ASKING THE WRONG QUESTIONS ABOUT GENERATIVE AI: EMERGENT ETHICS & AESTHETICS IN MACHINE COLLABORATION
Generative AI tools like ChatGPT or Midjourney have hundreds of thousands of unwitting co-authors——whose content was scraped from Reddit, GitHub, or the other public sites to create their proprietary models.
This talk investigates how creators interested in using generative tools should consider these collaborators, plus how machine learning systems shape whose voices are heard and silenced. Ciston presents alternative techniques that can make space for new aesthetics and ethics to emerge through community-centred approaches to machine learning. They argue that we should move away from sucking up more data carelessly and building larger, generalised, centralised models, and instead move toward approaches that include more expansive, diverse and conscientious imaginaries: namely, conscientious dataset stewardship, small dataset curation, data sovereignty, and reimagining machine learning models from scratch.
Ciston’s artistic practice works to reveal automated systems’ inner workings, their errors and inconsistencies, while pushing them to their hyperbolic limits. They are currently building alternative datasets of queer texts collected through intersectional tactics, such as community input and contributor consent, to be released with transparency about their contents, licensing, and methods——making them distinct from the large language models and datasets that support generative AI.
In this talk, Ciston explores the ethical and aesthetic implications of large-scale collaboration with machinic and human coauthors, and discusses the critical impact of automated systems on language use and creative collaboration. The talk helps us to better understand the mechanisms by which generative AI systems operate at massive scale, and teases out the stakes of their increasing impact on text, image, and their creative use.
Artist Talk, by Sarah Ciston
About the artist
Sarah Ciston is a poet-programmer who loves building tools to bring Intersectional approaches to machine learning and building community through accessible creative-critical coding. They are an AI Anarchies Fellow with the Akademie der Künste, a Mellon PhD Fellow in Media Arts and Practice at the University of Southern California, and an Associated Researcher at the Humboldt Institute for Internet and Society, plus author of "A Critical Field Guide to Working with Machine Learning Datasets" from the Knowing Machines research project. Their work has been featured in Leonardo Electronic Almanac; Ada Journal of Gender, New Media, and Technology; and CITAR: Journal of Science and Technology of the Arts. They also lead Creative Code Collective, a student community for co-learning programming using approachable, interdisciplinary strategies. Ciston’s projects include the Intersectional AI Toolkit, as well as an interactive natural language processing (NLP) database to ‘rewrite’ the inner critic and a bot that tries to explain feminism to online misogynists. They are currently developing a ‘queer love corpus’ that experiments with alternative methods of conscientious data stewardship in order to counter large language models like ChatGPT.