Live data from Hacker News

Here Come the Fake Videos, Too

nytimes.com

41–50 of 107 posts

Re: Here Come the Fake Videos, Too

#41
post #38

Anyone else rather unimpressed by the realism of these fake videos? I mean, technologically it’s cool and all, but it seems more like a proof of concept than anything that can be used to fool humans. All the videos I’ve watched — out of technical curiosity, naturally — have had some sort of glitch that made it obvious it was fake. I think this technology will have a serious problems just mapping the facial expression…

Some of them are pretty incredible - you can not tell at all they are fakes. I think it depends on how close the face matches the person being faked and how generic the lighting is. If you scroll down in the OP article you'll see some impressive examples.

Re: Here Come the Fake Videos, Too

#42
post #5

The era of verifiability-- when photographic and video evidence could plausibly be trusted-- is over. Photos have been dubious since photoshop, but still required expertise and artistry to use. Fake moving pictures required considerably more extensive expertise, lots of time, and usually specialized equipment. None of that is true anymore. We are back to a pre-industrial age of discourse, where rumor and hearsay and…

No, this is an opportunity for new technology that adds additional data stream(s) to audio/video to ensure authenticity. Perhaps something like a one way hash of physical properties related to time/position and the rasterized data, in a way that cannot be faked.. or maybe a move away from rasterized data. Idk. Just brainstorming. I’m confident there will be an innovation in this space since there is clearly a very ur…

The difficult part is the "in a way that cannot be faked". If you can record it using a device you control, you can generate a fake made to appear as if it had been recorded with that device. If you do not fully control the recording device (say because it uses a secure enclave to generate signatures), you'll need some way to generate realistic inputs for fake data, but that might be as easy as photographing a high-resolution print.

If the effort to document the truthfulness of data is much higher than for unverified recordings, the majority of the evidence anyone can come up with will be of the variety that is trivially to fake for someone determined.

Re: Here Come the Fake Videos, Too

#43
post #38

Anyone else rather unimpressed by the realism of these fake videos? I mean, technologically it’s cool and all, but it seems more like a proof of concept than anything that can be used to fool humans. All the videos I’ve watched — out of technical curiosity, naturally — have had some sort of glitch that made it obvious it was fake. I think this technology will have a serious problems just mapping the facial expression…

Eh - https://vimeo.com/257360045

Definitely not perfect, but this is hobbyist grade work. With the recent work around parallelized WaveNet synthesizing 10 seconds of audio for every second of wall clock time, a live fake that fools 50% of regular people is probably a couple of years away at most. Particularly if you can control the setting to ensure lighting/angles/etc match up reasonably well.

Re: Here Come the Fake Videos, Too

#44
post #29

Earlier quoted context omitted.

I don't think Jennifer Lawrence was particularly saddened about increased media attention around her that was timed precisely at the peak of her career.

Shameful assumption. I can't imagine she (or any other victim of the leak) enjoyed having their private lives and most intimate moments being accessible to the whole world.

I assure you none of them lost any sleep over this. All those carefully planned and shot takes were "leaked" completely accidentally at exactly the right time to ensure maximum media uproar and their personas appearing on the front pages of every tabloid in the western world. If you think they play by the same rules as you and me, you are naive.

Re: Here Come the Fake Videos, Too

#45
post #16

What about just having videos be watermarked and signed by their authors as well as the equipment used? Then you could trust their authenticity, no?

You could trust the authenticity of the fact that someone used the author's key to sign it at some point in time.

You could not trust any claims the author makes about the circumstances under which the signed object was produced. They could have put their signature on a deepfake. They could have put their signature on the work of someone else. The author could have lost their signing key. It could have been created much earlier than it was signed.

A signature tells you very little about the thing being signed besides the fact that it was signed.

Re: Here Come the Fake Videos, Too

#46
post #42

Earlier quoted context omitted.

No, this is an opportunity for new technology that adds additional data stream(s) to audio/video to ensure authenticity. Perhaps something like a one way hash of physical properties related to time/position and the rasterized data, in a way that cannot be faked.. or maybe a move away from rasterized data. Idk. Just brainstorming. I’m confident there will be an innovation in this space since there is clearly a very ur…

The difficult part is the "in a way that cannot be faked". If you can record it using a device you control, you can generate a fake made to appear as if it had been recorded with that device. If you do not fully control the recording device (say because it uses a secure enclave to generate signatures), you'll need some way to generate realistic inputs for fake data, but that might be as easy as photographing a high-r…

Tamper-proof HSMs in recording equipment might do the trick. The point is not that they can be broken. The level of sophistication has to be high enough that most bad actors will be incapable of doing that. This is why chip cards for security work.

Re: Here Come the Fake Videos, Too

#47
post #44

Earlier quoted context omitted.

Shameful assumption. I can't imagine she (or any other victim of the leak) enjoyed having their private lives and most intimate moments being accessible to the whole world.

I assure you none of them lost any sleep over this. All those carefully planned and shot takes were "leaked" completely accidentally at exactly the right time to ensure maximum media uproar and their personas appearing on the front pages of every tabloid in the western world. If you think they play by the same rules as you and me, you are naive.

You shouldn't "assure" stuff you have no idea about. Fact is JLaw was very upset about the leaks and reacted so badly to it that most of reddit style sites fell out of love with her because she didn't appreciate having her photos leaked or didn't just laugh it off and that was considered out of character for her.

Re: Here Come the Fake Videos, Too

#48

Earlier quoted context omitted.

No, this is an opportunity for new technology that adds additional data stream(s) to audio/video to ensure authenticity. Perhaps something like a one way hash of physical properties related to time/position and the rasterized data, in a way that cannot be faked.. or maybe a move away from rasterized data. Idk. Just brainstorming. I’m confident there will be an innovation in this space since there is clearly a very ur…

A short-lived opportunity for sure. This is the very definition of an arms race, amd I can't see "real" winning out over fake. Anything that can be checked can be preempted, so "real" is always one step behind, as far as the present is concerned.

Historically, arms races have been pretty long-lived opportunities for weapons developers.

Re: Here Come the Fake Videos, Too

#49
post #5

The era of verifiability-- when photographic and video evidence could plausibly be trusted-- is over. Photos have been dubious since photoshop, but still required expertise and artistry to use. Fake moving pictures required considerably more extensive expertise, lots of time, and usually specialized equipment. None of that is true anymore. We are back to a pre-industrial age of discourse, where rumor and hearsay and…

Nikon used to provide "Image Authentication Software" which would authenticate images and show if they had been altered since taken by the camera. This was useful not only in news but also court cases.

This, obviously, could not be used on tweaked published images but would work if you had access to the original frame. Most photojournalists work to a code whereby they don't amend the images at all, even tidying up, but that's only viable if they and their editors are trustworthy.

The Nikon process was cracked a few years ago. This allowed edited images to pass the validation.

Re: Here Come the Fake Videos, Too

#50
post #21
post #7

Earlier quoted context omitted.

Is this true though? Sure, for a casual observer, these new methods for generating videos appear convincing, but is that the right bar to judge "ability to fake evidence"? As far as I know, there have always been more sophisticated techniques and forensics to determine if an image is doctored, and likewise for video. I've not seen any research tackling fooling those methods yet, and I would bet that naive implementat…

Does that matter? On Internet the judgement is quick, long before the truth has revealed itself lives has been destroyed, jobs lost and reputation irreparably damaged.

In addition to this, despite all our cutting edge forensics, we still fumble the ball in the courts, based on sham forensics, and send innocent people to jail.
Post reply on HN