When I browse social media platforms like TikTok, I sometimes come across AI-generated videos that imitate politicians. These videos are so well-made that it’s difficult to tell what is real and what is fake,especially when watching on a small mobile screen.
I then came across this interesting paper by Groh et al.( 2024) on how humans detect political speech deepfake across transcripts, audio and video.
According to Groh et al. (2024), participants had the lowest accuracy when evaluating deepfakes from text transcripts, performed somewhat better with audio alone, and achieved the highest accuracy when both audio and video were available.
The study does not indicate whether the videos were shown on mobile devices or desktop computers.
The questions I have for peers is:
- Do you think the results would have been different if the deepfake videos were shown on mobile devices instead of desktops?
- What resources do you think would be most valuable for students in combating deepfakes?
Disclaimer: I have used ChatGPT 5 to fix grammar and improve accuracy of the sentences.
Reference:
Groh, M., Sankaranarayanan, A., Singh, N., Kim, D. Y., Lippman, A., & Picard, R. (2024). Human detection of political speech deepfakes across transcripts, audio, and video. Nature Communications, 15(1), 7629. https://doi.org/10.1038/s41467-024-51998-z
