Research from MIT Media Lab has shown that watching highly realistic AI videos undermines trust in authentic video content, even when the use of neural networks is clearly indicated in disclaimers.
What Happened
In an experiment involving 100 participants, it was established that previewing synthetic content causes increased doubt in the authenticity of subsequent real videos. This psychological effect, termed "reality distress," leads to a decrease in users' confidence in their own judgments regarding the authenticity of visual information.
Context
As the quality of generative video grows, the problem is shifting from a purely technical plane of deepfake detection to the realm of psychological perception. Even with the presence of labels or watermarks, the human brain begins to systematically doubt the authenticity of the entire visual landscape.
Why It Matters for the Industry
For developers of generative models and technology platforms, this necessitates a transition from simple deepfake detection methods to strategies for ensuring "perceptual safety." In the long term, the industry will require the creation of "provenance-based ecosystems," where verification mechanisms are built into data transmission protocols, such as C2PA.
Why It Matters for Users
Viewers may face a long-term effect of declining trust in any visual information. Even if a video is labeled as AI-generated, users may develop an obsessive doubt regarding the authenticity of all other content they see in media and social networks.
What Remains Unknown / Limitations
The spectrum of impact ranges from a purely psychological effect to the necessity of creating new verification infrastructure (provenance) and changing UX patterns.
Sources
Author
Look at AI, Editorial Team