Chess: The Drosophila of AI

Chess has been called the “Drosophila of artificial intelligence” — just as the fruit fly became the standard organism for studying genetics, chess became the standard problem for studying machine intelligence. The quest to build a machine that could play chess has driven some of the most important advances in computer science.

Early Dreams (1770–1950)

The Mechanical Turk (1770)

The first “chess machine” was a hoax. Built by Wolfgang von Kempelen, the Mechanical Turk was an elaborate automaton that appeared to play chess independently. In reality, a strong human player was hidden inside the cabinet, operating the mechanism. The Turk toured Europe for decades, defeating Napoleon Bonaparte and Benjamin Franklin before the secret was revealed.

Turing and Shannon (1950)

The real story of computer chess begins with two pioneers:

  • Alan Turing wrote the first chess algorithm in 1948 (executed by hand since no computer could run it). He called it “Turochamp.”
  • Claude Shannon published a seminal 1950 paper, “Programming a Computer for Playing Chess,” outlining two approaches:
    • Type A (brute force): evaluate every possible move to a fixed depth
    • Type B (selective): evaluate only promising moves, searching deeper in critical positions

These two approaches would define the field for the next 50 years.

The Early Programs (1950s–1970s)

Early chess programs were weak by any standard — losing to average club players — but each generation improved:

  • 1956: The first program to play a complete game (Los Alamos chess, on a 6×6 board)
  • 1958: Alex Bernstein’s program played on a full 8×8 board
  • 1967: Richard Greenblatt’s Mac Hack VI became the first computer to compete in a human tournament
  • 1970: The first computer chess championship was held, marking the beginning of competitive computer chess

The Rise of the Machines (1980s–1990s)

Dedicated Chess Hardware

The 1980s saw the development of purpose-built chess hardware that could evaluate millions of positions per second:

  • Belle (Ken Thompson and Joe Condon, Bell Labs) — the first strong chess computer, achieving a rating around 2200 (expert level)
  • ChipTest → Deep Thought → Deep Blue (at Carnegie Mellon, then IBM) — a lineage of increasingly powerful machines

Deep Blue vs. Kasparov

The defining moment in computer chess history:

  • 1996: Garry Kasparov defeated Deep Blue 4–2 in Philadelphia
  • 1997: An upgraded Deep Blue defeated Kasparov 3½–2½ in New York

The 1997 rematch was a global event. Deep Blue could evaluate 200 million positions per second. Game 2 became infamous — Kasparov resigned in a position later shown to be drawable, apparently unsettled by the machine’s mysterious play (a bug may have caused a random move that Kasparov interpreted as deep understanding).

The Software Revolution (2000s)

After Deep Blue, chess programming shifted from specialized hardware to software running on standard processors. Open-source engines democratized computer chess:

  • Crafty, Fritz, Shredder, Rybka, Houdini — successive generations of stronger engines
  • Stockfish (2008) — an open-source engine that became the gold standard, eventually reaching a level where no human could compete

By the 2010s, even a chess engine running on a smartphone was stronger than any human who ever lived.

The Neural Network Revolution (2017)

AlphaZero

In December 2017, Google DeepMind’s AlphaZero shocked the chess world. Starting with nothing but the rules of chess, AlphaZero taught itself to play through self-play reinforcement learning — playing millions of games against itself.

After approximately 4 hours of training, AlphaZero defeated Stockfish (the world’s strongest traditional engine) in a 100-game match: 28 wins, 72 draws, 0 losses.

What stunned chess experts wasn’t just the result — it was the style. AlphaZero played with a Romantic flair reminiscent of 19th-century masters, willingly sacrificing material for initiative and long-term positional advantages. It seemed to rediscover principles that human players had debated for centuries.

Leela Chess Zero (Lc0)

Inspired by AlphaZero, the open-source Leela Chess Zero project reproduced the neural network approach with distributed volunteer computing. Lc0 became a top-tier engine rivaling Stockfish, and the competition between traditional engines and neural networks drove both to new heights.

NNUE: The Best of Both Worlds

Stockfish eventually integrated neural network evaluation (NNUE — Efficiently Updatable Neural Network) into its traditional search framework, combining:

  • The deep positional understanding of neural networks
  • The precise tactical calculation of traditional alpha-beta search

This hybrid approach produced the strongest chess engines ever created.

How Engines Changed Human Chess

Preparation

Top players use engines to prepare opening novelties — new moves in well-known positions that surprise opponents. The depth of modern preparation has reached the 25th or 30th move in some lines.

Analysis

Engine analysis has deepened our understanding of chess. Positions that were considered winning for one side have been shown to be drawn (or vice versa). Endgame tablebases solve all positions with 7 or fewer pieces on the board with mathematical certainty.

Playing Style

The influence of engines on playing style is debated, but several trends are clear:

  • More accurate play — top players make fewer mistakes than ever before
  • Greater strategic complexity — engine preparation forces players into deeper, more nuanced positions
  • Appreciation for computer-like moves — moves that seem “inhuman” (quiet retreats in sharp positions, long-term sacrifices) have become part of human play

Anti-Cheating

With engines available on any smartphone, cheating has become a significant concern. Online platforms employ sophisticated statistical analysis to detect abnormal play patterns, and over-the-board tournaments use signal detection and behavioral monitoring.

The Future

Chess AI continues to advance. Current frontiers include:

  • Explainable AI — understanding why engines evaluate positions the way they do
  • Chess as a benchmark for broader AI research
  • Human-computer collaboration — “centaur chess” or “advanced chess” where humans and engines work together

The relationship between humans and chess computers has evolved from competition to collaboration. Engines haven’t replaced human chess — they’ve enriched it, revealing new depths in a game that has fascinated humanity for 1,500 years.