Intelligent qubits?

Quantum hardware may be a good match for AI

New manuscript describes analyzing image data in a quantum processor.

John Timmer – Apr 11, 2025 10:46 am | 47

Credit: Jason Marz/Getty Images

Concerns about AI's energy use have a lot of people looking into ways to cut down on its power requirements. Many of these focus on hardware and software approaches that are pretty straightforward extensions of existing technologies. But a few technologies are much farther out there. One that's definitely in the latter category? Quantum computing.

In some ways, quantum hardware is a better match for some of the math that underlies AI than more traditional hardware. While the current quantum hardware is a bit too error-prone for the more elaborate AI models currently in use, researchers are starting to put the pieces in place to run AI models when the hardware is ready. This week, a couple of commercial interests are releasing a draft of a paper describing how to get classical image data into a quantum processor (actually, two different processors) and perform a basic AI image classification.

All of which gives us a great opportunity to discuss why quantum AI may be more than just hype.

Machine learning goes quantum

Just as there are many machine-learning techniques that fall under the AI umbrella, there are many ways to potentially use quantum computing to perform some aspect of an AI algorithm. Some are simply matters of math; some forms of machine learning require, for example, many matrix operations, which can be performed efficiently on quantum hardware. (Here is a good review of all the ways quantum hardware might help machine learning.)

But there are also ways in which the quantum hardware can be a good match for AI. One of the challenges of running AI on traditional computing hardware is that the processing and memory are separate. To run something like a neural network requires repeated trips to memory to look up which destination signals from one artificial neuron need to be sent to and what weight to assign each signal. This creates a major bottleneck.

Quantum computers don't have that sort of separation. While they could include some quantum memory, the data is generally housed directly in the qubits, while computation involves performing operations, called gates, directly on the qubits themselves. In fact, there has been a demonstration that, for supervised machine learning, where a system can learn to classify items after training on pre-classified data, a quantum system can outperform classical ones, even when the data being processed is housed on classical hardware.

This form of machine learning relies on what are called variational quantum circuits. This is a two-qubit gate operation that takes an additional factor that can be held on the classical side of the hardware and imparted to the qubits via the control signals that trigger the gate operation. You can think of this as analogous to the communications involved in a neural network, with the two-qubit gate operation equivalent to the passing of information between two artificial neurons and the factor analogous to the weight given to the signal.

That's exactly the system that a team from the Honda Research Institute worked on in collaboration with a quantum software company called Blue Qubit.

Pixels to qubits

The focus of the new work was mostly on how to get data from the classical world into the quantum system for characterization. But the researchers ended up testing the results on two different quantum processors.

The problem they were testing is one of image classification. The raw material was from the Honda Scenes dataset, which has images taken from roughly 80 hours of driving in Northern California; the images are tagged with information about what's in the scene. And the question the researchers wanted the machine learning to handle was a simple one: Is it snowing in the scene?

All the images were sitting on classical hardware, of course. To classify an image on quantum hardware, it had to be converted to quantum information for processing. The team tried three methods of encoding the data, which differed in terms of how the pixels of the images were sliced up and how many qubits the resulting slices were sent to. The researchers used a classical simulator of a quantum processor to do the training steps, which identified the appropriate numbers—again, think in terms of the weights of a neural network—to use during the two-qubit gate operations.

They then ran the hardware on two different quantum processors. One, from IBM, has a lot of qubits (156) but a slightly higher error rate during gate operations. The second is from Quantinuum and is notable for having a very low error rate during operations, but it only has 56 qubits. In general, the accuracy of the classification went up as the researchers used more qubits or as they ran more gates.

In general, though, the system worked; accuracies were well above what you'd expect from random chance. At the same time, they were generally lower than what you'd get from a standard algorithm run on normal hardware. We're still not at the point where existing hardware has both enough qubits and a low enough error rate to be competitive on classical hardware. Still, the work was clearly able to show that real-world quantum hardware is capable of running the sorts of AI algorithms that people have been expecting it to. But like everyone else, people hoping to solve useful problems will have to wait for further improvements on the hardware side.

John Timmer Senior Science Editor

John is Ars Technica's science editor. He has a Bachelor of Arts in Biochemistry from Columbia University, and a Ph.D. in Molecular and Cell Biology from the University of California, Berkeley. When physically separated from his keyboard, he tends to seek out a bicycle, or a scenic location for communing with his hiking boots.

47 Comments