Truth, a million frames a second: Towards a critique of World Models in AI

FBK Aula Piccola

Fondazione Bruno Kessler - Polo delle Scienze Umane e sociali

FBK Aula Piccola

Fondazione Bruno Kessler - Polo delle Scienze Umane e sociali

World Models are a rapidly growing area of contemporary AI: systems that generate video and interactive environments as synthetic training worlds for robots. This talk opens up this new technical and media form to historical critique. My central claim is that the transition from image generation to video generation is not merely a technical extension, but an epistemological shift. Image generators have often been publicly framed as toys, tools for visual production, or engines of “AI slop”. Video models such as Sora, Genie and Cosmos are often critiqued as moving “poor images”, but are increasingly presented by AI companies as simulations of the world. The addition of one axis, time, transforms the claim: the system is no longer said simply to know what things look like, but how they behave. Beginning with Harun Farocki’s concept of operational images, the talk traces World Models through longer histories of military vision, simulation and machine vision. Because these systems are now being developed as training environments for robots, their politics extends beyond virtual representation into behaviour in the world: the labour of drivers, cleaners, factory workers and soldiers. The question is no longer only what machine vision sees, but what machine vision makes machines do.

 

LEONARDO IMPETT | Bibliotheca Hertziana

 


Seminar cycle: “AI and history

Organization:
Federico Mazzini (Università degli Studi di Padova)
Andrea Pojer (Università di Trento – FBK-ISIG)
Massimo Rospocher (FBK-ISIG)
Sandra Toffolo (FBK-ISIG)


ISIG is accredited or qualified to provide training for school staff in Trentino.

The talk will be held in English.

The presentation will be in-person in the FBK Aula Piccola while seats last and online.

Registration is mandatory by July 3, 2026 at noon.

 

 

***
Image: AdobeStock_1037404632

Speakers

  • Leonardo Impett - Speaker
    Bibliotheca Hertziana
    Leonardo Impett is Research Group Leader at the Bibliotheca Hertziana, Max Planck Institute for Art History, where he directs the Machine Visual Culture group, and Associate Professor in Digital Humanities at the University of Cambridge. His research sits at the intersection of art history, digital humanities, computer vision and critical AI studies. He works on computational visual culture, multimodal models, datasets, embeddings, and the aesthetic, epistemological and political implications of machine learning systems applied to images: in short the image theory of AI. His book on the historical epistemology of AI, Vector Media, with Fabian Offert, was published earlier this year.

Registration

Registration to this event is mandatory.

Register

Contacts

Organizers

The initiative was also realized thanks to the contribution of "Direzione generale Educazione, ricerca e istituti culturali" of the Ministry of Culture.

Privacy Notice

Pursuant to art. 13 of EU Regulation No. 2016/679 – General Data Protection Regulation and as detailed in the Privacy Policy for FBK event’s participants, we inform you that the event may be recorded and disclosed on the FBK institutional channels. In order not to be filmed or recorded, you can always disable the webcam and/or mute the microphone during virtual events or inform the FBK staff who organize the public event beforehand.