Rotate your phone to navigate through the site properly
We are not only participating in the Visual Science of Art Conference 2022 in Amsterdam next 24-27 August, but we are also organizing a symposium entitled: “Unexpected realities: how uncertainty and imperfection influence perception and interpretation in digital visual studies“
• What is the right task? Prof. Nanne van Noord, University of Amsterdam, Netherlands
• Imperfect tools – When uncertainties of automated recognition reveal pictorial peculiarities. Valentine Bernasconi, University of Zurich, Switzerland
• Poetic Contingencies – Uncertainty and Imperfection in AI Art. Prof. Dejan Grba, University of the Arts, Belgrade, Serbia
• Happy Accidents of the Artificial Unconscious. Robin Champenois, SACRe – ENS – PSL University, Paris, France
• The doors of multimodal perspectives: Deep learning and the kaleidoscopic embeddings of culture. Dr. Eva Cetinić, Dr. Darío Negueruela del Castillo, University of Zurich, Switzerland
Computational methods are becoming increasingly integrated in research and creative practices in arts and visual studies. Computer vision and deep learning methods are being applied for various tasks such as detecting objects in paintings, quantifying stylistic concepts, recognizing gestures, classifying iconographic themes, or generating novel artistic content from visual or multimodal latent spaces.
Showcasing results of those tasks usually focuses on advancements and successful examples, while keeping unsuccessful attempts, errors, and misfits out of the picture. However, uncertainty and imperfection have played and keep playing a very important role in the visual arts, being a constitutive part in the process of creation, interpretation, and attribution, as well as contributing to the emergence of new artistic and scholarly iterations and productions.
The last decade saw the development of more robust deep learning models and computational tools, which is fostered by the creation of very large training datasets and numerous competitions for the completion of specific tasks. In this competitive context, most research is driven by the goal of improving specific metrics and limited by conventional methodological workflows imposed by the nature of specific tasks. When these methods and models are applied within a domain that cultivates the notion of dynamic and open interpretation, rather than a problem–solving viewpoint, then uncertainties and imperfections can play a significant role in assessing and adapting novel interdisciplinary research and creative practices.
The integration of artificial intelligence in artistic practice allows a critical reflection on the new medium, expanding our considerations of the aesthetical role of imperfections. The creative process casts light on misfunctions and broader social impacts of these technologies from which the scientific community can benefit and improve their work. The symposium aims to bring together researchers and practitioners in the field of computational art history, digital visual studies and generative art to discuss the creative potential of computational imperfections and unexpected results. Instead of focusing on the increased accuracy or successful examples of retrieval, classification, or style transfer, we aim to consider the fruitful aspects of errors – how they open new research directions and guide methodological decisions, how they contribute to the explainability of complex computational systems, how they influence the emergence of new poetics and expressive features, how they foster serendipity and discovery.
Through a series of five talks, the panel thus aims to open the discussion on the impact of upstream choices on the results of artificial intelligence, from the definition of datasets to the selection of tasks to which they are subject; how the vision and perception of the user exploiting the computational methods may influence the degree of imperfections; how artists may reveal the creative potential of these machines and offer a less strict apprehension of the artificially created content; and how these
acknowledgments and the interdisciplinarity of the field of digital visual studies can foster new
methodological perspectives, allowing the branch to evolve within its own framing.