Reproducibility
Exhibition
curated by Spiros Hadjidjanos
18 - 27 October 2024, Wintertuin, Academy
Opening Thursday 17 October, 18:00
With work by Minne Atairu, Emmanuel Van der Auwera, Alice Channer, Elisa Giardina Papa, Wade Guyton, Spiros Hadjidjanos, Mashinka Firunts Hakopian, Holly Herndon & Mat Dryhurst, Benjamin Lallier, Sarah Meyohas, Mimi Ọnụọha, Seth Price, Michael Reisch, Thomas Ruff, Philippe Starck, Jenna Sutela, Kelley Walker, Nushin Yazdani
We are in this AI moment. The exhibition Reproducibility explores the transition from digital tools to artificial intelligence models in technologically mediated artistic production. Through a multisensory assemblage, the exhibition groups together seminal positions from the past two decades, focusing on the present moment.
Over two decades ago, American artist Kelley Walker disseminated his work(s) as digital files in CD-ROMs, printing on their envelope: ‘The disc and the image it contains can be reproduced and disseminated as often as the holder desires. (…) All forms of reproduction/deviation derived from this image … perpetuate a continuum correlating to the art-work.’ In 2021, artists Holly Herndon and Mat Dryhurst, after training and interacting with their own custom model, debuted CLASSIFIED, a “handcrafted” self-portraiture series exploring how the classification ‘Holly Herndon’ is embedded within OpenAI's CLIP neural network. Separated by two decades and different technologies, these photographic works share striking similarities. Both are distributed in JPEG or PNG formats and both focus on endless technological production and infinite variation. Can we identify a dichotomy today between a static and a dynamic thinking in technologically mediated artistic production, analogous to the contrast between handcrafted works and automated fabrication?
In his book The Eye of the Master – A Social History of Artificial Intelligence, Matteo Pasquinelli argues that classical algorithms differ significantly from AI systems. In classical algorithms, input data are passive and processed by fixed rules. In contrast, machine learning systems dynamically adjust their internal rules (parameters) based on input data, making data active rather than passive. Can generative AI models genuinely innovate and lead to breakthroughs? Are AI's current capabilities enough to replace human, forward-thinking creativity?
Questions pertinent to generative technology are explored historically with works that draw from archival material to form alternative possibilities such as Thomas Ruff’s Andere Porträts (‘Other Portraits’, 1994–95), a series of superimposed photographs, using a technique sometimes employed by police to produce imagery of likely suspects. Similarly, Nushin Yazdani, pointing to predictive analytics, uses AI technology to generate future prisoners from an archival dataset of inmates (Neither fate nor coincidence: Visualizing structural discrimination and machine bias, 2019); Sarah Meyohas trained a generative adversarial network (GAN) on a dataset of 100,000 physical rose petals to generate endless, new and unique petals (Infinite Petals, 2017); and Jenna Sutela’s nimiia cétiï, 2018 speculates on the potential functionalities of multimodal AI models.
The exhibition is titled after the Reproducibility section of the machine learning library PyTorch. Despite identical initial conditions, generative AI outputs are by default non-reproducible with varying outcomes due to random seed variation, GPU-induced randomness, hyperparameters and dataset splits. The conceptual space of current AI diffusion models appears vast when viewed through Flusser's camera program. There is only one kind of image missing from their terminal generative (AI) program: noise. This is because these models are built to remove noise, not generate it. Unlike the photographic apparatus of a camera, neural networks process data in ways the human operator cannot (yet) comprehend or predict. They are a kind of contingent machine within the framework of Deleuze’s philosophy of Difference; rhizomatic complexes of potential, non-activated possibilities that emerge from a multitude of contingencies and materialisations.
The AI-generated representations in the exhibition are anticipated to lead to breakthroughs in AI-driven material fabrication, much like the breakthroughs achieved with digital technology. This potential can be conceptualised by examining the pleated surfaces of Alice Channer, Wade Guyton’s inkjet paintings, Benjamin Lallier’s sun-bleached velvet works and Seth Price’s digitally ‘handcrafted’ materialisations. Do the strategies of these artists have the potential to be updated via AI models and the variation in massive datasets? Although these artists utilise digital technology in an input-output process, their approach follows a nondeterministic production logic that is similar to that of the AI works in the exhibition, which are based on statistical probabilities and unpredictable outcomes.
(image: Jenna Sutela: nimiia cétiï, 2018)
→ This exhibition is part of the research festival ARTICULATE 2024 I ANONYMOUS CREATIVITY - ART WITHOUT ARTISTS