Impossible evolutions: textillic thinking with machine learning models

Abstract This paper discusses the creative project ‘Impossible Evolutions’, which uses generative machine learning models in the design of woven tapestries. This project is used as a conduit to unfold highly relational ways of thinking about the entanglements of human and machine assemblages within generative artificial intelligence. The project leverages interconnected ecological stories and the language of textiles to provide novel perspectives on the emerging relations between human and machine intelligences. The project uses Generative Adversarial Networks (GANs) and diffusion models to imagine new iterations of endangered Australian butterflies and wildflowers. The generated images are composed into three textile weavings of place: tapestries of the interconnected lives that generate each creature’s ecosystem. By reflecting on the interweaving of conditions that has disrupted each ecological niche, space is opened to think about unseen sensory worlds (Richmond Birdwing butterfly), symbiotic exchange (Bulloak Jewel butterfly), and stewardship of the land (Sunshine Diuris orchid). Each story becomes a fabric both literal and metaphorical, with this ‘textillic thinking’ offering speculative vantage points for approaching artistic and social practice with ML models. Textillic thinking interweaves creativity, collaboration, and care: conditions which are foregrounded in textile-making practices and disrupted in each creature’s ecological story. The creative work is diffracted with the neomaterialism of Rosi Braidotti and Karen Barad, and with the notions of care offered by both Maria Puig de la Bellacasa and the Indigenous Protocol for Artificial Intelligence. Through this reading and practice, the project offers material language for discussing the processes and effects of ML, while emphasizing the responsibility of the human in their design.