World Models and Simulated Physics: From Digital Twins to Generative AI (Joel McKim)

• 16.12. / 2 - 4 pm / C40.530 / Smartness as Wealth Talk
• Lecture by Joel McKim (Birkbeck, University of London)
• There isanoptiontoparticipateonline.

This presentation considers the ‘world modelling’ and ‘world simulation’ claims of two intersecting technological domains: planetary scale digital twins and generative AI. Projects such as Nvidia’s Earth-2 digital twin platform seek to simulate global climate dynamics, while AI research is moving beyond the successes of large language models towards the development of “world models” capable of generating synthetic environments that resemble the physical world. OpenAI and Google DeepMind both claim that their large world models, such as Sora and Genie, demonstrate a grasp of “intuitive physics” and represent a path towards the production of effective world simulators. Drawing on recent discussions of embodied computation and the philosophy of simulation, from Eric Winsberg to Katherine Hayles, this presentation will question the epistemological and political implications of this technological turn towards “world modelling.” In particular, it examines the recursive risks that arise when simulations are used not only as tools for understanding but also as training data for AI systems deployed in real-world settings, and the gap between synthetic and physical worlds begins to collapse.

Joel McKim’s work focuses on the study of digital images and the impact of digital technologies on architecture, art and design. He is the Director of the Vasari Research Centre for Art and Technology and currently a Visiting Research Fellow at the V&A Research Institute where he is working on a project entitled “A Prehistory of Machine Vision: Exploring the V&A Computer Art Collection.”