How should designers structure human-GenAI interaction systems?

Reviewing a corpus of 291 papers, the authors of this HCI survey "synthesize the research fields from both the user and GenAI perspectives... into the following dimensions of the design space: 1) Purposes of Using GenAI, 2) Feedback from Models to Users, 3) Control from Users to Models, 4) Levels of Engagement, 5) Application Domains, and 6) Evaluation Strategies." The paper states this is meant to consolidate a literature where "key design considerations, common practices, and future research opportunities are still hidden and scattered across many broad topics," giving researchers a single framework to situate new systems within.

The taxonomy follows a User-Centered Design logic the authors spell out explicitly: a system is designed by "(1) orienting the system through the needs of the end-users... (2) considering the capabilities and functions of the GenAI models... (3) exploring the possible actions of the users, the corresponding results of the actions, and the mapping between the actions and the results... (4) enabling different levels of engagement in the interaction for different tasks... (5) deciding the application domain the system can be generalized into... (6) and evaluating the system via designated metrics." The taxonomy itself emerged from a multi-stage collaborative coding process: each author first tagged a subset of 25 papers, the group iteratively merged and split categories, and three authors then coded the full corpus, opening new discussion whenever a paper resisted the existing categories.

This sits one level above narrower taxonomies in the vault that cover a single slice of the same space. How should generative UI be organized as a design space? proposes audience, constancy, control, and timeframe as axes specifically for generative UI — a framework that maps most directly onto this survey's "Control from Users to Models" and "Levels of Engagement" dimensions, applied to one class of GenAI output. Similarly, Do generated analysis UIs really work better than chat? proposes a three-axis trade-off framework (malleability, specification, interoperability) for one application domain this survey's sixth dimension would file under "Application Domains." Read together, this survey functions as the umbrella structure those narrower, system-specific taxonomies instantiate.

The six dimensions were derived by qualitative consensus among three authors reading and tagging papers, not from a controlled study or a formal ontology-validation method, and workshop proposals, surveys, and literature reviews were deliberately excluded from the final tally and used only as supplementary guidance. The excerpt does not establish which dimension matters most for any particular design decision, nor does it validate that the six dimensions are jointly exhaustive or mutually exclusive — only that they recurred often enough across 291 papers for three readers to agree on them as a shared vocabulary.