Anna Ridler — the British artist known for treating the dataset as her primary medium — continues to make some of the most thoughtful work in AI art precisely by refusing its usual shortcuts. Where most generative practice leans on models trained on billions of scraped images, Ridler builds her training data by hand: in her landmark Mosaic Virus, she photographed and individually labeled ten thousand tulips herself, then trained a model on them. The labor is not a means to an end; it is the work, making visible the human decisions, categorizations, and effort that every “automated” AI system quietly depends on and conceals.
Watch: Artificial AI: Datasets, Ethics and Deep Learning — Anna Ridler (YouTube)
The dataset as the artwork
Ridler’s insistence on hand-making her data reframes what AI art can be about. A dataset is never neutral — someone chose what to include, how to label it, what counts as which category — and those choices shape everything a model produces. By authoring her datasets openly and laboriously, she turns that hidden foundation into the visible subject: questions of classification, bias, value, and human labor move from footnotes to the center of the piece. It’s a pointed alternative to the frictionless-prompt paradigm, and part of the broader push this site has tracked of artists reasserting craft and intention against AI’s default of effortless output.
Why her approach matters now
As generative tools become ubiquitous and their training data more opaque and contested, Ridler’s method reads less like a quirk and more like a critical stance the whole field needs. She demonstrates that you can work deeply with machine learning while interrogating rather than obscuring how it works — that the interesting art question isn’t “what can the model generate” but “what did we teach it, and who decided.” Among this site’s digital-artist coverage, she stands distinct from the generative-code and data-sculpture artists: her material is the epistemology of the machine itself.