A Renoir? Artificial intelligence says yes (probably)
November 20 2022
Picture: Guardian
I recently went to a conference organised by the Art Loss Register on authenticity in art, and whether science and artificial intelligence can do better than existing connoisseurship. Naturally, I made the case for connoisseurship (after, as ever, the examples of a few bad connoisseurs were used to tarnish connoisseurship in general), but my conclusion was that we need to worry less about worrying about authenticity itself. Not that it isn't important whether, say, Leonardo da Vinci painted the Salvator Mundi - it really is - but that we should accept we can't be entirely certain. In the past, an attribution, whether given by science or connoisseurship, has tended to say 'this is a Leonardo' with an implied probability of 100%, when it should probably be more like 80-90%. This is especially the case for major artists who employed studio assistants.
Anyway, one of the speakers at the conference was Carina Popovici. CEO of Art Recognition, a new firm based in Switzerland offering certificates of authenticity based on the verdict of a computer programme, or as we must call it now, Artificial Intelligence. I'm in no doubt AI can help with attributions, but am wary of systems which imply they can determine who painted what based simply on an uploaded iPhone photo. If I was to declaim on an attribution based just on a phone photo, I'd rightly be criticised. A recent example of Art Recognition's decision, on the above disputed Renoir (the Wildenstein Institute says 'non', AI says 80% yes), is covered by Linda Geddes in The Guardian here.
