AI can now animate the Mona Lisa's face or any other portrait you give it. We're not sure we're happy with this reality AI code can breathe life into portrait paintings, photos of dead celebrities, and your Facebook selfies, transforming single still images into moving and talking heads. In one demonstration of the software's creepy abilities, the Mona Lisa, famous for its ambiguous expression, is animated just like one of the moving paintings …
My worry isn't that videos will be faked to make someone say something that they didn't but the reverse. I think it will give plausible deniability to politicians and others to say that it wasn't them in the video. You could imagine a certain political president when faced with a video of him directing women paid for company to entertain him in a hotel room once used by another president from that same country.
"That's deep fake news folks, it wasn't me, just deep fakes"
That'll never take off.
Now, instead, if you bury a watermark into every frame of video, and sell it as a way to track people for surveillance and/or advertising, now that might get it spread widely.
(Or as an 'anti-piracy' measure).
Technologies usually get taken up because they make someone a lot of money.
"Now, instead, if you bury a watermark into every frame of video, and sell it as a way to track people for surveillance and/or advertising, now that might get it spread widely."
So.... How would you verify that the watermark wasn't faked to implicate the owner of the watermark in some scandal or other?
Who keeps track of the watermarks/owners/serial numbers database... Google and FB because they're so good at it? Do we get something rather like a keyserver but for cameras, or will they all simply have OpenPGP built in? BUT THEN, even with signed watermark based on EXIF and frame hashed using a secret factory-fused UUID for a salt and whatever else makes you feel good-- how would you verify that the bad guys didn't just steal your camera and make a hopefully moire-free 720p video of a 4k screen running a deepfake puppet show? Fingerprint scanners in the buttons? Yubikey or else no worky? Seems like it's Tuttles all the way down.
This is going to be one more of those things that curiously seems to work very differently depending on your net worth - footage incriminating hoi polloi will continue to be accepted as evidence with any objections of could-be-a-fake dismissed immediately, while "real people" with mansions will enjoy the benefit of doubt and "fake news" even if they're shown shooting someone point blank (assuming there's no body). Meanwhile, back in reality, no video is proof of anything anymore.
"Anything's a dildo if you are brave enough." - Abraham Lincoln
A great article showing twitter replies to the governor of Texas for his crappy twitter quote fact checking (and first place I heard the Abe quotes). Darn funny imo.
https://gizmodo.com/texas-governor-tweets-fake-churchill-quote-about-fascis-1828159181
This post has been deleted by its author
The best part is that you can find the right face for your movie in a crowd of people at the mall/train station/grocery store, and instead of teaching them to act, you just pay them for 2 minutes of talking on camera. No more dealing with actors and agents. $200M movies now only cost $2M because you only have to use the previous 'second unit' film crew to do background shots, all of the actors are on disk. The other side effect is that movies now only cost $0.50 again.
Doctored videos of Nancy Pelosi slowed down to make it sound like she's drunken slurring their words are making the rounds tonight. You don't need fancy AI tricks able to work off a single photo if you have actual video of people to mess with.
The downside of this trend is that even if you get a video of someone admitting to / committing a crime, they will be able to cry "fake video!" and it will be difficult if not impossible to prove whether or not it is legitimate. It used to be that audio and video evidence was the most convincing proof you could come up with in a trial, but before long it will be seen as on the level of something bearing a signature - easily forged and therefore easily dismissed as not being "proof beyond a reasonable doubt".
So the question is, could someone theoretically come up with a way to work around this? Maybe have video recording devices use a private key to encode their identity into the recording, so if you could present the phone that made the recording into evidence it could be proven that the video shown is as was originally recorded by the device and not modified?
Maybe have video recording devices use a private key to encode their identity into the recording, so if you could present the phone that made the recording into evidence it could be proven that the video shown is as was originally recorded by the device and not modified?
That'll work until someone manages to set / recover the secret key, or finds out the algorithm* that sets them.
I estimate about 3 days after it gets to market, after a 6 month pre-launch advertising / promotional drive saying how such a thing it can't be done.
* "But we seeded rand using time() and the serial number and then encrypted it with base32 AND uuencode, and then rot13'd the result nine hundred thousand times, just like it says in the stack overflow comments! No one can break that!"
The technique of photo and video watermarking can ensure any doctoring of the images or frames can be shown, by perturbation of the watermark. You should be able to identify the camera that took the footage as well. In theory, deep learning methods might be able circumvent that in future, but my guess would be that the deep network would need lots of training data of the camera in question.
Some cameras already allow for this digital watermarking.
However at the same time the cameras are getting more powerful and adding analytics capabilities. With this amount of processing power, the camera can generate the deepfake and authenticate it as well.
I wonder if traditional film cameras could be seen as more trustworthy? Of course, one could try to doctor the film by first making a doctored video, then transferring that to film, but an expert would very probably detect this, especially if he can examine both the film and the camera that it was allegedly shot with. The imperfections of cameras leave fingerprints. Besides it would make the faking much more laborious.
This reminds me of two "ideas" I had back in the day (scare quotes because ideas are cheap, execution is the real issue).
