AI, photography and the new image. Ten key positions in AI image theory – from infrastructure and datasets to authorship and power.
Holly Herndon & Mat Dryhurst – AI as cultural infrastructure
Herndon and Dryhurst treat AI not as a mere tool but as cultural infrastructure. Training data is part of a shared cultural heritage and should be governed transparently; consent, compensation and governance belong at the centre of model training, and artists should help write the rules. Their initiative Spawning and its tool Have I Been Trained? put these ideas into practice.
Hito Steyerl – Images inside data and power
Steyerl describes images as parts of global structures of data and power. Digital images are used by people and machines alike, and their meaning also depends on technical and political infrastructures. AI changes what an image does. In her work, making images and critiquing the conditions they come from go hand in hand.
Key texts: In Defense of the Poor Image (2009), Duty Free Art (2017), Mean Images (2023).
Trevor Paglen – How machines see us
Paglen investigates machine vision and the hidden infrastructures of AI. Training datasets carry social prejudice; ImageNet Roulette (2019, with Kate Crawford) made this visible by letting people see how a model labelled their own faces. His work joins photography and AI critique around one question: how do machines classify people?
Key text: Invisible Images (Your Pictures Are Looking at You) (2016).
Kate Crawford – AI as a socio-technical system
Crawford understands AI as a socio-technical system: datasets, data centres and human labour belong to it as much as algorithms. Atlas of AI (2021) traces the material and political foundations of artificial intelligence. AI is never neutral; power and resources shape how it develops.
James Bridle – More data, less understanding
In New Dark Age (2018) Bridle argues that more data does not automatically mean more knowledge. Complex systems create new forms of opacity, algorithms shape decisions, and technology must be placed in its social context. Critique and reflection remain necessary.
Joanna Zylinska – Creativity as a network
Zylinska examines AI art philosophically, questioning authorship, creativity and originality. Art emerges in networks of people and machines; AI changes artistic production without replacing it. Ethics and aesthetics belong together.
Key books: AI Art (2020), The Perception Machine (2023).
Boris Groys – Images in circulation
Groys analyses digital images as parts of global information flows. Meaning arises from circulation and context; the relation between original and copy shifts in digital space, and art becomes increasingly communicative. His theory explains where AI images stand within the network.
Lev Manovich – Software shapes culture
Manovich connects computer science and art history. Software shapes contemporary culture, and generative AI changes creative processes: person and model work together. Read through his work, design and selection remain human contributions.
Key books: The Language of New Media (2001), AI Aesthetics (2018).
Kate Crawford & Trevor Paglen – Datasets are political
Together they analyse training datasets as political infrastructure. Excavating AI: The Politics of Images in Machine Learning Training Sets (2019) shows that classifications carry social values; datasets are never neutral. Artistic research makes these structures visible, and transparency becomes a precondition for responsible AI.
Timnit Gebru – Accountability for data
Gebru calls for transparency about datasets. Datasheets for Datasets (2018, with co-authors) proposes documenting where data comes from and what biases it may contain. Accountability should be part of building AI; fair datasets are a precondition for trustworthy systems. Her work has shaped the debate on AI ethics.
Why this matters for FACE & FICTION
Every image in FACE & FICTION carries a provenance record: model, seed, parameters, LoRA, prompt and hash. That practice answers the demands running through these positions – Gebru’s datasheets, Crawford and Paglen’s excavation of datasets, Herndon and Dryhurst’s call for consent. The training data here is my own face, given with consent and documented in the open.
More on Groys, Manovich, Fontcuberta and Zylinska: Theoretical context →