Part of Temporal Constructs Appendix, which contains details behind the research: dates, quotations, descriptions of instruments, and sources.
General definition
A systematic review by Saloni Singh, Koen Hindriks, Dirk Heylen and Kim Baraka (2025) covers 62 co-creative systems. It sorts them by phase of the creative process, creative task, proactive behaviour, user control, embodiment and model type.
Relation to the practice
It is the field-level account of what co-creative systems are and what has been built. The review is cited for what it leaves out. Its applications are visual arts, design and writing, and music is not among them. That absence is the gap the practice works in, and it is where Co-creativity is studied here.
Scope of the review
The review collects 62 papers on systems where artificial intelligence takes part in the creative process as a collaborator. It reads them along six dimensions: the phase of the creative process, the creative task, the system’s proactive behaviour, user control, system embodiment and the type of model. The review gives two findings the greatest weight.
Systems that leave the person high control produce greater satisfaction, trust and a sense of ownership over the result. Proactive systems deepen collaboration where they are adaptive and sensitive to context. From these findings the review draws 24 design considerations. Among the 24 are encouraging the person to externalise their thinking and raising the system’s social presence and transparency so it can be trusted.
The review also names what is missing from the field. It records little support for the early phases of a creative process, such as clarifying the problem. It records difficulty in getting people to adapt to the systems built for them.
Verification
The abstract and the first page were read on 7 November 2026 from the PDF in the library (39 pages, arXiv:2506.21333). The claim that music is absent from the list of applications rests on the abstract’s own wording. That wording names visual arts, design and writing. The authors give the count of 62 papers themselves. A full reading is owed before the review is cited for anything beyond the abstract.