Scientists Propose Quantum Computers Could Generate Data to Train AI For Chemistry

Two researchers from IonQ and Microsoft have proposed that quantum computers could generate highly accurate data to train artificial intelligence models for simulating chemical systems. Chi Chen of IonQ and Matthias Troyer of Microsoft described the hybrid approach in an essay published in IEEE Spectrum. Quantum computers would produce precise details on electron behavior in molecules. The data would then train AI models running on classical computers to deliver faster predictions. The idea targets limitations in computational chemistry. Classical methods such as density functional theory rely on approximations that reduce accuracy for complex electron interactions. Exact modeling of all possible electron arrangements grows too demanding for conventional computers as molecules increase in size. Quantum computers could supply small volumes of highly accurate training data that would be prohibitively expensive to generate classically. Once trained, the AI models would estimate chemical properties at much higher speed. The researchers indicate the method could support design of catalysts, batteries and other materials while large-scale fault-tolerant quantum computers remain under development. The essay cites recent work by Microsoft and Pacific Northwest National Laboratory as an illustration of AI-assisted materials research. Models evaluated more than 32 million candidate materials for battery electrolytes. The screening reduced the set to roughly 800 promising options in less than a week. Traditional computational approaches would have required decades for the same task.