Chess after AlphaZero: what changed in how we understand the game

Classical engines are built on search: they evaluate an enormous number of positions using rules people put into them - piece values, king safety, pawn structure.

AlphaZero was built the other way round. It was given the rules of the game and left to play against itself, training a neural network on the outcomes. No human evaluation terms, no opening book, no grandmaster games.

The match conditions were criticised afterwards - the opponent's settings and time control were not the most favourable for Stockfish. But the fact that a program with no human input at all reached that level in hours was never disputed.

The real discovery was not technical but chess-related: AlphaZero played in a style humans had written off as outdated or reckless.

• Long-term sacrifices. It gave up a pawn or the exchange with no immediate compensation - purely for piece activity that paid off twenty moves later.
• Marching the rook's pawn. Pushing the h-pawn up the board, long taught as a loss of time, turned out to be a powerful attacking device.
• Activity over material. Evaluations where it considered itself better with fewer pieces became routine.
• Restricting the opponent. It often played not for its own attack but to leave the other side with no good moves at all.

A curious outcome: that style sits closer to the romantic chess of the nineteenth century than to the strict positional school of the twentieth. Only now the sacrifices rest on millions of games rather than on intuition and daring.

AlphaZero itself was never publicly available, but the idea spread immediately.

• Open equivalents appeared. Projects that reproduced the approach on volunteer computing made neural-network play available to everyone.
• Classical engines adopted neural networks. Modern versions of traditional programs use a neural evaluation instead of hand-written rules - and got markedly stronger for it.
• Preparation changed. Professionals stopped hunting for «the best move» and started hunting for positions that are awkward for a specific opponent - the places where engines disagree with each other.
• Forgotten openings came back. Lines written off as dubious were rehabilitated by computer analysis and returned to tournament practice.

A side effect: the gap between engine and human grew so wide that the question of who is stronger lost all meaning. The computer stopped being an opponent and became a tool.

The practical takeaway for an amateur: use the engine as a tool for reviewing your own mistakes, not as a model to copy blindly - see fair play over the board.

The fear had been around since the nineties: once a machine beats humans, the game loses its point. Three decades on, the opposite happened.

• More people play than ever. Online platforms, streams and apps made chess a mass activity.
• Watching got better. A live evaluation bar turned broadcasts into something a non-specialist can follow.
• Players got stronger. Access to analysis lifted the average level: a club player today knows things that used to be the preserve of masters.
• The point never changed. People play not because they are the best calculators, but because it is interesting - the same as running, where machines have been faster for a century.

What did change is the value of playing in person. When perfect analysis is available to everyone, what gains value is precisely what a computer cannot give: an opponent across the board, a ticking clock, and no way to take the move back.