Pandora's Box is open. The recent release of text-to-image AI models (DallE, MidJourney) is poised to have seismic impacts on the work and role of artists in society. In this talk, Scott gives a broad account, and critique, of the current state-of-the-art in AI, talking about the past, present, and future of these disruptive technologies from the point of view of an artist. In an accessible way, he explains the underpinnings of AI, how these systems learn and are trained, and the fascinating capabilities they enable for artists. He examines the wonders contained in the ‘visual cortex’ of AI models trained on internet-scale data but also introduces the idea of ‘craft’ AI – locally-trained machine learning models using highly curated datasets and workflows available to individual artists. In the talk he gives many examples from his recent work training and using these tools.
With a background in engineering, fine arts, and computer graphics, Scott was an early adopter in the field of creative AI. He was recently artist in residence at Meta AI Research and one of the first external beta testers of OpenAI’s DallE2.