Multimodal AI such as DALL·E, Midjourney or Stable Diffusion are capable of generating very complex text-image meanings. The “learning” of these models consists in mapping an input – i.e. the structures and patterns of human culture, memories and concepts – with an output by drawing a function that approximately describes their tendency, and then applies that function to future inputs to predict their outputs. In the process, missing parts are “guessed” through interpolation (projection and prediction of an output that falls within the known) or extrapolation (projection and prediction of an output beyond the limits of the known). In this way, AI systems recompose our cultural heritage without plagiarising.
The generated images are “statistical renderings” and reproduce certain patterns of addressing the audience, but also poses and styles as patterns. In this project I am interested in:
‒ What historical image traditions can be identified with regard to these patterns?
‒ How exactly do they manifest themselves in the present, and especially in connection with AI-generated images?
‒ What makes these image patterns so attractive for contemporary acts of communication?
‒ How do artists refer to these pictorial traditions; which ones do they favour? How do they use them?