A research team from the Xinjiang Astronomical Observatory (XAO) of the Chinese Academy of Sciences has developed an interpretable artificial intelligence framework called the Convolutional Kolmogorov–Arnold Network (CKAN), offering new insights into the nature of dark matter at the scale of galaxy clusters. Unlike traditional black-box AI models, CKAN is designed to provide transparent and understandable predictions, allowing scientists to examine how and why the model arrives at its conclusions. This interpretability is crucial in astrophysics, where understanding the underlying physical mechanisms is as important as making accurate predictions. The team applied CKAN to analyze gravitational lensing data—distortions in light caused by massive structures like galaxy clusters—which serve as a key probe of dark matter distribution. By leveraging the model’s ability to reveal hidden patterns in complex data, the researchers were able to infer properties of dark matter, such as its spatial distribution and potential non-ideal behavior, that challenge some standard assumptions in cosmology. The results suggest that dark matter may not be as smoothly distributed as previously thought, and could exhibit more complex structures or interactions at large scales. The success of CKAN marks a significant step forward in using AI not just to process vast astronomical datasets, but to actively contribute to scientific discovery in a way that aligns with the principles of physical reasoning. The framework could become a valuable tool for future studies of dark matter, cosmology, and large-scale structure formation in the universe.