Labeling AI-generated media online

Area of Science:

• Artificial Intelligence
• Media Studies
• Psychology

Background:

• Generative AI advancements raise concerns about audio-visual misinformation.
• Policymakers and tech companies are pushing for labels on AI-generated media.
• The effectiveness of different labeling strategies remains unclear.

Purpose of the Study:

• Evaluate process-based vs. harm-based labels for AI-generated content.
• Assess the impact of labels on viewers' beliefs and behavioral intentions.
• Determine the most effective labeling strategies for online AI media.

Main Methods:

• Two preregistered survey experiments with 7,579 American participants.
• Focus on misleading, AI-generated images.
• Comparison of different labeling strategies: process-based and harm-based.

Main Results:

• All tested labels significantly decreased participants' belief in presented claims.
• Process-based labels (identifying AI generation) had little impact on engagement.
• Harm-based labels showed greater potential in influencing beliefs and intentions.

Conclusions:

• Labeling AI-generated media can mitigate belief in misinformation.
• Process-based labels alone are insufficient for curbing engagement with misleading content.
• A combination of labeling strategies may be necessary for effective misinformation control.