Nvidia scientists and their counterparts at a range of academic, scientific, and quantum computing institutions late last year published a research paper about the ongoing convergence of AI and quantum computing, and how each will play increasingly important roles in the development of the other.
“There is ample intuition to motivate exploring AI as a breakthrough tool for [quantum computing],” the researchers wrote in the paper published in Nature Communications in December. “The inherent nonlinear complexity of quantum mechanical systems makes them well-suited to the high-dimensional pattern recognition capabilities and inherent scalability of existing and emerging AI techniques
.”
This AI-quantum partnership is a central part of the larger classical-quantum hybrid datacenters that will be standard operating environments as quantum computing matures. It manifests itself in myriad ways, such as classical AI handling data cleaning, encoding, and optimization while the quantum system performs calculations.
A key area is the use of AI in solving the challenge of error correction in quantum computing. Qubits are notoriously fragile and can essentially fall apart – or decohere – due to environment factors around them, from noise and temperature to light and the activity of other qubits. This can lead to errors, and error-prone quantum systems are neither practical nor useful.
AI For QEC
A lot of work is being done to see how AI – both generative and agentic – can help address error correction in quantum systems. For example, scientists from the University of Pennsylvania and Hao Tang of Peking University in China addressed the issue in a research paper early last year. Nvidia in April ran out Ising, a family of open source AI models that the vendor said can provide up to 2.5 times faster performance and three times higher accuracy for the decoding process for quantum error correction. Nvidia co-founder and chief executive officer Jensen Huang said that “with Ising, AI becomes the control plane – the operating system of quantum machines – transforming fragile qubits to scalable and reliable quantum-GPU systems.”
The same month, researchers with Amazon Web Services, Quantum Elements, the University of Southern California, and Harvard University wrote about their use of AI digital twin technology and a cloud-based HPC system to create and run simulation models to advance research into quantum error correction.
In another paper, IBM researchers in June outlined a large-language model (LLM)-based framework that can sort through thousands of code variations used to solve the error correction problem to find the right one for a particular situation. It’s “one example of the growing two-way interplay between quantum computing and classical AI, where each is beginning to inform and accelerate the other,” they wrote.
This month, a group of researchers led by Volodymyr Sivak, a research scientist with Google Quantum AI, said that one of AI’s features – its ability to learn from experience – could be used to enable quantum systems to run for longer periods of time – days, weeks, or even months – without having to stop their calculations to adjust calibrations when errors are detected.
The reinforcement learning (RL) method the researchers detailed in a study published in Nature, illustrated below, uses the encoded information that the quantum computers already collect as they monitor for errors to run them through an AI system that can learn from the errors and make adjustments to the quantum system’s control settings and operating parameters based on what it found.
Fault-tolerant quantum error correction relies on errors leaving detectable signatures that can be detected, signaling that an error has occurred in a particular region of the circuit. A decoder can then look at the detection events to suss out likely error patterns. After the decoder corrects the logical state, the remaining error is quantified using the logical error rate (LER), which the researchers said is the principal way of measuring the quality of the quantum error rate process.
Reinforcement learning has been used successfully in other areas of complex computing, from robotics to refining the behavior of LLMs, the researchers noted.
Keeping The Quantum Computers Running
In this case, the method addresses a challenge that quantum computers face when it comes to error. When errors are detected, the systems have to be periodically stopped and recalibrated, interrupting the calculations they were running, which can hobble their ability to run the kinds of complex, long-term jobs they’ll be asked to do as the industry moves into the era of practical commercial and fault-tolerant quantum computers.
Reinforcement learning essentially lets a quantum computer make the necessary adjustments needed to correct errors on the fly while continuing to run its calculations.
The researchers created a framework for Google’s Willow superconducting quantum chip (below), which the hyperscaler introduced in December 2024, touting breakthroughs at the time for error correction.
They introduced hardware drift – when a system’s physical properties, like qubit transition frequencies or gate resonance parameters, change gradually over time – and used the AI learning algorithm to continually ingest the error data and make adjustments based on them.
The researchers found that when using the technique, logical error rates were made 3.5-times more stable and reduced them by some 20 percent when compared with traditional recalibration methods. At the same time, the system’s performance remained steady even as the reinforcement learning was underway.
“The agent is able to reach high performance even starting from randomized initial control parameters, suggesting the potential to augment or replace elements of the traditional calibration stack,” the researchers wrote.
The group tested the framework over multiple surface and color error-correcting codes, with the agentic framework able to manage more than 1,000 control parameters. In numerical simulations, the researchers found that the framework could manage as many as 40,000 of control parameters.
Going forward, the reinforcement learning framework can be enhanced so that “it is conceivable that in the future, with sufficient enhancement of our learning framework, a quantum processor could be calibrated for QEC [quantum error correction] ab initio fully by RL, with no reliance on the traditional calibration paradigm or human experts,” the researchers wrote. “By empowering the quantum computer to learn from its errors, we unlock a scalable pathway to optimize performance in real time, replacing disruptive calibration routines with uninterrupted computation. Our work suggests that the path to fault tolerance will be built not only on better hardware but on more intelligent control.”