> How do you deal with the same in Photoshop?
You're making a false equivalence. "AI" image generation is so much easier to do and requires so much less skill than Photoshopping something that you're really dealing with an entirely different problem.
Back in 2005 or so, kuro5hin (a now gone discussion site) closed signups because somebody photoshopped the founder's wife's head onto some porn. That was 18 years ago.
True, doing it with Photoshop took a bit of skill, but it is a skill a lot of people have, for whom it would be doable in minutes. For a newbie, figuring out Stable Diffusion is probably more work than figuring out how to do it in Photoshop.
And IMO the training argument is long term a pointless waste of time.
It's perfectly clear if you realize similar is not same.
> Back in 2005 or so, kuro5hin (a now gone discussion site) closed signups because somebody photoshopped the founder's wife's head onto some porn. That was 18 years ago.
So what? Everyone here understands that's possible. If you think that example somehow addresses the concern, you missed the point.
> True, doing it with Photoshop took a bit of skill, but it is a skill a lot of people have, for whom it would be doable in minutes.
Also, IIRC, that particular Photoshop of Rusty's wife was terrible, as in obvious.
The skill "a lot of people have" is to make bad photoshops. Generative AI has the ability to near-effortlessly make ones that most people could confuse for a real photo.
> For a newbie, figuring out Stable Diffusion is probably more work than figuring out how to do it in Photoshop.
And how long will that last? You do realize someone's going to come up with an easy-to-run "revengeporn.exe" sooner rather than later?
The answer is a better quality than if they are trying to figure out stable diffusion.
Prove it. Give some newbie Photoshop and a week of time, and show me how well they can Photoshop the face of a particular person (say Tom Vilsack, Secretary of Agriculture) onto some porn.
Literally all some kid has to do is use the Photoshop magic wand feature to copy someone's face and post it on another image.
High schoolers don't care about the difference. They'll harass the target just as much if those kinds of pictures show up around school.
And that would only take a couple minutes to produce.
High schoolers don't need a 3000$ gaming computer, and the skills of figuring out how a GitHub repo and install process works to harass their fellow students.
You are entirely confused about what the problem is here. Slight differences in technology are not the cause of the problem of sexual harassment.
Do it, show me the results.
> High schoolers don't care about the difference. They'll harass the target just as much if those kinds of pictures show up around school.
You're moving the goalposts. I don't care if highschoolers will run with obvious, low quality crap. In fact, throughout this whole conversation, I didn't have highschoolers harassing each other on my mind at all. I was thinking about the more general problem, which includes things like harassing exes and potential employers coming across pictures during a job search.
> which includes things like harassing exes and potential employers
You are completely missing the point. Also, your repeated requests that I engage in sexual harassment is weird.
The damage of sexually harassing messages being sent to your friends and families can be just as damaging regardless of whatever small quality issues that you think exist.
You are confused about what the issue is. People are still significantly harmed via this harassment, even if there are quality issues. The "accuracy" isn't the determining factor here. Instead, it is the sexually harassment messages and images being spammed to people surrounding the victim.
Is this a reasonable summary?
> AI = easy, so it should be regulated. It has passed the "threshold of simplicity" (and realism) where new legislation should be enacted.
> Photoshop = harder, so it should not be legislated.
If so, what happens when Photoshop releases a "copy/paste a face" feature (a desired general photo editing capability) that uses GenAI to merge background, skin tone, lighting, etc.? That could easily be a beginner-level feature (ctrl-c/ctrl-v with auto-segmentation) and be used to create porn.
Are you proposing that the feature be regulated because it's become too easy? That artificial barriers of difficulty be implemented?
Again, what are you proposing be the outcome?
It's a problem software engineers (or some superset that contains them) seem particularly prone to.
And Photoshop is so much easier to do and requires much less skill than carving marble statues.
I say we also ban Photoshop and only allow chisels and stones.
You are frankly missing the point. The issue isn't easier, the issue is crossing a threshold of easiness or scalability that means the problems can't be managed the same way as they were in the past.
For instance: the problem of revenge porn was severely limited before cameras existed and before the internet made mass distribution effortless. Digital cameras + internet distribution removed technological limiting factors, and made the problem significantly more severe and widespread.
But there's still limiting factors that make individual-infraction level enforcement possible: the perpetrator needs some physical access to the victim such that the victim usually knows who took the photo. Generative AI even removes that limitation.
That might no mean anything should change, but it does mean "LOL. I can find similarities so they're the same," isn't a reasonable response.
But I do see mass-scale generation of nonconsensual images, well beyond individually commissioned works -- and actual CSAM blending in with it because it becomes indistinguishable -- as a real problem.
We're having nuanced discussions about how to deal with misinformation, including mass-scale generation of it from LLMs. Why can't we have those discussions about AI porn so that we can come up with solutions that are somewhere between "change nothing/don't be a prude" and "go full authoritarian"?
To me, it seems like we quickly get into the AI image equivalent of prosecuting the actor in a movie for committing the murder of a fictional character. We even go as far to have the entire genre of horror movies at mass Hollywood scale based around fictional death and murder.
It gets even stranger with sex and nudity when such a large % of people are online porn consumers. I have seen the figure for men at 90%. It is where I think the market will actually figure out the solution and that the porn consumer will want assurance the person is real and not AI.
What is disingenuous in this debate to me is that chatGPT estimates there might be over 100 million pictures of real naked people on the internet as of 2021 lol. It is like we are pretending these pictures don't already exist of real people at a huge scale and that no one really looks at them at huge scale now. It is only when AI starts cranking them out people will be so tempted and THAT is a big problem.
If anything, this is the much bigger problem that we can no longer discern the magnitude of a problem in the real world vs the R0 of the idea of a problem online.
"Since everyone is posting about AI generated nudity/porn, it must be a big problem."
Spam email and annoying door to door sales might look like similar problems, but they demand very different solutions.
But we have laws specifically for weapons of mass destruction, instead of just suing those who use nuclear bomb with murder
Then it became easy for everyone and became a problem, so laws were created specifically to combat electronic spam, because current laws didn't cover that situation.
If we encumber the tech with regulation, only the giants will be able to engineer the adequate protections that fit the letter of the law. You'll effectively be choosing the winners and creating a massive barrier to entry for everyone else.
Burdening these systems early on with legislated regulations only makes it so that the incumbents win. Regulations are a moat.
I am angry at EU legislation. While it was heavily involved in research, legislation are again on the course to fortify dominance of established big tech players. Probably because their interests often align with media interests, which have a huge influence on EU politics. And I do not mean the open and independent part of media, rather the established players that have direct business relations with huge tech firms.
I think enforcement should be focused on disclosing what's AI generated/fake, not on banning altogether.