AI joins the battle over painting authenticity: what's the current situation and what are the challenges?
Artificial intelligence (AI) is making its mark on the art world, even extending its reach into finicky fields like
AI meets Old Masters in the fight to authenticate paintings
https://www.ft.com/content/101f6fde-0817-4fc6-9fe8-2b8d4b0f2c65
Founded in 2019, Art Recognition is a company that provides an AI system capable of performing 'accurate and objective authentication of paintings.' The company has performed over 500 authentications to date, and has a track record of verifying works of art that have sparked controversy over their authenticity, such as Vincent van Gogh's self-portrait housed in the National Gallery of Art in Oslo .
Authentication determines whether a work is genuine or fake, determining its value and supporting the study of art. The painting 'The Adoration of the Kings,' which was thought to have been created by the workshop of Rembrandt Harmenzoon van Rijn , sold for 860,000 euros (approximately 140 million yen) at auction in 2021, but was later discovered to be a painting by the artist himself, and is expected to sell for 10.9 million pounds (approximately 2.07 billion yen) at auction in 2023.
According to Art Recognition CEO Karina Popovic, the company is also used by asset management services and legal professionals, and adds that 'the development of AI in the art field is being watched with interest by many in the art industry.'
Art Recognition - Shaping tomorrow's Art Authentication with AI
Joe Lawson-Tancred, author of ' AI and the Art Market ,' a book summarizing the relationship between AI and the art market, said, 'AI is good at pattern recognition, so if you show it enough works by a particular artist, it can learn their characteristics more easily than humans.' However, Tancred added, 'AI is not good at understanding context, so human reasoning is still essential.'
Carlo Milano, who works at London-based art gallery Callisto Fine Arts , said, 'For example, if a questionable catalogue raisonné is used to input data about an artist, the results could be questionable.' Milano explained that the work of an art dealer 'involves a wide range of psychology,' and said that using AI will provide more information and reduce errors, but added, 'it will not completely take over human work.'
Art conservators, on the other hand, are concerned about whether AI can take into account factors like dirty layers of varnish, wear and tear, and other factors. In general, they are very skeptical that AI can supplement or even replace the human eye when judging art.
In 2023, Art Recognition became embroiled in a furor over a painting called 'Brecciatondo,' attributed to
In January 2023, researchers at the University of Bradford and the University of Nottingham in the UK analyzed Bresci Tand using AI-assisted facial recognition software and determined that the face depicted in the painting was identical to that depicted in another Raphael painting, the Sistine Madonna, concluding that Bresci Tand was a Raphael painting. Similarly, Art Recognition also conducted an authentication test on Bresci Tand and determined that it was not a Raphael painting with an 85% probability of authenticity. This controversy highlights the advantages and disadvantages of AI-based authentication.
Subsequently, in December 2023, a research team at the University of Bradford published a peer-reviewed paper announcing new findings about Raphael's works. The team fed details of authentic Raphael paintings from a database into an AI program to examine the color tone, gradation, hue, and brushstroke patterns of Raphael's works. The results concluded that the face of Joseph in Raphael's 'Madonna della Rosa,' housed in the Prado Museum in Madrid, Spain, may not have been hand-painted by Raphael. According to Professor Hassan Ugair, director of the University of Bradford's Visual Computing Centre, the AI model they developed can authenticate Raphael's works with 98% accuracy.
However, Art Recognition CEO Popovich disputed the research team's findings after examining the training data for the AI model published by the University of Bradford research group. Popovich pointed out that the dataset the research team used to train the AI model did not include negative examples, such as 'artworks that are not Raphael's works but resemble his style.'
Like experts who learn from examples, an AI's ability to recognize and evaluate patterns depends heavily on how representative its training data is. 'Without exposure to both real and imitation examples of a subject, an AI will tend to classify work that resembles images from the positive training set as real,' Popovich said.
Professor Ugair responded by saying that there are other ways to train AI models besides the one mentioned by CEO Popovich, and added, 'This is not to say that AI will take away human jobs. It shows that AI can be an important tool, like spectroscopy and dating technology.'
Art historian Wendell Grosvenor said: 'AI can be useful to art connoisseurs. However, its main drawback at present is the quality of the input given to the AI appraisal programs currently in use. It is impossible to determine whether a painting is by Rubens based on a poor-quality image that is less than half the size of a Rubens painting. However, the same is true for human appraisers.'
Related Posts:
