AI-Assisted Fake Porn Is Here
motherboard.vice.com
motherboard.vice.com
Imagine an AI generated 5 second clip of Weinstein saying something scandalous about raping women, or imagine Trump retweeting _anything_. People choose to believe something that fits with or reinforces how they think - emotion and reaction is easier than checking if something is true. Even if it's reported to be a fake after, people would rather question the reporting ("well that's what he'd want you to believe" / "of course they'll report it as a fake") than acknowledge the fake.
I hate to be a harbinger of trouble, but I can see a lot of people having their lives destroyed with this.
It used to be you had to have two independent sources before you took an individual source's word. Now rumor becomes newsworthy --it's like the Enquirer won.
They are also owned by big cos (Amazon, Conde Nast, Carlos Slim, Murdock, etc.) and have very clear biases. NOt that biases didn't exist before (Hearst papers) but on day to day stuff they played it pretty level. Now, they go for the incendiary whereas before they were focused on big issues --the issues that mattered to society at large rather than smaller divisive issues. But divisiveness gets you clicks.
http://wilkins.law.harvard.edu/projects/2017-08_mediacloud/G...
Remember that image? I would bet good money that the amount of green has gone down steadily the last 50 years, and Journalism is at the lowest point it has ever been. I would also predict that the amount of politician donations and lobbying has never been as high in the Journalism industry, nor the amount of quid pro quo between journalists and politicians.
What's more troubling is that as media becomes falsifiable, solid evidence of...well, anything, becomes hard to have. The ultimate loser there is the truth, sadly.
https://www.smithsonianmag.com/history/infamous-war-worlds-r...
One could even hypothesise that social media bubbles are the first step precisely in that direction.
Maybe this new tech will force people to bet their credibility on what they say and spread.
Just take the massive global interest in fact checking sites after the 2016 US election. It seems as a strong trend in competing with media for carrying the banner for truth, and the best part is that many such sites seems to have a non-partisan policy in stark contrast to most media out there. The loser don't seem to be the truth, but rather those who wish to use a few seconds of video with no context to push a point.
No, it will force the development of AI to detect whether things are fake or not.
We'll soon be in an arms race as better-and-better AIs alternately try to fool or not be fooled.
The people who are able to understand and wield these technologies will have incredible amounts of power over those who are not.
This is one route to the singularity - to deal with the consequences of ML and AI advancements, you need to invest even more. This ever-increasing arms race sucks vast amounts of human and financial capital into the 'tech sphere'.
Ideas: https://github.com/pjlsergeant/multimedia-trust-and-certific...
It wasn't the good old days or anything, but that was how it worked for a while. There will be value in being a trustworthy source of information again.
of course, talk is cheap. i dont have the chops to actually spec this out.
How are you not seeing the obvious issues with your idea? People are not going to put compromising photos and videos on this blockchain because they are compromising. And yet some compromising photos and videos taken by other people are real. The fact that it's "on the blockchain" means basically nothing in this case.
Other users who were discussing "hardware signing" are closer to a working system. This is not a problem for a blockchain to solve.
Secure enclave based verifiable recording devices.
So, a camera that uploads a signed hash of every image it takes using mobile connectivity as soon as it takes it. Or the same for video and audio recordings.
It's loosely inspired by Town Crier[0]. The notion is that instead of regarding arbitrary blocks of data as "recordings" we demote those to arbitrary blocks of data, and anything you want to claim as verified truth needs a time-stamped hash proving what device took it, where it was taken, when, and that it hasn't since been modified.
Disclaimer: I'm the furthest thing from a crypto person, this just seems like a nice use case to me.
Maybe cryptographically verifying that a porn video happened is someone's fetish, but I don't think most people involved in the production or consumption of porn would care.
I would make a merkle-tree-like hash to be able to obscure parts of the picture and still be able to prove the date.
The normal risk is the service simply replaces an old hash for video with a new hash secretly. However, if you put data + old hash you can create a long verifiable chain and someone can record a specific block and know the chain is unchanged from the point they saved a specific hash.
http://imaging.nikon.com/lineup/software/img_auth/index.htm
The catch is that it has been hacked before:
https://blog.elcomsoft.com/2011/04/nikon-image-authenticatio...
The Secure Enclave helps a lot but the Nikon hsck illustrates a hazard of needing devices to be updated, especially since there’s an obvious opening for someone to use a plausible old vulnerable device – and it seems like, say, a compromising photo of a politician would spread a lot further and faster than the observation that the older phone has been compromised.
