If you try to fill a 3×3 grid with only +1 and −1 so that every row multiplies to +1 and every column to −1, you’ll quickly run into a contradiction—no arrangement fits both rules. This puzzle, known as the magic square game, highlights a fundamental classical limitation. Now, Google Quantum AI’s team has used their superconducting quantum processors to show that quantum systems can sidestep this classical impossibility, and they've measured this effect in systems as large as 105 qubits.
The experiment reframes the puzzle as a game between two players, Alice and Bob. Each is given a row or column and must fill it according to the corresponding rule, without any communication. If their answers agree at the intersecting cell, they win the round. Any classical approach, which is essentially just writing all possible answers in advance, will always fail at least once in nine rounds. But with quantum resources, the odds change dramatically.
Surpassing classical strategies in the magic square game
On the Willow quantum processor, Alice and Bob each control two qubits and share two entangled Bell pairs. Instead of relying on predetermined answers, they use quantum measurements set for their assigned row or column. The source notes that classical strategies can win at most 88.89% of the time. In Google's experiment, their qubits achieved a win rate of 98.3% (0.9830 ± 0.0001) across 100,000 plays per version of the game, only losing about one in sixty rounds. This level of coordination is only possible because, in quantum mechanics, the outcome of a measurement can depend on the context—what other compatible measurements are performed at the same time.
The improvement over classical strategies isn’t just a curiosity. The experiment demonstrates that quantum contextuality—where measurement outcomes are not predetermined and depend on measurement context—can be harnessed on real hardware at scale. Noise and imperfections still cause occasional losses, but the quantum advantage is statistically clear.
Quantifying contextuality using hardware benchmarks
To go beyond the game and put a number on this quantum effect, the team used a Bell–Kochen–Specker inequality. This bundles the correlations from the game into a single value, ( \chi ), calculated as χ = R1 + R2 + R3 − C1 − C2 − C3, where R and C are the products of the rows and columns. Any classical system is limited to ; quantum mechanics allows values up to 6. What’s notable is that this test is state-independent: it doesn't matter how the qubits are prepared, so the outcome reflects the machine’s properties, not just a specially chosen input.




