New No-Code Tool Lets Designers Blend AI Feedback With Human Judgment

Designers rarely judge their own work on a single opinion. In real practice, they mix expert reviews, tests with actual users, and computer models that predict how people will respond. Each source has blind spots. An expert might miss what real users do; a simulator might be fast but wrong. The problem is that the machine learning methods for combining all these sources have stayed locked away behind code and technical know-how, out of reach for most designers.

A team of researchers tackled that gap with MUSE, a tool that needs no coding at all. Designers load in their different feedback sources, then use simple controls to decide how much weight each one gets. Because MUSE models each source separately, it can do things a single blended score cannot: it handles slow sources and fast ones at the same time, and it visibly surfaces disagreements between them. If the expert says one thing and the simulator says another, you see it.

The team tested MUSE in a study where participants worked on visualization designs — charts and graphics meant to communicate data clearly. They moved back and forth between their own instincts and the simulator's suggestions. Participants reported feeling more confident in their choices and more flexible in how they worked, rather than being pushed toward whatever the algorithm preferred.

The honest catch is that this is research, not a product. MUSE was tested with a small group in one specific design task, and the paper is published in an academic journal, not shipped as software. The bigger idea, though, is promising: AI that helps you weigh many opinions — including your own — rather than quietly replacing your judgment with its own.