By analyzing a single brushstroke, this AI is able to pick a real Picasso from the fakes

Artificial intelligence is already able to imitate the work of great artists, so why shouldn’t it also be able to spot genuine works from forgeries? In a new paper, researches at Rutgers University in New Jersey and the Atelier for Restoration & Research of Paintings in the Netherlands examined how machine learning can be harnessed to more effectively spot fakes.

The researchers tested the AI using a data set of 300 digitized drawings consisting of over 80,000 strokes from artists including Pablo Picasso, Henry Matisse and Egon Schiele, among others. Using a deep recurrent neural network (RNN), the AI was able to learn which strokes were typical of each artist, and then used that information to make educated guesses.

The results showed that the AI was able to identify the individual strokes with an accuracy of between 70 to 90%. The researchers also commissioned artists to create fake drawings similar to the originals in the AI’s data set, and in most test settings it was able to detect forgeries with 100% accuracy, simply by looking at a single brushstroke.

The use of artificial intelligence in art has the potential to make the work of art historians much easier, who currently use time-consuming techniques like infrared spectroscopy and radiometric dating when verifying the origins of a piece of art.

“It is quit[e] a tedious process for a human to trace individual strokes to provide segmentation of them, specially [sic] such task requires [a] certain level of expertise,” the researchers write in the paper.

While the results are impressive, the AI is not yet perfect. For instance, it cannot identify paintings where the brushstrokes are not visible. For this, the researchers say they plan to conduct further experiments using impressionist paintings to bolster their current results.


Lead Image: Clem Onojeghuo | Unsplash

Artificial intelligence is already able to imitate the work of great artists, so why shouldn’t it also be able to spot genuine works from forgeries? In a new paper, researches at Rutgers University in New Jersey and the Atelier for Restoration & Research of Paintings in the Netherlands examined how machine learning can be harnessed to more effectively spot fakes.

The researchers tested the AI using a data set of 300 digitized drawings consisting of over 80,000 strokes from artists including Pablo Picasso, Henry Matisse and Egon Schiele, among others. Using a deep recurrent neural network (RNN), the AI was able to learn which strokes were typical of each artist, and then used that information to make educated guesses.