Slashdot Mirror
Algorithm Automatically Spots 'Face Swaps' In Videos (technologyreview.com)
yagoda shares a report from MIT Technology Review: Andreas Rossler at the Technical University of Munich in Germany and colleagues have developed a deep-learning system that can automatically spot face-swap videos. The new technique could help identify forged videos as they are posted to the web. But the work also has sting in the tail. The same deep-learning technique that can spot face-swap videos can also be used to improve the quality of face swaps in the first place -- and that could make them harder to detect. The new technique relies on a deep-learning algorithm that Rossler and co have trained to spot face swaps. These algorithms can only learn from huge annotated data sets of good examples, which simply have not existed until now. In semi-related news, the Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) says it's "fighting back" against the dangers posed by new face-swapping technologies that have been used to digitally superimpose the faces of its members onto the bodies of porn stars.
"SAG-AFTRA has undertaken an exhaustive review of our collective bargaining options and legislative options to combat any and all uses of digital re-creations, not limited to deepfakes, that defame our members and inhibit their ability to protect their images, voices and performances from misappropriation. We are talking with our members' representatives, union allies, and with state and federal legislators about this issue right now and have legislation pending in New York and Louisiana that would address this directly in certain circumstances. We also are analyzing state laws in other jurisdictions, including California, to make sure protections are in place. To the degree that there are not sufficient protections in place, we will work to fix that..."
"SAG-AFTRA has undertaken an exhaustive review of our collective bargaining options and legislative options to combat any and all uses of digital re-creations, not limited to deepfakes, that defame our members and inhibit their ability to protect their images, voices and performances from misappropriation. We are talking with our members' representatives, union allies, and with state and federal legislators about this issue right now and have legislation pending in New York and Louisiana that would address this directly in certain circumstances. We also are analyzing state laws in other jurisdictions, including California, to make sure protections are in place. To the degree that there are not sufficient protections in place, we will work to fix that..."
Where is the "danger" in that? Porn stars usually have great bodies, so you'll end up looking great. And you don't have to worry about leaked sex tapes anymore because you always have plausible deniability.
I think the real reason the screen actor's guild is so up in arms about this is because it makes it much easier for movie producers to mix and match acting ability with looks: they can go for an unknown actor and paste exactly the kind of face on him they want. And licensing your face to be pasted on an unknown actor isn't as lucrative as acting yourself.
They correctly used the term algorithm instead of AI.
Yes! that's what the whole principle of adversarial learning is based on.
Some drink at the fountain of knowledge. Others just gargle.
Some grad student figured out a way to get a PhD for watching pornography all the time!
#DeleteChrome
Why are they trying to break one of the laws of the internet? Arrest those fucks.
If a fake video algorithm could fake a fake video algorithm, how much could a fake video algorithm fake?
My question is, who wins, the face swap detecting algorithm or the face swap detecting algorithm trying to create the perfect face swap?
The face swap algorithm wins, it is the only one with a winning state.
At some point it will reach a quality where the output quality is equivalent with what it would have been in a non-faked video, at that point the competition ends.
While your logic is on track it omits a crucial ground truth. It's like saying all crypto can be cracked. It's not the same as saying it can be cracked cheaply. If the detection is cheaper to implement than the synthesis then there will be a resource cap in the arms race that favors the detection.
Some drink at the fountain of knowledge. Others just gargle.
It's clear that the SAG is on this because there is a massive amount of money at stake. The real story here is that, should we evolve the technology to the point where no AI or person can detect it, there is a real hazard to liberty. Political enemies could be placed (virtually) in embarrassing or illegal positions, surveillance footage of crimes could be faked, criminals could go free by claiming their surveillance footage was faked. This really shakes the core of things we've learned as "truth." We've known photos were able to be faked, with various amounts of success/believability, for a long time, but the idea of producing videos that can't be detected as fakes is crazy.