Why do GenAI interfaces favor sliders over voice and gesture?

Reviewing 291 papers on human-GenAI interaction, the authors report that "direct control modalities are more applied (e.g. widgets, controllers, drawings and highlighting, and text, N= 234) than the intuitive ones (e.g. gestures, brain signals, and voice, N= 49) to control the output of the GenAI system." This is Finding-1 of the survey's discussion section, titled "Mediums of Control: Direct but not Intuitive."

The authors explain the distinction by how each modality handles the mapping between a user's action and the model's output. "Direct control modalities allow users to modify the attributes of the models or data straightforwardly, while intuitive control modalities require mappings from the users' intuition to the functions of the models." Treating GenAI systems as black boxes with unknown input-output correspondence, they argue intuitive interactions should, in principle, "foster smoother learning curves of this I/O correspondence" because researchers have preset the mapping for users, whereas direct interactions leave users to construct that mapping themselves — for example, using sliders to adjust the weights of the attributes of a GAN model is direct yet not intuitive, because novice users... do not possess the technical knowledge to understand the correspondences between the attributes and the outputs. Despite that theoretical advantage, the literature still skews heavily toward the direct, non-intuitive end.

This finding sits inside the "Control from Users to Models" dimension of the taxonomy proposed in the sibling note, How should designers structure human-GenAI interaction systems?, giving that dimension an empirical base rate: most systems researchers have built default to direct manipulation rather than gesture, voice, or brain-signal input. How should generative UI be organized as a design space? also isolates control as a design axis, but at the level of a single UI-generation framework rather than a literature-wide tally; this finding suggests that even as generative UI work matures, the control surfaces it builds are likely to keep leaning on sliders, widgets, and text rather than intuitive input.

The 234-to-49 count describes what researchers built and published, not what end users found easier or preferred — the survey does not report a usability study comparing direct and intuitive modalities head to head, so "intuitive" here is the authors' categorical label for modality type, not a measured outcome. It also does not say why direct modalities dominate (easier to implement, easier to evaluate, or genuinely better for most tasks are all left open), so the finding should be read as evidence of a design practice gap — intuitive interaction is underexplored — rather than evidence that intuitive control is in fact superior.