Shauli Zacks
Published on: April 23, 2024
In a recent interview with SafetyDetectives, Dr. Carina Popovici, CEO and co-founder of Art Recognition, delves into the innovative intersection of art and artificial intelligence. Hailing from a rich background in theoretical physics and finance, Dr. Popovici shares her journey from academia and banking to pioneering AI in art verification. The story of Art Recognition began with a challenge: to tackle the rampant issues of authenticity in the art world—a sector notoriously reliant on the subjective judgments of human connoisseurs. Driven by her passion for art and equipped with a formidable technical acumen, she founded Art Recognition, aiming to transform the art authentication landscape with AI’s precision and impartiality.
Can you tell us a bit about your background and how you came to work with Art Recognition?
I grew up near Bucharest, Romania, and studied Physics at the university there. I completed my doctorate in Theoretical Particle Physics at the University of Tübingen, Germany. After positions as a research associate in Coimbra (Portugal) and Giessen (Germany), I moved to Zurich where I worked as a financial risk analyst at Credit Suisse. In 2019, I founded Art Recognition together with my colleague Christiane Hoppe-Oehl.
The inspiration for developing this program arose from discussions with an art historian back in 2018, who made me aware of the issue of authentication in the art market. As an art enthusiast, I was very eager to put my technical skills at the service of art.
At that time, there were no existing computer programs capable of addressing this challenge. Therefore, I began to write a program myself during my spare time. Eventually I decided to leave my position at a bank and establish a company. In the first year, we were fortunate to secure a research grant from the European Union. This grant provided us with the necessary funding to develop our program without the pressure of seeking investors or facing the typical challenges often encountered by early-stage start-ups.
How does Art Recognition’s AI system differentiate itself from other art verification methods traditionally used in the market?
Traditionally, the authentication of art has relied on the expertise and subjective judgment of human connoisseurs which are still the dominant authority in art authentication. However, this approach is subject to biases, individual interpretations, and the potential for human error. Also, relying solely on the expertise of one individual poses a significant problem as it concentrates an immense amount of power in the hands of a single person.
The introduction of AI technologies offers a promising solution to mitigate these concerns, since AI can provide objective and data-driven analyses in the authentication process. This shift from subjective opinions to evidence-based assessments can reduce the potential for human biases and enhance the overall reliability of authentication outcomes.
How does Art Recognition handle the nuances of different art styles and eras, particularly when dealing with less-documented artists?
Indeed, various art styles and periods present unique challenges. For instance, during the Renaissance and Baroque periods, many great masters had large studios, where apprentices were involved in the artwork creation process. In such cases, we use the available literature to find out what pieces were created by the master himself, and which ones were produced collaboratively with the workshops. Another challenging situation arises when multiple Catalogues Raisonnés exist for a single artist, like Modigliani. Here, our art historians must identify artworks recognized as authentic by all experts, which are then used for training the AI. Also for the less documented artists we rely on available literature to gather the necessary information. From a practical point of view, artists with scarce documentation are hardly subject to authentication queries, since in general they have a lower monetary value. Conversely, higher-valued artists are very well-documented.
In summary, we always ensure that our training datasets align with Catalogues Raisonnés and scholarly literature, with each training image thoroughly documented and verified.
In what ways do you think AI technology will continue to evolve in the art authentication field over the next decade?
I am confident that AI will not only continue to evolve but has already started to trigger a revolution in the art market. AI has the potential to drive profound and positive changes in this field, with Art Recognition emerging as a leader in this movement. Our AI authentication system is already trusted by hundreds of art collectors and institutions who appreciate the objectivity and transparency we provide. Major institutions like Kunsthaus Zurich have even started to take a closer look at paintings from their collections, based on our insights. The adoption of our technology is rapidly growing, and it is just a matter of time until AI will become mainstream.
With the rise of digital art and NFTs, how does Art Recognition position itself in this new market?
We can already see that generative AI models are transforming artistic creativity, opening new possibilities for artists to produce spectacular work with AI-assisted technologies. Yet, there are also significant concerns linked to AI, including its potential to mimic the styles of renowned artists and create digital forgeries. To counteract this, we included AI-generated forgeries as negative examples into our dataset, to train the AI to differentiate between authentic pieces and deep fakes. This strategy ensures we stay ahead of the curve in addressing the evolving challenges AI presents in the art industry.
Beyond authentication, are there other areas within the art world where you see potential applications for your AI technology?
A new project we are currently developing focuses on art attribution. Clients frequently ask us to identify the unknown creators of artwork they possess. This isn’t simply about verifying well-known artists, like confirming a Van Gogh; it is about finding the artist behind a treasured painting they might have inherited. To accommodate these requests, we need to refine our technology and mostly to expand our database of artists to ensure comprehensive coverage and accuracy.
