2025-07-10
Until recently, the attribution of an artwork, i.e. assignment of its authorship, remained the domain of qualified specialists. In recent years, however, even art historians have recognised the potential of artificial intelligence for analysing images.
Tools using this technology are proving particularly useful when navigating the fast-growing online art market. In The Guardian, we were able to read about an algorithm that identified forged paintings by Claude Monet and Auguste Renoir on eBay from photographs. The prices of the forgeries reached hundreds of thousands of pounds.
Similar technology was used by Steven and Andrea Frank to assess the authenticity of 'Salvator Mundi', a painting attributed to Leonardo da Vinci sold for $450.3 million at auction at Christie's.
The programme, which recognises formal inaccuracies, is based on artificial neural networks – so-called convolutional neural networks (CNNs for short), for example, are used in face recognition technology.
However, they do not work perfectly and their implementation requires adequate preparation. The programme can only use a small area of the image so that it is not overloaded. It also needs a large amount of material to work with. This causes a problem when the artist to whom one wants to assign a work using the algorithm has only created a dozen or so works in his or her entire life.
Programmes that use neural networks to analyse images work by comparing multiple samples. But what if the artist was willing to experiment? So far, the technology has been used to establish the attribution of works of ancient art or modern representational art. Will it develop to the point where it also finds application in contemporary art?