1) A digital camera that would encrypt any images, with a generated key, and then dole out <n> partial keys, such that all subjects (participants?) would need to consent to later viewing. Or using Shamir's "Secret Sharing", some quorum of the subjects.
2) A way to record (on analog tape) a conversation with potential risk of prosecution for both sides, including some preamble and postamble around an agreed course of action. Neither party would want the recording to be seen, but the threat of "blackmail", in pursuit of a "less formal court system" could be used to strengthen the "contract". The basic idea was to shoot the video in a room with a television playing local news and a window onto a busy street, plus whatever other "continuity canaries" are agreed.
As with previous ideas about ways to route (landline) telephone calls to make it harder to associate number with physical location, it was not the work involved (although I am lazy), but the risks of dealing with the most likely customers. Groups like Camorra don't have great retirement plans.
Anyway, also seconding the mentions that the tech from the article will almost certainly find its main use to establish "plausible deniability". Who you gonna believe? Me, or your own lying eyes?
It seems to me that they've just used x-rays to see what's inside Pandora's Box - they haven't quite cracked it open just yet; though it won't be long.
If you watch the video, take note of the background in all the examples - including the Mona Lisa. The animation of the face, where head movement is added, causes the background to be messed up, and the more there is on the head that needs to blend with the background (Einstein's messy hair*, beardy bloke's beard), the more obvious the fakery.
This will improve, of course - but for now, just avoid having your picture taken against a flat background, and for added security make sure your hair is a mess. (Burns passport, throws away all combs...)
* In that instance it's worse, the system didn't appear to work out where the hair really ended, so some of it was the background.
That's amazing, watching Mona Lisa talk has to be a quantum step towards the singularity...
- plus nothing has been real, ever.
The universe is holographic - synaptic responses to electrical stimuli. Humanity has believed in the out there for far too long.
William Blake (who died in the 18th century) said...
“Nothing is real beyond imaginative patterns men make of reality.”
"They say the world is a stage. But obviously the play is unrehearsed and everybody is ad-libbing his lines." — Calvin
There is a method to prove what is internal and what is external. Degrees of freedom and control. I can control internal thoughts/actions greater than external actions, and some external actions are out of my control.
Hence, there is an internal world (mind/neurons/code/maths) a physical body (flesh/bone/matter) and a physical outside world (rocks, gravity, earth, sun etc).
Add to that the observation of people and of self, and you get other implications that are unavoidable.
Because then it's not you thinking. It's Descartes' Evil Demon. The "people" in the book I write do not "think", they are writing in a book.
Thus, if anything acts on me, that is not me, so as to *try* to fool me that I am acting (not them), then "myself" does not experience it as me acting, but as someone acting on me.
Proof of this is already available. Have someone move your arms. You may have partial feeling you are moving, but also have partial feeling they are moving you.
Thus if I suggest I do not exist while existing (suggest Descartes' Evil Demon can exist) I hit a paradox in statement due to making a *wrong* assumption. If I take the assumption away, and instead try to test Descartes' Evil Demon, I find out it cannot fully exist, and would be detectable in some form.
Thanks! :)
Think meta. Don't think of people writing in a book. Think of controlling the person who's writing the book. Don't think of moving a person's arm but compelling the actual person to move that arm, all the while the person thinks he/she did it of his own volition. Mental suggestion, persuasion, subtle cues, etc. That's the true power of Descartes' Evil Demon. You don't know he exists because he's outside your context, yet he can alter your thoughts to make you do what he wants, all the while making you think it was our thoughts to begin with.
I did. You missed the point. The "meta" is, you and I have self. If we assume a self exists, then there is one. If we assume there is not, then there is not.
Descartes' Evil Demon by definition, has self in the example. Thus either we both have self, and I can detect the demon, or we do not, and neither can it influence me.
"Don't think of moving a person's arm but compelling the actual person to move that arm"
Assumes movement of an arm is not compelled by thought? If I move my arm, I compel it. If someone else moves it. They did not compel me. The question is, can they compel me with no detection? That is the question, we cannot thus assume in the thought experiment, and if we do, we get erroneous conclusions.
Remove the assumptions you wish to test from the thought experiment. Else it's not a test, it's a set paradox (if use the thought experiment "I *can* go through time and kill my grandfather", I have defined myself as being a paradox, not proven time travel is possible or not ;) I have to ask "*Am I able to* go through time and kill my grandfather" ).
"yet he can alter your thoughts to make you do what he wants, all the while making you think it was our thoughts to begin with." Citation needed, or proof. In that, I can theorise "an undetectable action", but an action is by definition detectable, else it would not be an action! So really, don't get in a twist over the theories, but look to break their mistakes!
"Descartes' Evil Demon by definition, has self in the example. Thus either we both have self, and I can detect the demon, or we do not, and neither can it influence me."
Why? Prove that it MUST be both or neither. I propose the demon has self while you don't AND that the demon is FOOLING you into believing you have self. That's the true evil of the demon. He's tricking you into believing you have self-determination.
