We present an AI model that not only achieves cutting-edge precision for nuclear masses, but does so in an interpretable manner. For example, we find (and explain why) that the most important dimensions of its internal representation form a double helix, where the analog of the hydrogen bonds in DNA here link the number of protons and neutrons found in the most stable nucleus of each isotopic chain. Furthermore, we show that the AI predictions can be factorized and ordered hierarchically, with the most important terms corresponding to well-known symbolic models. Remarkably, the improvement of the AI model over symbolic ones can almost entirely be attributed to an observation made by Jaffe in 1969.

Mechanistic Interpretability promises a path toward fully understanding how neural networks make their predictions. We argue that neural networks can learn low-dimensional representations of their training data that are surprisingly faithful to human-derived domain knowledge. As a case study, we extract nuclear physics concepts by studying models trained to reproduce nuclear data.

We introduce Nuclear Co-Learned Representations (NuCLR), a deep learning model that predicts various nuclear observables, including binding and decay energies, and nuclear charge radii. The model is trained using a multi-task approach with shared representations and obtains state-of-the-art performance, achieving levels of precision that are crucial for understanding fundamental phenomena in nuclear (astro)physics.