For centuries, art authentication has exclusively relied on the eye and expertise of art historians and connoisseurs. These experts examine the minute details in a work of art including subject matter and continuities in the artists’ oeuvre in terms of palette, brushwork and temperament. Every individual connoisseur, invariably, focusses on the unique characteristics of an artwork informed by their idiosyncratic mental dataset, each connoisseur having been exposed to a different selection of artworks throughout their lifetime and in their own specific scholarly context. No connoisseur looks or thinks in the same way as the next. As the judge, Ingeborg Baümer-Kurandt aptly claimed in the 2018 Wiesbaden forgery scandal involving major examples of Russian avant-garde art, “ask 10 different art historians the same question and you get 10 different answers” (1). Although originally articulated in the context of litigation and the, oftentimes, diverse vested interests of experts, this quotation points to idiosyncrasies of connoisseurship and expertise more generally.
In the same way as how each connoisseur looks differently at any artwork, Pictology analyses high-resolution images of artworks to recognise formal patterns invisible to the human eye. Whereas a connoisseur might focus on brushwork, our stylometric algorithms identify key features of artworks, translated into quantifiable data that does not correspond to conventional categories of connoisseurial analysis.
There have been serious concerns expressed that AI poses a threat to replace connoisseurship, but this is quite simply untrue. AI offers an additional means of analysing an artwork which is separate from connoisseurship. After all, AI and connoisseurs are useful to divergent ends – no connoisseur simply conducts a mechanical analysis of an artwork in the way that AI does. Connoisseurship, in this way, will always be integral for a conclusive and reliable attribution to be made for an artwork, and it is for this reason that AI only forms one part of Hephaestus’ authentication protocol; a protocol that also consists of connoisseurship, provenance research and scientific analysis.
AI is being increasingly employed across sectors as a means of supplementing human analyses. In the healthcare sector, for example, AI is being used as a powerful tool to provide fast, accurate and affordable solutions for screening programs including breast cancer screening, targeted lung health checks and diabetic eye screenings to name a few. The applicability of the technology extends to sports, where, for example, in Hawk-Eye computer vision systems, a statistical model is employed, based on inputs provided by cameras across the stadium, to determine, based on probability, where the ball touched the ground.
Across these diverse applications of AI, outputs are calculated from a statistical analysis of input data. In this way, through training Pictology, our unique stylometric algorithms, on a data set of bona fide authentic artworks, we can calculate the likelihood of an artwork’s authenticity accurately, in statistical terms. A connoisseur might notice obvious differences at a glance but Pictology can identify subtle, numerical inconsistencies that indicate as to the authenticity of an artwork. Of course, Pictology cannot evaluate and communicate the beauty, emotional depth or importance of an artwork in the same way as a specialist connoisseur but it can analyse data patterns impossible for the human eye to identify. Unaffected by biases or pecuniary conflicts of interest, Pictology supplements traditional and scientific forms of analysis to enable a wholly conclusive and reliable attribution.
(1) Quoted in Catherine Hickley, ‘Unknown or Unreal? The Shadow on Some Russian Avant-Garde Art’, The New York Times, April 6, 2018 <available at: https://www.nytimes.com/2018/04/06/arts/design/russian-avant-garde-art-forgery-disputes.html>
