BlueSky Thinking Summary

As the U.S.

election season heats up, new concerns are growing about deepfakes and AI-generated media that are designed to mislead.

Matt Groh from Kellogg School and his team performed research that estimated how well people can tell such deepfakes.

They tested 2,215 participants on their ability to separate real from fake speeches by Presidents Biden and Trump, presented in various formats like transcripts, videos, and audios.

Results indicated that individuals are about 74% accurate when they have video and audio but that the accuracy goes down with less contextual information.

Groh's findings run somewhat against the conventional wisdom that deepfakes are everywhere and uncontrollable, pointing out that making good deepfakes is still a difficult and computationally intensive task.

The study emphasizes that developing critical-thinking skills and media literacy forms part of the battle against deepfake misinformation.

As deepfake technology evolves, it is clear that knowing what it cannot do and being better informed are keys to treading the digital information landscape.