A three-day conference asks how images train the real

Simone C. Niquille wants to show you what a machine has to be fed before it can recognize a chair. The Wonderful World of Chair, Saturday afternoon at FACT Liverpool, walks through how computer vision systems are trained to identify bodies, chairs and rooms: the training data, the simulated scenes, the furniture built for software to look at before any person sits in it. It is the closest thing on the program to a tour of a game engine, rooms authored for a viewer that is not human. Niquille follows it with a conversation with the conference curator.

How To Read An Image is a three-day conference at FACT, Oct. 2 to 4, curated by Nora O'Murchú of the University of Limerick and presented with the Institute of Art and Technology at Liverpool John Moores University. The announcement sets the question: how do you read an image that feels more real than reality? Most sessions run in The Box on FACT's ground floor, with live captions throughout.

McKenzie Wark, author of Gamer Theory, opens Friday night with a keynote on the "cis gaze." Maya Man performs StarQuest, her piece setting competition dance against generative AI's synthetic dancers, and Georgina Voss reads trust in images through her own retinal detachment in 20/Happy. On Sunday, Rosa Menkman maps the vocabulary of contemporary image types, Hadi Afif and the philosopher Vid Simoniti take on AI propaganda and images of Gaza and Lebanon, and Alex Quicho and Luiza Prado split Saturday's opening on visibility and the cloud.

Why go: Saturday alone is worth the trip. Niquille, Voss and Man in one afternoon cover the whole pipeline, from how a machine learns a room to how a body learns to perform for one. If you can make Friday too, the weekend pass costs less than buying the days separately.