A recent AI-based attribution of a Rubens painting has reignited the debate about the opportunities and limits of the technology in the art world. The work in question is Diana Bathing, long regarded by many scholars as a copy of Rubens.
The Swiss authentication firm Art Recognition now suggests that the master himself may have created the canvas, at least in part. CEO Carina Popovici explains that, in addition to training their algorithm on Rubens’ undisputed works, they also fed it “images of copies, imitations, and pieces produced by followers. From all this training material, the AI learnt Rubens’ unique style and, just as importantly, how to distinguish genuine paintings from skilful imitations.”
The analysis concluded that Rubens painted some sections of the picture with a probability greater than 80 per percent. Yet the finding that the entire canvas may not be by his hand is hardly unusual: like many painters of his era, Rubens ran a large studio, and assistants often copied and completed his compositions.
But not everyone is convinced. Nils Büttner, president of the Centrum Rubenianum – the leading authority on Rubens’ oeuvre and catalogue raisonné – disagrees. Büttner cites technical inconsistencies and, to his critical eye, a level of execution too low to warrant attribution to the artist. How, then, to reconcile these views?
Traditionally, artworks are authenticated through three main avenues. First, the connoisseur’s eye, which relies on visual and mental comparison and the aesthetic sensitivity of seasoned experts. Second, provenance research, which traces historical records, catalogue raisonnés (where available), exhibitions, and past transactions. Third, scientific analysis, employing methods such as carbon dating, pigment examination, and X-ray fluorescence to verify whether a work’s materials match those of the period to which it is ascribed.
Although these approaches are often combined, connoisseurship still carries decisive weight in the art market. As nineteenth-century painting specialist Rachel Kaminsky notes, “The intuitive component occurs within the first few seconds an expert stands before a painting. Almost instantaneously, in the blink of an eye, the brain processes an enormous amount of information… It’s hard to explain how this happens, yet these instant reactions are, surprisingly, often correct.” This intuition, though subjective, is widely accepted in the art world and can be pivotal in legal disputes over authenticity and, of course, in the pricing of works.
Just a few months ago, for example, the LMI Group announced it had found a “lost” Van Gogh purchased for under $50 that could fetch US$15 million if authenticated. The Van Gogh Museum swiftly denied the attribution in a brief statement without physically examining the piece, while the company had invested more than four years, US$30,000, a nearly 500-page report, over fifteen specialists, and a blend of advanced scientific methods and archival research. Despite that rigour, when the acknowledged Van Gogh authority says a work is not genuine, the resulting cloud of doubt is hard to dispel. So, while the judgment is subjective, there is little room for debate once the expert declines to recognise a painting, and the impact of that stance on the market and potential legal battles is immense.
It would seem, then, that certain roles in the art world remain reserved for human perception. Yet it is precisely this expert eye that is now being complemented by sophisticated AI as a new tool for authentication. The question is whether the technology currently has – or will soon develop – the capacity to rival professionals and alter the evidentiary landscape.
AI may well uncover patterns imperceptible to humans, decentralise expertise, cut costs, and, perhaps most intriguingly, bring unprecedented transparency to a market where up to 40 per cent of works are estimated to be forgeries. Still, its accuracy depends on the quality, diversity, and granularity of the training data. It also confronts a sector in which human subtlety, experience, and intuition have long been the ultimate arbiters of authenticity.
Diana Bathing will not be the last case of its kind. While AI has yet to earn the full trust of the art market, it places a powerful new instrument on the table. Will this task, until now entrusted primarily to humans, also be swept up in the unstoppable technological tide?
Access the Spanish version here.
