AI Models Revolutionize Research on Quantum Systems

Since the formulation of the Schrödinger equation, quantum mechanics has driven the development of transformative technologies such as semiconductors and lasers, laying the foundation for the modern information society. Entering the 21st century, the so-called “it from qubit” paradigm of the second quantum revolution is gaining momentum, with quantum computing, quantum communication, and quantum sensing transitioning from laboratory research toward real-world applications. These fields are widely regarded as key engines of the next technological revolution. However, as quantum systems grow in scale, a fundamental challenge has become increasingly apparent: how to efficiently describe and characterize complex quantum systems. The Hilbert space dimension of a quantum state grows exponentially with the number of qubits, rendering traditional methods ineffective for large-scale quantum systems—posing a major bottleneck to further progress in quantum information science. In 2022, the emergence of large AI models like ChatGPT not only accelerated the practical deployment of artificial intelligence but also significantly advanced the field of AI for Science. Problems once considered intractable in quantum information science are now being addressed through novel data-driven approaches. Over the past three years, data-driven AI models have demonstrated remarkable efficiency in learning large-scale quantum systems, shifting the research paradigm from one based on prior theoretical knowledge to a data-driven framework. This transformation has dramatically improved the efficiency of learning quantum systems generated by quantum simulations and quantum computers. Driven by a strong belief in the potential of AI for quantum science, Associate Professor Ya-Dong Wu from Shanghai Jiao Tong University has been conducting continuous research in this area over the past three years, aiming to establish AI as a new paradigm in quantum research. In 2024, Wu was invited to visit Singapore, where he met co-author Yu-Xuan Du for the first time in person through a mutual friend—ending a long period of intellectual exchange conducted solely through reading each other’s work. Their in-person conversation revealed a shared recognition that the field had made significant cross-disciplinary advances over the past three years: from statistical learning models to deep neural networks, and now to large foundation models, all have taken center stage. Yet they also identified a critical gap—a systematic, comprehensive review was still missing, making the field difficult to access for newcomers and researchers from other disciplines. This realization led them to collaborate on a comprehensive review paper, systematically exploring how AI can be used to discover and predict the physical properties of large-scale quantum systems. The paper reviews three major categories of AI models—machine learning models, deep learning models, and foundation models—in the context of understanding and representing large quantum systems. These systems include complex quantum states generated by quantum simulations and results produced during quantum computation. The review highlights two key roles AI plays: first, enabling more accurate prediction of quantum system properties; and second, constructing classical surrogate models—simplified representations that approximate quantum behavior, thereby reducing reliance on costly experiments. These models allow researchers to study quantum systems efficiently without directly confronting the exponential complexity of Hilbert space. The paper also outlines current challenges in the field and discusses future directions, particularly the potential of cutting-edge AI techniques to accelerate quantum computing research. Looking ahead, AI models for quantum science are expected to play vital roles in multiple areas. They could enable faster and more accurate calibration of experimental quantum systems, improve the efficiency of quantum circuit execution, and autonomously discover new quantum phases within vast parameter spaces. Even more promising is the prospect of AI becoming a “virtual scientist”—capable of identifying patterns in experimental data that humans might overlook, and even designing novel quantum experiments to explore previously unknown quantum phenomena. If realized, AI would evolve from a mere research tool into a key partner driving breakthroughs in quantum science. In the early stages of writing, the authors conducted an extensive literature review, mapped the evolution of the field, and established a framework centered on machine learning, deep learning, and language models. Given the highly interdisciplinary nature of the topic, balancing technical depth for AI researchers with clarity for physicists proved challenging. The team iteratively revised the manuscript, adjusting its focus and structure multiple times. Recognizing that a single perspective was insufficient, they invited several leading experts from both fields to contribute. The team includes renowned figures from physics—such as Barry Sanders, a member of the Royal Society of Canada, Jens Eisert from Freie Universität Berlin (a leading expert in quantum information and many-body physics), and Giulio Chiribella from the University of Hong Kong (a pioneer in quantum information and quantum causal structures)—as well as top AI researchers, including Dacheng Tao, an Australian Academy of Science Fellow and IEEE Fellow. Throughout the revision process, intense discussions arose between AI and quantum physics perspectives. Scholars from North America, Europe, and Asia brought diverse insights, helping the team refine a coherent classification system and develop clear, accessible language to convey complex concepts to a broad audience. Although the initial draft was completed in just two months, the extensive revisions and peer feedback—aimed at achieving high quality—continued for nearly ten months before the final version was completed. The authors plan to further explore the application of advanced AI, particularly foundation models, in quantum information and quantum computing. Future work will focus on areas such as learning complex quantum systems, optimizing quantum hardware design, improving quantum circuit performance, and advancing quantum error correction. Reference: https://arxiv.org/pdf/2509.04923 Editorial/Formatting: He Chenlong