Why don't those developing AI Deepfake detectors use two differently trained detectors at once that way if the Deepfake was trained to fool one of the detectors the other would catch it and vice-versa?

To be clear this is really a question of can deepfakes be made to fool multiple high-accuracy detectors at the same time. And if so then how many can they fool before they become human detectable from noticeable noise?

I've heard of papers where they injected a certain noise into their deepfake videos which allows them to fool a given detector (https://arxiv.org/abs/2009.09213, https://delaat.net/rp/2019-2020/p74/report.pdf), so I thought well if they simply used two high-accuracy detectors then any pattern of noise used to fool one detector would interfere with the pattern of noise used to fool the other detector.


1 Answer 1


Because it is possible to fool many different models at once. See table 2 in this paper, for an example using adversarial perturbations: https://arxiv.org/pdf/1610.08401.pdf

That being said, there is no reason to think that using two detectors at once will not increase chance to detect deepfakes. It will just not resolve the problem completely.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .