Trevor Paglen's New Book Argues AI Has Fundamentally Changed What Images Do
The artist Trevor Paglen has published a new book arguing that computer vision and generative media have transformed visual culture more profoundly than the invention of perspective, photography, or mass consumer culture. The book, How to See Like a Machine: Images After AI, was published in 2026 by Verso Books. It collects six essays with an introduction, running to 192 pages with 53 illustrations in its hardcover edition, drawing on writings spanning the past decade. (The Art Newspaper; Collector Daily)
Paglen's central claim is that digital images have stopped being representational — that is, pictures of something — and have instead become operational interfaces, tools that serve corporate or state interests by intervening directly in people's daily lives. An image is no longer just something you look at. It is something that looks back, sorts you into a category, and acts on that classification. (The Art Newspaper)
To make this concrete, Paglen interrogates ImageNet, a widely used training dataset for machine-learning systems — the labelled collections of images that teach algorithms to recognise and classify the visual world. He finds its categorisation of people riddled with racist and gender biases, encoding discriminatory assumptions into the very infrastructure that increasingly governs automated decision-making. (The Art Newspaper)
The book also argues that the traditional theories and vocabulary for understanding visual culture are now obsolete and require radical updating. The frameworks developed to analyse painting, photography and cinema were built for images meant to be seen by human eyes. Paglen contends that the dominant images of our era are built to be read by machines, and that the gap between those two regimes is wider than most critics and scholars have reckoned with. (The Art Newspaper)
Paglen is known for a body of work that has long examined the invisible infrastructures of surveillance, from military black sites to undersea data cables. The new book extends that investigation into the territory of artificial intelligence and the training data that shapes what automated systems can "see." City Lights Bookstore in San Francisco has announced an event at which Paglen will discuss the book, though a date has not been specified. (City Lights Bookstore)
What makes the book timely is its insistence that the stakes are not aesthetic but structural. If images now function as instruments of classification and control rather than as representations, then the questions of who builds the datasets, what biases they carry, and who is sorted by them are not side issues. They are the issue.
