Quantum computing’s AlphaGo moment

Back in 2016, DeepMind’s AlphaGo beat the world’s best human player at Go, a game long considered too complex for machines to master. It wasn’t the moment AI suddenly became useful to everyone – we had to wait until 2023 for that – but it showed that a computer could do something that it wasn’t supposed to be able to do.

Quantum computing has now had its version of that moment.

Understanding how energy and information flows through matter

In a joint demonstration with IBM, we designed a model of heterogeneous quantum matter, the messy, disordered matter that real things are actually made of, rather than the perfect crystalline order assumed in most simulations

Run on an IBM Quantum Heron processor, the model turned the machine itself into a programmable stand-in for that material, allowing researchers to see how information moves through it.

Rather than just running a simulation, the model lets researchers engineer the microscopic structure and couplings of a material, controlling where information flows, localizes or interferes as it would in a real substance. It is engineering matter, not just modeling it, precise, verified control over one of the most complex kinds of physical system there is. This is an idea Richard Feynman first envisioned in 1982, of simulating quantum matter on a reconfigurable quantum processor built from the same physics.

For eight months and counting, the world’s leading classical simulation methods have not been able to reliably reproduce the result. We believe this is the strongest quantum advantage claim published to date: a quantum computer doing something no classical method can match.

To see why, let’s return to Go. The number of possible board combinations in Go is around 10¹⁷⁰, greater than the number of atoms in the observable universe. For that reason, evaluating every move by brute force is computationally impossible. But AlphaGo still won on an ordinary classical computer; it just used a cleverer method, combining neural networks and a search algorithm. The difference is that Go’s complexity has a 19x19 board to constrain it. Quantum matter does not. The number of states increases exponentially with every particle you add. And, unlike Go, the best classical strategies break down here.

More than that, it is a problem that matters. How information, energy and particles travel through real materials governs things like catalysts and battery electrolytes, the building blocks of real-world products that are extremely difficult and expensive to model classically.

This is what makes it a demonstration of useful quantum advantage. It is a physically meaningful insight that no known classical method can reproduce, with real downstream value for material design. Researchers can design materials and architectures that exploit, rather than avoid, disorder.

A new way to trust quantum results

But how do we trust this result when no classical computer can verify it? That’s the second part of this breakthrough.

In AlphaGo’s second game, the AI played a move – move 37 – that the best players in the world thought was a glitch, because none of them would have played it. But as the game went on, it became apparent that move 37 was the one that gave AlphaGo the game. For a short while, however, no one knew if the move was genius or a mistake.

Quantum computing now has a similar uncertainty, and one that can’t be resolved so easily, because no expert human player or classical computer can play the position to tell us whether we’re right.

Quantum results have, so far, been verified by checking them against a classical simulation. So what happens once you go beyond what a classical computer can do, as we have here? When we tested the best classical methods on this problem, in collaboration with world-leading classical simulation researchers, including those at the Simons Foundation’s Flatiron Institute and EPFL, their methods disagreed with one another.

So we created a new framework of trust, by stress-testing the processor. We deliberately changed the noise affecting the quantum circuits and ran the problem on multiple IBM Quantum processors, and the result remained stable, showing that the quantum computers were producing outputs we could trust.

AlphaGo was heralded at the time as a breakthrough moment for AI. But what has followed in the decade since has been even more remarkable. The technology has moved beyond theoretical demonstrations and is now recognized as both useful and commercially valuable, with hundreds of applications across every sector.

I believe that, in time, today’s announcement will be seen as equally significant for quantum computing. It is the moment the technology can begin to move beyond academic research and start to prove its utility and commercial value across industries, from life sciences and aerospace to chemical and material manufacturing, energy and finance.