
Andrés Arias–Rosales, Arthur K. Wheelock, J. Kenneth Moritz, Philip R. LeDuc, Jonathan Cagan · Proceedings of the National Academy of Sciences 2026 · 2026
DOI: 10.1073/pnas.2626989123
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
With questions that persist for centuries, art attribution is a complex detective effort that shapes cultural identity. Our goal is to complement human expertise, providing statistical and visual arguments that shed new light on discussions about great masters like Rembrandt and beyond, with long-term implications for art analysis and human-AI collaboration. We present an AI-driven explainable approach to assist painting attribution with up to 97.9% recall, 95.8% accuracy when highly confident, and methods that identify the most relevant areas. This work supports human-AI collaboration by incorporating expert guidance, expanding the outputs with explainable visualizations, tailoring settings to users' needs, thorough calibrations to avoid overconfidence, and building a dataset that balances quality with quantity and focuses on comparable artists like a master's pupils and teachers.
No comments yet — start the discussion below.