Article: How Artificial Intelligence Is Revolutionizing the World of Forgery Detection

Photo Source: Flickr

Recently artificial intelligence has made waves in the art world not for AI-derived art but for its ability to attribute paintings to master artists and to spot forgeries. Five years ago, the high stakes world of art masterpiece sales jumped to nearly a half billion dollars. In November 2017, Christie's Auction House sold a painting by Leonardo da Vinci titled Salvatore Mundi for $450.3 million to an unknown buyer in the largest artwork sale ever. Other masterworks have sold recently in the nine-figure range, such as William de Koonings Interchange for $300 million, Paul Cézanne's The Card Players for $250 million, and Andy Warhol's Orange Marylin for $225 million. (theartwolf.com) The prices for masterworks from the Renaissance to today continue to climb to stratospheric heights driven by billionaires fighting for an ever-shrinking pool of masterpieces not already in museums.

Such a thirst for great art greatly incentivizes art forgers to generate undetectable fakes. It also puts tremendous pressure on art historians and authenticators to discern true masterworks from forgeries or painters from the same period that imitated the masters. The process of art authentication began with art historians studying the style, content, and form of paintings to determine who may have painted an unattributed painting. Today art authentication includes art history, detective work, and analytical science. Art authenticators use sophisticated scientific instruments such as x-ray to see into a painting, chemical analysis to determine if the paint could have existed at the time of the painting's creation, and isotope dating analysis to determine the age of the paint and canvas of the picture. Many scientific analyses help date and place a painting but cannot tell who painted the painting. Enter artificial intelligence.

Researchers have used a type of artificial intelligence called computer vision to identify elements in a picture, such as a cat, dog, or human face. The process of computer vision uses deep learning. Deep learning allows a computer to analyze an image by looking at it in different size chunks, from tiny pieces of a picture to significant sections. For example, the program learns the difference between human and cat eyes and hundreds of other features. By examining hundreds to thousands of pictures of faces or pieces of art, deep learning programs become very accurate and sensitive to different elements of an image. Deep learning trained on artwork can determine very fine details in brushwork. Deep learning finds that each artist has a distinct style of brushwork, as distinctive as a fingerprint.

In shocking news for the art world recently, a deep learning AI trained on 148 uncontested paintings by the renaissance master, Peter Paul Reubens, found the painting Samson and Delilah (ca. 1609/10) to be a fake! (artnet.com) The National Gallery of London famously purchased the work in 1980 for over $5 million, a landmark purchase for the time. The deep learning AI concludes with a 91% certainty that the painting is not Reuben's.

AI can also help determine who painted a painting when unknown. Recently, the Wall Street Journal reported using a similar deep learning AI in attributing a painting found in an antique shop to one of the greatest renaissance painters of all time, Raphael. (wsj.com) The painting depicts Madonna and child and Elizabeth and the child John the Baptist. Since only 22 known Raphael paintings exist, discovering a 23rd Raphael would truly rock the art world. According to the Zurich-based art authentication group called Art Recognition, their deep learning has concluded with 97% certainty that Raphael painted the faces of Mary and Jesus in this painting.

The appetite for masterworks continues to grow among extremely wealthy art collectors as the supply of masterpieces not in museums shrinks. The sale of a painting by Leonardo da Vinci for 450 million dollars in 2017 demonstrates the hunger for art by historical masters. Unfortunately, the surging prices for art only encourage forgery and further challenge art historians and art authenticators. Over the past decades, art history has teamed up with analytical scientists to use chemistry and physics to help determine the time and place from which a painting comes. However, such analytics cannot tell who painted the work, only that it comes from the correct time period and sometimes place. In breakthrough fashion, artificial intelligence has altered the authentication market for art with deep learning. Deep learning trained on the known works of art masters can observe the strokes of a paintbrush at a level beyond human capability. Deep learning sees brush strokes as unique as fingerprints. Such deep learning has recently supported suspicions that the Reuben's at the National Gallery of London is fake. Moreover, deep learning developed at Art Recognition strongly suggests that a new Raphael may join the tiny collection of Raphael's left in the world. As a result, forgery may have just gotten exponentially more difficult.

Dr. Smith’s career in scientific and information research spans the areas of bioinformatics, artificial intelligence, toxicology, and chemistry. He has published a number of peer-reviewed scientific papers. He has worked over the past seventeen years developing advanced analytics, machine learning, and knowledge management tools to enable research and support high-level decision making. Tim completed his Ph.D. in Toxicology at Cornell University and a Bachelor of Science in chemistry from the University of Washington.

You can buy his book on Amazon in paperback and in kindle format here.