CAN AI DETECT FORGED ART?

CAN AI DETECT FORGED ART?

Tonight I would like to show you a beautiful painting. It has the title, lepore and has been attributed to the French impressionist. This painting belonged to Dr Arifitmar, an avid collector who owned more than 600 watts a lot. In 1939 he donated the painting to the Museum of Fine Arts in Osaka, where it can still be admired to this day. Now, just imagine you. are the museum creator and it's your responsibility to make sure that this beautiful painting goes through a proper authentication process before being put in the museum for the visitors to enjoy. In the past if your goal was to find out whether art pieces original or not, you will have to submit it to an expert for authentication. But experts require to see the art in original so you will have to pack your painting and ship it off to a different country for appraisal. So your museum wouldn't pay just for the evaluation itself, but also. for the transportation and for the insurance. Then you will have to wait for some months or sometimes even for some years to receive an answer. So this process is very slow, not not practical, expensive and particularly laborious. If a larger number of paintings need to be authenticated, But let us suppose that you did it. OK, so you are the curator. You shifted off to that expert and now you're waiting for his answer. How many of you think that this painting that we've just seen is authentic? Please raise your hand. Well, The answer to that question. is we don't know And the reason is that expert assessments are known to be fundamentally subjective And very often in practice, different experts have contradictory opinions so it can easily happen that you are going to an expert getting a positive answer, going to another expert, getting a negative answer. What do you do? You quickly reach an event. but now just imagine that there is a technology which can authenticate works of art simply by using a photograph taking with your smartphone. That technology actually exists. It uses artificial intelligence to learn the main features of an artist from photographs of authentic word command by that artist We can choose any artist, regardless of the style. The only condition is that he manifestly uses brush strokes to create his masterpieces. The process of learning the style of an artist is called training. And right now I just want to show you how the training works and I'll do it by very by using a very basic example, which is myself applying a stroke on that campus. OK, Let's just take a moment to look at this stroke. First of all, you see it's blue. It's just a color that I've chose. Then you might notice the lower part is a bit thicker. The upper part is a bit thinner. It has a bit of a curvature, not much. I didn't apply too much pressure on the canvas So all of these elements, they are unique to my own style, but if any of you here in this room would apply another brush stroke, it would look different. And that is really the basic principle underlying the training. The fact that the brush stroke of every human is unique, just like the signature, for example, or the handwriting. So if the AI would be presented with many of this masterpieces, you know, by myself, then it would learn the particularities of my own unique style. In practice, when training a new artist, we take all images of verified works of art by that artist and feed them into the program. I. dearly, we should have say at least 200 images to create a reliable data set. I know 200 doesn't sound like a huge number for artificial intelligence applications, but each image covered on the whole surface contains a large amount of information which can be processed by the AI. And furthermore, we split the images into smaller patches and zoom into every patch separately to make sure that we are capturing all the relevant information. from all this training images, the AI doesn't learn just the brush stroke, which is for sure the most important characteristic, but also other elements such as the use of colors or or object placement or or higher level compositional elements or even other characteristic characteristics which are really specific to the artist that we are training on and once the training is completed, the AI is able to compare the features learn during the training with those identified on the new work of art and based on that comparison, it returns a probability for the new artist. This technology has three main advantages. First and foremost, it is very precise. The AI has been tested many times by being asked to authenticate genuine art mixed with forgeries and has returned the correct answers. Which brings me to the second of advantage the objectivity. You know, we said that the art experts, they are quite subjective. They tend to become emotional, whereas the AI is fully unbiased and has no emotions. And the third advantage is the time. While the training itself can take several days checking whether a new art piece is by the same artist is just a matter of minutes. And now I'd like to return to this beautiful painting which you. as a museum curator, are in charge or having it authenticated. Let's just assume that after all, you've decided to submit an image for the AI to authenticate it instead of waiting for that expert. OK, the program needs about three days to learn photons made characteristics from around 700 training images and less than five minutes to calculate the probability for the authenticity of the new art piece. After awaiting time or say 34 days, you will receive an answer to your AI authentication query. Let's check this answer together. Great news. The program has evaluated the painting to be authentic with the 97% probability. This score is very high and is really no room for doubt. And therefore. you can finally take this wonderful painting and put it in your museum for the visitors to rejoin. in order to better understand and mostly visualize the results on the painting itself, we call we overlay so called heat map The region highlighted in bright red have been identified during the training as characteristic for the style of the artist Therefore confirming the authenticity of I'm convinced that such a technology offers huge possibilities. You know, back in 1939 that the curator of the Museum of Fine Arts in Lausanne had to go through the extensive process of verifying authenticity with an exploit But today every music. of the world can have their collections verified in a file affordable and objective manner. Every curator can find out immediately if the works he's bidding on are real or not And this will help to preserve our own cultural heritage. Thank you.