For this talk, researcher Leonardo Impett will argue that AI video models (like OpenAI’s SORA) should not be understood primarily as tools for cinema, but as infrastructures for the automation of physical labour. Drawing on Harun Farocki (who argued that since the Lumiere brothers’ Workers Leaving the Factory, cinema has always ignored the factory), Impett reframes generative video as a technology for simulating work rather than producing spectacle. 

While public debate remains fixated on deepfakes, copyright, and Hollywood, companies such as OpenAI, NVIDIA, and Amazon are building world simulators to train robots for warehouses, logistics, domestic labour – attempting to automate away low-paid (often migrant) workers in rich countries. Partly this is through what cinema has always done – temporal compression – allowing machines to learn in video-generated simulations at speeds impossible in real environments. 

Leonardo Impett is PI of the “Machine Visual Culture” research group at the Bibliotheca Hertziana in Rome, and assistant professor of digital humanities at the University of Cambridge. He has a background in information engineering and computer vision. His 2026 book Vector Media (with Fabian Offert) is published by Meson and the University of Minnesota Press. 

Organised in collaboration with the Centre for the Study of the Networked Image / Digital x Data Research Centre at LSBU.