Can a machine determine whether a painting is authentic? This question marked the beginning of one of the most compelling collaborations in my work: Together with the renowned American research team Steven and Andrea Frank (Art Eye-D Associates), we developed an AI system that distinguishes works by Peter Robert Keil from forgeries – with 98% accuracy.

The impetus was real: Keil’s oeuvre – over 20,000 paintings, popular and affordable – represents an attractive target for forgers. During more than 15 certification sessions, over 7,000 works were examined and approximately 600 forgeries identified. However, traditional authentication requires that each work be physically brought to Heidelberg. Our question: Can AI complement expert assessment – and enable remote authentication?

How the system functions

At its core is a Convolutional Neural Network (CNN) – “The A-Eye” – trained on 150 verified Keil works and 150 comparative works: from artists who influenced Keil (Miró, Picasso, Beckmann, Kirchner), through contemporaries (Appel, Baselitz, Penck, Warhol), to companions of the “Junge Wilde” (Lüpertz, Fetting, Salomé). Critical to the project was a rare asset: authentic forgeries identified by the artist himself from the collection, which we were able to contribute as training material. This enabled the network to learn not only broad distinctions, but the finest differentiations – even with forgeries that appear deceptively authentic to the human eye.

My role: Bridge between research and market

The concept for this project originated with me: I had followed the work of Steven and Andrea Frank and contacted the research team on my own initiative – resulting in a transatlantic collaboration. Subsequently, I coordinated the project: from compiling the verified body of works, to coordinating between the researchers, the Keil Collection Heidelberg, and art historian Dr. Kristina Hoge, to addressing how such a system would be implemented in practice. In parallel, I examined the subject academically: In my master’s thesis, I investigate the acceptance of AI-supported authentication in the art market – among collectors, dealers, and experts. Because the most advanced technology is of no value if the market does not trust it.

What this means for the art market

The project demonstrates exemplarily how I understand digitalization: not as an end in itself, but as a tool that solves a real market problem. The AI makes authentication scalable and location-independent – while simultaneously strengthening what every art market is built upon: trust.

Further information: Digitalization & AI · Keil Collection Project Report · Study: Art Eye-D Associates – Peter Robert Keil