
Because they worked in fundamentally different ways, and only one of them learned anything.
Deep Blue, in 1997, searched around 200 million positions per second and evaluated them with a scoring function hand-written by engineers working with grandmasters. It was a triumph of specialised hardware and human-encoded knowledge. It could not do anything but play chess, and it did not learn from playing.
The famous incident is telling. In game one, Deep Blue made a move Kasparov found inexplicable and read as evidence of deep strategic understanding, which shook him for the rest of the match. It was later reported by one of the engineers that the move was the result of a bug — the machine, unable to choose, had picked a move essentially at random.
AlphaGo, in 2016, beat Lee Sedol at Go, a game with a vastly larger branching factor where the same brute-force approach cannot work. It was built from neural networks trained first on human games and then on millions of games against itself, and it evaluated positions by learned intuition rather than by written rules.
Its move 37 in game two was calculated by the system as having a roughly one-in-ten-thousand probability of being played by a human. Commentators called it a mistake. It won the game, and professional Go players have since incorporated the idea.
AlphaZero in 2017 removed the human games entirely: given only the rules, it played itself for hours and reached superhuman strength at chess, shogi and Go. That is the real turning point — a system that acquires the knowledge rather than being given it.
Has chess been solved?
No, and it almost certainly will not be.
A game is solved when the outcome under perfect play from the start is known. Checkers was solved in 2007 after eighteen years of computation, and the answer is a draw. Chess has a legal position count estimated somewhere around 10⁴⁴, and a game-tree size vastly larger — the often-quoted Shannon number of about 10¹²⁰ for possible game sequences exceeds the number of atoms in the observable universe.
What has been solved is the endgame. Complete tablebases exist for all positions with seven pieces or fewer, computed backwards from every checkmate. They contain positions requiring over 500 perfect moves to force a win, sequences no human would ever find and which look to a strong player like a draw.
The strong presumption among players is that chess with perfect play is a draw, based on how hard it is to win at the top level. That is an intuition, not a proof.
Go, and the others
Why is Go harder for computers than chess?
Three reasons that compound.
The board is larger: 19 by 19, 361 points, against chess's 64 squares. A typical Go position offers around 200 legal moves against chess's 35, so the tree of possibilities widens far faster.
Positions are much harder to evaluate. In chess, a decent estimate of who is winning can be made from material — counting pieces with weights. In Go, all stones are identical, and the value of a position depends on subtle judgements about influence, shape and the eventual life or death of groups. There is no equivalent of counting pieces.
And the effects are non-local: a stone placed in one corner can decide a battle on the far side of the board a hundred moves later.
The result was that Go programs remained at strong-amateur level long after chess programs surpassed humans, and the field needed a different approach — Monte Carlo tree search combined with learned evaluation — before progress came.
What other mind sports are there?
Bridge, a card game of imperfect information played in partnerships, where the bidding is a constrained language for passing information to your partner within rules that also let the opponents hear it. It resisted computers for a long time precisely because information is hidden and because inferring what your partner knows is central.
Scrabble was solved to superhuman level early, because a computer with the full dictionary and a good simulation of tile draws simply outplays humans on vocabulary and endgame counting. The best programs have been beating champions since the 1990s.
Poker fell later, and the milestone matters: in 2017 and 2019, programs beat professionals at both two-player and six-player no-limit Texas hold'em. That is a game with hidden information, bluffing and an opponent modelling you back, so the achievement is closer to a general result about decision-making under uncertainty than a game result.
Rubik's cube speedsolving is not a mind sport in the same sense but belongs here: the world record for a single solve is under four seconds, and the underlying mathematics established in 2010 that any cube position can be solved in at most 20 moves — a result called God's number, proved by classifying billions of positions with a large computation.
Chess and India
Why is India suddenly producing so many grandmasters?
Because of one player, and then an infrastructure.
Viswanathan Anand became India's first grandmaster in 1988 and world champion in 2007, holding the title until 2013. Before him India had essentially no presence in international chess despite having invented the game; the country now has over eighty grandmasters, the large majority of them since 2000.
The mechanism is documented rather than mystical. Anand's success made chess visible and made it a plausible career; sponsorship and state support followed; Chennai and Tamil Nadu in particular built a dense network of coaching and tournaments, which is why a disproportionate share of Indian grandmasters come from one region.
Cheap online play then removed the geographical constraint entirely. A child in a small town can now play thousands of rated games and study with engines, which previously required travel and a club.
The current generation — Gukesh, Praggnanandhaa, Erigaisi, Vaishali and others — reached the world's top ranks in their teens, and India won both open and women's sections of the 2024 Chess Olympiad.
What is the Immortal Game, and why do chess players name games?
Because a small number of games are studied for a century as objects rather than as results.
The Immortal Game, Anderssen against Kieseritzky in London in 1851, was a casual game in which Anderssen sacrificed a bishop, both rooks and his queen, and delivered checkmate with three minor pieces. It is the standard example of the Romantic style of the nineteenth century, where attack was valued above material and declining a sacrifice was considered poor form.
The Evergreen Game, the Opera Game — Morphy beating two aristocrats sharing a board at the Paris opera in 1858 — and the Game of the Century, in which the 13-year-old Bobby Fischer sacrificed his queen against Donald Byrne in 1956, are the others everyone knows.
Naming games is unusual among sports, and it exists because the game record is complete and reproducible. A chess game is a text that can be replayed exactly, forever, by anyone. Very few sports produce an artefact like that.
What comes next
The last page of this Part covers everything that did not fit — the sports with the strangest rules, the records that will not be broken, and the ones invented by accident.