AI-Assisted Descriptor Discovery for Electrochemical Interfacial States via Latent Organization of High-Throughput EIS

Measurements do not automatically yield meaningful descriptors when existing theory cannot uniquely determine how complex observations should be reduced. Using high-throughput electrochemical impedance spectroscopy as a model case, we present a human-guided, AI-assisted workflow in which a conditional autoencoder organizes electrochemical responses into a latent space for descriptor construction and comparison, while experimental potential of zero charge (PZC) is used only as an external validation reference. The learned representation remained strongly ordered by electrochemical bias, enabling the construction of trajectory-derived descriptor candidates indexed by applied bias ($V_b$). Grounded comparison among these candidates identified an anchor-relative sine-like trajectory coordinate as the most credible scalar summary within the tested family. These results show that latent organization can support auditable descriptor discovery from complex electrochemical measurements in underdetermined settings.