Rembrandt through the eyes of the machine: Explainable Human–AI collaboration in art attribution

What is it about?

We developed an AI-based approach to assist in centuries-old, culturally consequential painting attribution discussions, such as whether a painting is a genuine Rembrandt. Seeking to support and expand human-AI collaboration, we propose a 7-step explainability framework to help explore the reasons why our AI model classifies a painting a certain way. Our approach can help experts raise new questions, identify interesting relationships, and discover details that help statistically characterize a genuine painting by a great master.

Featured Image

Photo by Meizhi Lang on Unsplash

Photo by Meizhi Lang on Unsplash

Why is it important?

Much as X-rays gave art experts new ways to examine paintings, AI introduces a layer of analysis based on statistical pattern-finding that can reveal potential new evidence for or against an attribution. Our approach assists for making these patterns visible, helping experts identify subtle distinctions among artists. Addressing common limitations, we go beyond the "black-box” toward explainability, improve the quantity-quality data balance, test data-separation robustness, and implement careful calibrations.