"Remove the assumptions you wish to test from the thought experiment. Else it's not a test, it's a set paradox"
That's what I propose; that you CAN'T test it, as the thing you're trying to test is outside your context. Like testing for God.
"Citation needed, or proof."
As noted above, I claim an article of faith, which asserts its truth while also being impossible to prove. Actions need not involve motion. Is not thought an action as well, meaning the Demon can implant a thought in my head that includes the concept that I self-thought the thought spontaneously, as many minds often do when they drift?
I've been predicting for years now that, as 3D rendering gets more lifelike and AI gets better at this kind of interpolation, the obvious culmination of mixing the two is going to be in the movie industry. Why bother paying hundreds of millions to the zzz-elebs of today when, for a fraction of the amount, you can have a walking talking, virtually indistinguishable from life, legend from the past appearing in your lattest blockbuster?
Humphrey Bogart, Marleine Dietrich, Paul Newman, Marilyn Monroe, Steve McQueen, etc. etc....
Their careers have only just begun!
Not sure about that. The ad with the fake Audrey Hepburn was creepy because, although an excellent likeness, it lacked some kind of behavioural nuance that was presumably what made her a great star. Movie acting is an odd, minimal, sort of art - look how stage actors often overdo it. I'm definitely not starstruck, but lead actors who can 'carry' a movie seem to have the ability to project personality.
Whilst you are absolutely correct that great actors add more than just their image to the performance, that's still only an algorithm* away.
Alternatively you end up with 5 brilliant actors who do all the acting in every film and then the studio just maps the face and image onto them to get the final desired product.
We're already seeing de-aging used pretty successfully film and TV, it's not a huge step beyond that.
*perhaps quite a complicated one, but perhaps machine learning can be shown great films and great actors and adapt to understand what is needed to create that in a new film.
<tinFoilHat>
Or indeed anyone that the deep state decides is a nuisance. It used to be that the deep state would have to infiltrate your computer and insert some kiddy pron or pictures of animals before calling plod and let it take its course. Not too far in the future, they will be able to put forward realistic video of Nigel Farage (other controversial politicians are available) at a Nazi rally claiming that Hitler really was a good guy and was just misunderstood. The burden of proof will then weigh on the individuals to refute it. How are they going to do that?
Going down-scale a bit - it could be used to bribe businessmen or bankers into giving loans or business deals "otherwise we'll show this video of you asking a minor for sex to your wife, and also send it to the newspapers.". What are you going to do? How are you going to disprove it?
This technology will NOT be used for good purposes.
I want the 1980s back. Things were a lot simpler then.
</tinFoilHat>
Have to agree.
We already have people declaring real news fake.
And making up fakes and calling them real news.
If I don't agree, it's fake - and you're a nazi.
If I do - I don't check. Confirmation bias.
Sick of it. There needs to be some penalty for outright trolling or something.
Don't even need fake videos - already we have "unnamed sources" and "it's reported as" which people who agree take on faith - but those who don't point out that either can mean it's simply made up out of thin air. Of course in the latter it's even more tenuous. I can report the moon exploded last night to satisfy that one - and we know it didn't. But it would still be announced that "it's reported the moon exploded last night" - not even a lie as such, just ridiculous.
It's in the early part of the book "how to say whatever you want without actually telling an actionable lie".
It's all reality's fault for disagreeing with me.
One observer surmised the reason for Mona Lisa’s so-called “enigmatic smile” was because she was accustomed to hiding what must have been a very bad set of teeth.
Fact is, despite all humans using the same basic facial gestures, every single one of us uses his features in different and unique ways - some subtle; some blatant. Good example, captured well, is Jeff Bridges playing an alien in Starman.
Ergo it would be a very disturbing experience for me to find my wife’s face moving and her thereby suddenly expressing herself differently. Nothing to do with the size of her nose etc.
Recall some uni some years ago suggested communication is only about 37 per cent words.
We are already at that level of mercy. Have been for a looooong time. The "start a whisper in a crowd" way of controlling/confusing people has existed for as long as crowds do. "He said to do this" all the while when they did not, has been a trick for propaganda and lies and deceit.
One "solution" that strikes me is where an event requiring confirmation occurs, requesting back-up evidence from anyone who has it. More suitable for events that happen in a public space, of course, but the number of people routinely imaging their daily life is such that this could be feasible - this would not just be about visual images, sound would be necessary too. The task of stitching together disparate sources to prove chronology would be a monumental feat - which is its advantage, because those that aim to fake the evidence will need to fake the physics too: taking into account such things as the speed of sound between two device recordings sited at different points around the audible area, for instance. Wind direction and position of clouds affecting the lighting of the scene would also need to be taken into account, which is technologically out of scope at present, but if deemed important enough, will need to be considered.
I think that attempts were made to carry out such detailed analysis (within the confines of the technology of the day) were carried out subsequent to Kennedy's assassination, for example, despite it being later discredited:-
https://en.wikipedia.org/wiki/John_F._Kennedy_assassination_Dictabelt_recording
Then there is the analysis Inmarsat did for helping to determine the fate of MH370.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
