AI & creative practice
A new survey of human–AI co-creativity shifts attention from spectacular outputs to agency, iteration and the conditions of authorship.
In September 2026, researchers published Human-AI Co-Creativity: Advances, Opportunities, and Challenges on arXiv. The survey grew from a workshop held at ICML 2026 and maps questions emerging around generative systems in creative work. It is a preprint and workshop-informed survey, not peer-reviewed experimental proof that any particular tool improves creativity.
That status suits the subject. Co-creativity is not one settled interaction. It can mean prompting, selecting, revising, resisting, combining, discarding or beginning again by hand. The important unit may be less the generated object than the chain of decisions around it.
The term also contains a useful disagreement. Calling a system a collaborator can encourage designers to build for dialogue and revision, but it can blur the asymmetry between a person with intentions and a model generating from learned patterns. The language we choose affects which responsibilities remain attached to developers, institutions and users. It also affects whether audiences can recognise the human labour of setting constraints, checking results and deciding what should never leave the working studio.





