A few hundred years ago, authenticating whether a painting was created by a master like Rembrandt, or a talented copy-cat, was dependent only on the trained eye of an expert. Then, new tools like X-rays and infrared imaging allowed curators to see beneath the surface of a painting and analyze details invisible to the human eye.
Now, researchers at Carnegie Mellon University are exploring how artificial intelligence can work alongside human experts to add another layer of evidence towards the unsolved mysteries in art attribution.
Published in the Proceedings of the National Academy of Sciences and led by mechanical engineering professors Phil LeDuc and Jon Cagan, alongside alumni Andrés Arias-Rosales and renowned Rembrandt scholar Arthur Wheelock Jr., the researchers have developed an AI framework that can examine a painting for visual patterns associated with the work of Rembrandt to help experts make the final call on a painting's authenticity.
Attributing a painting to an artist like Rembrandt is no easy task. Experts consider everything from historical documents to the artist's subject matter, his varied styles of painting throughout the course of his career, including brushwork and the types of paints and supports he used. On top of that, they must keep in mind that Rembrandt's pupils and assistants may have sometimes painted alongside him, and that later restorers may have changed and added elements to Rembrandt's compositions. To help them in their analyses, experts also consider technical information gained from such examination techniques as X-radiography, infra-reflectography, and X-ray fluorescence analysis.
For an art collector, or insurer, the stakes are high. A true Rembrandt painting is a piece of history and cultural identity. To get it right, the engineering team learned from Wheelock, who spent more than forty years as a curator at the National Gallery of Art in Washington, D.C. and is an internationally recognized expert on Rembrandt and Dutch art.
"Working with him was a dream of mine," Arias-Rosales said. "We wanted to understand Rembrandt through the eyes of a world-class expert, then understand it through the eyes of 'the machine', our AI approach, and bring those together." Prior to diving into developing the new AI tools, it was crucial for the team to first go to the National Gallery of Art, DC, to study Rembrandt paintings and engage in in-depth discussions about the artist, his style, technique, and the questions and challenges surrounding his work.
The team then asked whether an AI approach could distinguish Rembrandt from paintings by artists whose work resembles that of the master. The team carefully curated a dataset of over thirteen hundred high resolution digitized images of paintings including 311 Rembrandts that were used for the AI analysis.
The AI model learned to identify statistical patterns across characteristics such as edges, color, light distribution, and visible brushstrokes to produce a score estimating how strongly a painting aligns with the patterns shown in Rembrandt's work. When the model was highly confident, it classified paintings with 95.8% accuracy.
"Experts need to understand how the AI model/algorithm came to its decision," said Cagan. "That is critical in successful human-AI collaboration, so we needed to make sure the algorithm could expose which areas of the painting influenced the algorithm's conclusion."
To do so, the AI approach classifies a painting as either Rembrandt or not, provides carefully calibrated probabilities and confidence scores, and generates visualizations that emphasize the painting areas and details that had the strongest influence in the AI-based classification. This becomes a useful tool for curators as they can focus on specific details within the painting that they may have otherwise overlooked. For example, if the AI was repeatedly highlighting the hands with a low score, suggesting that they likely weren't painted by Rembrandt, the researchers can digitally remove them from consideration and ask the model to classify it again. If the hands disappear, and there is high confidence in a Rembrandt attribution, a curator may consider whether the hands were painted by a Rembrandt student instead.
"The most exciting cases are the ambiguous ones," said Arias-Rosales. "We wanted to do more than bring opaque or simplistic answers to Wheelock; we wanted to bring him more questions and point to potentially interesting details for him to consider and help solve mysteries of art history."
The team has already tested their AI on a real mystery. In 2024 a painting titled Portrait of a Girl was discovered in an attic in Maine, and later sold for $1.41 million because of its possible connection to Rembrandt. When the CMU model analyzed the image, it determined an 84.7% confidence that the painting was a Rembrandt. "These results aren't an authentication, they are a statistical observation meant to be considered alongside historical research, material analysis, X-rays, and other tools with the final decision made by human experts," said LeDuc.
Eventually, the team hopes this tool could become an interactive application for museums, curators and other art professionals. With Rembrandt as the test case, they believe it could be adapted to investigate other artists as well.
For LeDuc this project underlines how Carnegie Mellon's interdisciplinary culture shines.
"Not everywhere will they bring together engineering, design, and art history to solve intriguing problems," he said. "It's such a CMU thing. We like to try to help with exciting fields in many different directions"
