Hasso-Plattner-InstitutSDG am HPI
Hasso-Plattner-InstitutDSG am HPI



Searching Through Seeing: Optimizing Computer Vision Technology for the Arts

What does it mean to see in the age of Artificial Intelligence? Can AI Be Creative? How is deep learning affecting our understanding of art? These are just a few of the thought-provoking questions that were addressed during a high-level symposium organized by The Frick Collection and the Frick Art Reference Library in New York City on April 12 and 13, 2018: Searching Through Seeing: Optimizing Computer Vision Technology for the Arts. The Hasso Plattner Institute for Digital Engineering and ARTORY LLC sponsored the event.

An impressive lineup of speakers as well as other experts in the fields of art history, technology, curatorial practice, business, and education attended the two-day symposium, which focused on how computer vision technologies are currently being used in the arts as well as their potential for the future. Among the institutions represented were Stanford University, Google, the Hasso Plattner Institute for Digital Engineering, Sotheby’s, IBM, Facebook, ARTORY, Universitӓt Heidelberg, the University of California, Berkeley, the Wildenstein Plattner Institute, Cornell University, Rutgers University, and the Frick Art Reference Library.

One of the primary goals of the symposium was to harness existing tools in computer vision science for art historical research and discuss how they can be developed further. In recent years, computer vision technology has been adopted in a variety of areas, including entertainment, sports, wearable devices, as well as autonomous and semi-autonomous vehicles, to name a few. As many of the speakers mentioned, the technology has not reached its full potential for curators, dealers, artists, and academics who work with images on a daily basis.

After opening remarks by Stephen Bury and Louisa Wood Ruby from the Frick Art Reference Library, the Library’s first fellow, Emily Spratt gave the keynote address, “Searching Through Seeing: Optimizing Computer Vision Technology for the Arts.” Spratt spoke about how machine learning techniques are becoming indispensable tools to find, sort, and analyze visual information. These and other technologies “are silently guiding the visual culture of our digital worlds.” Spratt believes that it is now time to recognize that the visual landscape of the digital world requires responsible curation and “the field of art history is uniquely positioned to advocate for this and could lead by example through its treatment of digital reproductions of art.” Pratt also set the stage for the conference by raising awareness of the philosophical debates on aesthetic perception and asking what it means to visually behold an image in a technocratic society that is extending our conception of what it means to see. 

The presentation that followed, “Rigorous Technical Image Analysis of Fine Art: Toward a Computer Connoissuership” was given by David Stork, a fellow at Rambus Labs. Stork has been involved in applying rigorous computer image analysis to problems in the history and interpretation of fine art for almost twenty years. Stork presented his vision of new art historical methodology, which involves a merging of traditional methods, such as close readings, textual analysis and comparisons of works, with new computer vision analysis methods. While recognizing that computer vision methods do not replace expert connoisseur judgments, he acknowledged that they can enhance as well as extend them.

Douglas Eck, a research scientist at Google who leads the Magenta Project, looked at how AI and Machine Learning fit into the creative process. The Magenta Project is a Google Brain effort to generate music, video, images, and text using deep learning and reinforcement learning. After describing the goals of Magenta, he looked at the relationship between machine learning and creativity. He argued that the issue is about augmenting and extending the artistic process rather than simply creating artifacts (e.g. paintings, songs) with machines. 

In his presentation, “Machine Learning: The Reality Behind Artificial Intelligence,” Prof. Christoph Meinel, the CEO and Scientific Director of the Hasso Plattner Institute for Digital Engineering (HPI) examined AI as a paradigm through which the digital transformation of our social and private lives is taking place. He also asked the question: What should we expect in a future world where the continuing advancement of digital technologies produces constant change? The presentation covered recent research and the development of digital technologies at HPI. Meinel placed particular emphasis on deep learning and demonstrated that it is relevant for the arts because it allows machines to perform radically new operations with images and texts. Not only was the presentation very well received by the audience, the vibrant slides also garnered extraordinary praise.  

Ahmed Elgammal, director of the Art and Artificial Intelligence Lab at Rutgers University introduced AI tools that have been developed for art recognition and visual searches. His presentation focused on the first public API designed and optimized for art, (ArtPI), which uses AI and deep learning models that have been trained using over a million images. ArtPI is able to recognize and predict the artist, style, and subject while simultaneously assessing formal elements. ArtPI also allows users to search large art collections that are available in the public domain.

In his presentation “Blockchain: The Holy Grail? Use of Immutable Records for Provenance and Art Historical Data,” Nanne Dekking, the CEO and Founder of ARTORY explained how blockchain technology can benefit provenance research and the authentication of artworks by providing art buyers with a tool that gives them more confidence in the purchases they make. Including all of the important, correct, and immutably connected information in the blockchain provides scholars with a greater sense of security in documenting the history of an artwork. The group that will benefit most from this technology, according to Dekking, is the living artists. “Through vetted galleries and estates, living artists can claim authorship of their works, reducing the incidence of forgery; a record of creation recorded as a block in the chain establishes the provenance of individual works from the moment they leave the studio; and sales are easily tracked, simplifying the payment of royalties under the terms of the Artist’s Resale Right.” Yet this benefit can also be a threat. Dekking admits that the blockchain is only as good as the quality of data that is entered. For this reason, it is essential to vet those who add to a blockchain, making sure that data integrity is a priority, along with security and transparency. 

These were only a few of the forward-looking presentations throughout the symposium, which also included engaging roundtable discussions and breakout groups that addressed future developments in the field. Thanks to the excellent organization, the wonderful location, and the generous sponsorship, the event was a resounding success.