Seeing Is Not Believing — Summary

A CHI 2026 study (N=100) from MIT Media Lab's Opera of the Future group finds that exposure to AI-generated videos erodes viewers' confidence in authentic media — even when the synthetic content is clearly labeled as AI-generated.

The two-phase experiment: an AI-exposure group first watched disclosed AI videos, then both groups watched identical human-generated content. Despite knowing the AI videos were fake, the exposed group reported increased doubt about authenticity, reduced confidence in their own judgments, greater perceptual disruption, and lower social connectedness.

The paper's central contribution is "perceptual safety" — a design principle arguing that labeling synthetic content addresses deception but not the deeper harm of losing the ability to trust what you see. The authors call for platform design strategies that preserve viewers' perceptual trust, not just their ability to spot fakes.