One other challenge is that you might be able to prove that a device took the photo but not that, say, it was a real scene rather than a high quality print or that the time stamp wasn’t subject to a manually set clock. Again, I’d expect countermeasures but worry that nobody would see them in time for the most damaging cases.
The high quality prints problem is more interesting. I’d hope that sort of trick would be detectable by inspection, but then we’re back to cat/mouse games with AI.
If we all switched to crypto verified light field photography, we could at least force hoaxsters to build dioramas :)
It's remarkable to me how little the ML researcher they interview in that episode seems willing to grapple with the consequences of the technology.
It would feel a little grating to me as the interviewee to have the interviewer keep pressing me on these questions that, while totally relevant, are obviously designed to make me have to "answer for what I'm doing" in some Frankenstein-esque sense. Yeah, I get it, it's scary, but this train left the station a long, long time ago and probably I'm not (even as someone working on the tech) gonna have a lot to say about the downsides that haven't already been said.
The thing is, we know what externalities are and how they play out in regards to technology, even if we can't fully predict them. It's trivial to infer that this technology could be abused for deception, and that some sort of authentication technology might help to prevent that. So the responsible thing to do would be to provide both the editing and authentication technologies at the same time, even if neither is perfect.
The editing tech in the OP is also off-the-shelf components, i.e. nothing was invented, just assembled in to the correct form. What exactly are they supposed to have done, just not posted anything about it until they found a way to thwart the technique they just discovered? Most likely this bottoms out in an arms race, so they'd be waiting forever.
Your request seems really absurd and I've never seen technological innovation work like this.
With VR porn, the new sex toys for men and women and now this I’m somewhat happy that this wasn’t around when I was a teenager as I’m really not confident I would have left the room if that was available.
"I Dated a Robot" actually has dating in it.... if we had a human or near human AI datebot that would be quite different since whilst you won't be participating in the continuation of the human species at least at that time you would also not be losing any social interaction skills.
This could be nothing and this could be the same thing as the rage against hustler in the 70's and internet porn in the mid 90's but for someone who grewup with every nasty available when I tried one of the higher end recent VR porn productions on my Oculus the feeling that you get from it is quite different.
I've tried VR porn as a joke like 1-2 years ago when the Google VR stuff started and back then it was pretty garbage it was mostly just barrel projection of a standard porn scene that wasn't shot with the intention of VR. The new stuff is shot with proper cameras, uses 3D audio and it's just creepy how convincing it has gotten so far despite the major drawbacks of the current VR headsets.
We are going to need technological solutions to validate against ai forgeries for our legal system to survive.
An issue that the same algorithms that can be used to differentiate between real and fake can be used to train AIs to make better fakes. But here, the fake detector has the advantage in the arms race.
Another approach is to use cryptographic signatures. Some cameras sign can sign pictures to ensure they are original.
Printing may be a better option with regard to resolution but because the print color gamut is different, simple analysis can probably give away the trick really easily.
Furthermore, assuming EXIF informations are signed along with the rest of the picture, you also make all the metadata match the picture. A sharp picture and bright daylight is not consistant with a slow shutter speed for instance.
Enough people are deceived by photoshopped pictures. Fake video will deceive a whole lot more.
Movie studios could get away with legal problems just by releasing videos without the faces, and letting the audience download faces through gray market channels.
Adam Sandler!
You could have someone who looks identical to Bogart, but that isn't Bogart.
This statement is confusing me in many ways. Can anyone explain the exact logic behind it? Specifically, for “forced to do” and “lack of respect”.
How does it, and implications of it, seem for you?
Maybe a blockchain single source of truth, using tiered identity verification where one can accumulate identifiers and use this ID to sign content (biometrics, other verified IDs can vouch for your verification etc higher score = less risk of fake). Seems like these things all already exist the big problems would be putting it together, scale, and easy clarity. For instance making it so my mom's browser or email automatically verifies signed content (and maybe more important denotes unverified or verified false) and clearly presents this like the green lock key in internet browsers.
I wonder if something like this could be extended to verify beyond like an MD5 checksum saying this email or photo is real - i.e. solving problems like 'fake news.'
For example, I want my bank to know my email address, phone number, and physical address because I want to receive fraud notifications, to make telephone banking transactions, and to receive paper statements.
I do not want forums that I shitpost on to have any of that information, nor do I want it to be written into law that any service I access over the Internet is required to ask for and log that information.
