Facebook is stepping up its efforts to detect deepfakes spread through platform .
As the company revealed in a blog post, its collaborative work with researchers has made modest progress. Their efforts are aimed at developing a framework that will automatically detect machine-learning-generated media, known as deepfakes, that are intended to manipulate audiences. The project is called the Deepfake Detection Challenge and has brought together thousands of participants in a joint effort to mitigate the spread of deepfakes, but with limited success.
Specifically, the Facebook post reports that the top performance, for most of the 35,000 models that were tested, managed to detect deepfakes with an accuracy of 65.18 percent.
“The results of the DFDC project show that deepfakes continue to be an unsolvable problem”, the post says in part. “None of the 2,114 participants, which included leading experts from around the world, achieved an average accuracy of 70 percent on invisible deepfakes in the black box data set.”
Facebook researchers are also having difficulty with this problem.
“ researchers participated in the challenge (although they were not eligible for prizes due to our role in organizing the competition),” the company’s post adds. “The team’s final submission did not appear in the final leaderboard due to runtime issues when evaluated on this private test set.”
Facebook has banned the use of deepfakes on its platform since January 2020. However, some politicians continue to ignore this ban and in some cases still manage to create deepfakes in order to manipulate users.
The popular social media platform is clearly eager to stop the spread of deepfakes, but at the same time it is making no effort to mitigate the misinformation being spread through its platform, which of course makes absolutely no sense.
