AI and Art Fraud: The New Frontier of Forensics

A new form of forensic evidence may soon be introduced in courts—evidence generated using artificial intelligence. Steven and Andrea Frank, based in Massachusetts, have trained an AI convolutional neural network to recognize art fraud. They report that their method has a success rate of 90.4 percent.

Using AI to process images is incredibly difficult, because of the high volumes of information that must be processed. A Rembrandt painting, for example, could contain hundreds of megabytes. The Frank’s method simplifies the amount of information being used to train the AI’s neural network by breaking these images down into 100 by 100 sized pixel tiles (and 200 by 200, 400 by 400, etc.) and measuring the ‘entropy’ (read: visual diversity) within each image. 

The alternative to the Frank’s method is to feed the AI less complicated, lower resolution images. This process, however, loses information that the AI can analyze, making its result less significant and reliable.

The results of this new AI analysis could be highly applicable in court. Consider the example of art fraud. Prosecutors can call an expert witness, such as someone who has studied the art of a certain period or even a specific artist. Such an expert, if qualified under Daubert, may be able to testify as to how an artist usually painted, or their known techniques. Chemical tests could also be done to determine the composition of paints on a canvas, which could determine whether the paint was one used at the time the painting was supposedly created, or if it is of a more recent, Sherwin-Williams origin. 

However, none of these methods would be able to crack the ‘perfect’ fake: the contemporaneous imitator. For painters like Rembrandt, who were famous during their own lifetime, there exist hundreds of paintings by contemporaneous imitators. These painters were likely not seeking to commit art fraud, but today’s forensic science would have difficulty disproving that their work is not Rembrandt’s. This is where the Frank’s AI technique would be valuable. The AI neural network can identify, at an astonishingly high reported level, whether an artwork is real, or fake; in other words, an evidentiary slam-dunk.

The AI neural network can identify, at an astonishingly high reported level, whether an artwork is real, or fake; in other words, an evidentiary slam-dunk.

The Frank’s method is not limited to artwork. Their AI neural network could be trained to analyze medical images as well. This could be especially useful for diagnosing rare diseases, including rare types of cancer.

The question looming over this exciting break-through, is whether this type of evidence is admissible in court. Rule 702, widely known as the Daubert standard, requires that an expert’s knowledge will assist the trier of fact and that their testimony is based on sufficient facts or data, the product of reliable principles and methods, and that those methods have been reliably applied to the facts of the case. To determine if these factors are met, the Court reviews whether the method can be tested for falsification (that is, can it be proved wrong), whether it has been submitted for peer review and publication, whether there is a known rate of error, and whether it has been generally accepted within the field.   

It is critical that this evidence assists the trier of fact—that it provides information that could help the jury reach their determination. After this threshold has been met, the remaining analysis focuses on whether evidence is reliable. To determine reliability, the court uses the above mentioned Daubert factors. The difficulty with ground-breaking new techniques, such as the Frank’s method of analyzing images, is that it takes time (and funding) to meet these standards. The Frank’s method is currently under peer review at IEEE Transactions on Neural Networks and Learning Systems. If the journal publishes it, and their findings engender further research and testing that supports their theory, the Franks could be well on their way to a new career—as expert witnesses. 

Ninamarie von Nyssen

Ninamarie is a second-year law student at UNC. She attended UCLA for undergrad and majored in English Literature. At UNC, she is the symposium editor for the Environmental Law Project and a competing member of Broun National Trial Team.