Child works and reposts are problematic too. Maybe could nest below origin content in a blockchain. Obviously re-saving jpgs & different video wouldn't hash to original so maybe it's more like you're cryptographically signing a perma-linked piece of content which is archived in something like IPFS or on this single source of truth blockchain. Maybe this would need scale and uniformity to create content standards; common meta tags to standardize media (og:video -> ipfs:mp4). Hard to get to content outside of always changing format and code let alone dealing with content which has changed in substance. Much more difficult than just PGP sign a block of text!
All of this, IMHO, will shape our general cynicism and distrust of recorded evidence, which will lead the way to us finding other heuristics to sniff out bs. I think this is an overall net gain, but I do agree, there will be some people who's lives are destroyed by what this enables, which really sucks, to put it lightly.
The upside, at least, is that we already have people looking to find ways to deal with this right now.
X people used your face as a swap in for a pornstar. Y$ a month and you can see which state they're from. Z$ to meet the top 10 reviewed individuals near your zip code.
Send a message? [Y/n/generate] Adjust profile picture to his preferences? [Y/n/half way there]
Have every public figure have their own key pair, and each official production have its own key pair. Then publish a signed hash or similar for each production the figure chooses to appear in. Any production without such a signature or with one that can't be verified is probably fake.
Would be a big shift in process, and would only happen if this sort of thing becomes a serious problem, but once it did such a scheme would stop it in its tracks.
Also — do you really need a signed signature to verify a celeb isn’t actually in a porn? It’s not like they’d sign that anyway...
First, the person creating the fakes is using an off-the-shelf algorithm[1], adapted accordingly. A system designed purely for the purpose of swapping faces (or morphing a particular face into others, one-to-many style) would obviously yield improvements.
Secondly, it's the work of one person without commercial backing[2]. A studio that invested in lidar or used multiple cameras would be able to incorporate volumetric information.
Third, it's limited to individuals where you have a performer that's an approximate body double for the target in question. Why involve real people at all? That limits the kinds of scenes you'll be able to shoot, and it's clearly a production bottleneck besides. Since it's feasible to create realistic animations of locomotion via deep reinforcement learning[3], applying the same technique would remove the need for human actors.
With that in mind, we can invent scenarios substantially more terrifying, for which this is but a teaser trailer.
As computation becomes cheaper and AI knowledge more widespread, the delay between when something becomes possible and when it becomes ubiquitous grows ever shorter. Once custom pornography is just another something-as-a-service, the question becomes how do we optimize it? IoT sensors and wearables can provide the data needed to customize an experience tailored to your particular tastes on a moment-by-moment basis. Facebook is sinister, but it's sinister in the sense of a malevolent external force with inscrutable goals; you can defeat it by not using Facebook. The idea of a machine that can produce something more compelling than most stimuli available in the external world is scarier, because all it would be doing is giving you exactly what you want.
Okay, so the riff on wireheading in the style of Infinite Jest might not be adequately terrifying. How about something more realistic, like character assassination campaigns? We've seen how intimations of sexual impropriety can damage a politician's electoral chances or outright cause them to resign. Sometimes, the thing that allows them to hold on to office is the fact that there's no convincing proof beyond circumstantial evidence. However, if you've already primed the public with rumors, a fake sex tape might be the just the thing to push past the tipping point. Imagine: a future where an individual could effect a bloodless coup using nothing more than a Twitter botnet to spread slander and a few GPUs to generate scandalous footage.
No? That's perhaps more in the vein of William Gibson. The last thought that came to mind was along the lines of "the future is weird and terrible, yet somehow boring and familiar" a la Neal Stephenson. The redeeming value of porn is that there's at least some intentionality behind it. Real people had to be involved, from the actors to the crew to the guy who wrote the script in traffic on the way to the shoot. The future will be endless porn spam, half-coherent plots generated by LSTM-char, just glitchy enough to be discernible as fake, yet seemingly unavoidable due to the sheer quantity made possible by automating the production pipeline.
------
0. And when it comes to AI, "could" should be read as "will" but with an indeterminate time frame.
1. Apparently, the face-swapping is accomplished following "Unsupervised Image-to-Image Translation Networks" (paper: https://arxiv.org/abs/1703.00848 video: https://www.youtube.com/watch?v=Bwq7BmQ1Vbc), with the face detection provided by FasterRCNN.
2. The reddit account of the artiste in question: https://www.reddit.com/user/deepfakes (not exactly safe for work, but not particularly unsafe, either).
Turns out you don't really need the scandalous footage
30 years away!
if fake porn does become ubiquitous as you argue, wouldn't the impact of such assassination be greatly diminished?
Anyway, the world is going to be a strange place. I'm sure that soon, nobody will believe anything they didn't see with their own eyes.
I mean, with the fall of eye-trust we will barely lose something important; it's only perception bias.