Artificial Intelligence has progressed to the point of making possible to create pieces of music, novels and newspapers articles, paintings… imitating the style of a given artist to the point that it can fool “experts” (take a look at the clip, where original and AI Bach pieces are played – can you tell which is which?). You can even give a try and create some AI painting masterpiece by yourself, or some new piece of music to go along with your new slide show, … And, just few days ago I spotted a news of a research team using AI to complete the unfinished Beethoven 10th symphony. You can listen it here.
It turns out that we can use AI to spot fake paintings… That is the case with the famous Samson and Dalilah painting at the London’s National Gallery. That is one of the most famous paintings by Peter Paul Rubens and when it was bought by the National Gallery it was the third most expensive artwork ever purchased at an auction.
Art Recognition, a Swiss company is providing services to analyse the authenticity of works of art and they do this by using their artificial intelligence based software. This software is trained on artworks of the artist you assume created the artwork and then apply the trained software to the one you want to assess.
The training takes place analysing both the complete artwork as well as hundreds of fragments (small areas of a painting) using deep convolutional neural networks.
Applying their software to the Samson and Dalilah they turned out a probability of 90% that such painting is a fake. Very bad news for the London’s National Gallery. This assessment is giving strengths to doubts expressed by few experts on the authenticity of that painting.
Indeed AI based software is now available to spot alteration created by Photoshop and other photo editing software (basically 100% of the photos you see on fashion magazines have been altered!). On the one hand AI helps creating fakes, on the other it helps spotting them: I guess it is just fair.