May (A)I Beautify Your Visualization? Expert Judgments of Acceptable Aesthetic Alterations

The paper reports an expert survey of 95 visualization researchers, practitioners, and domain scientists on 15 kinds of aesthetic alterations to 3D scientific visualizations. The set spans presentation-level changes such as lighting and camera position, plus data-level edits like removing errors or filling gaps.

The main result is that people judge acceptability based less on the technical layer being changed and more on the meaning of the transformation. In other words, a “presentation” tweak can feel more invasive than a “data” tweak if it changes interpretation, and vice versa. The paper also finds a stable ordering of which alterations are seen as more or less permissible.

A notable wrinkle: AI-generated transformations are consistently rated as less acceptable than the same changes made by humans. That suggests authorship still matters, even when the visual outcome is identical, which is relevant for any tool that auto-enhances charts, scientific renders, or in-engine debug visualizations.

For game developers, the practical lesson is about trust and constraints. If you’re building tools that beautify captures, telemetry views, level previews, or simulation outputs, you probably need explicit provenance, undoable edits, and clear boundaries between cosmetic changes and data-altering ones. The paper is a reminder that “better looking” can quickly become “less trustworthy” if the tool doesn’t make the transformation legible.