MP Maria Miller wants AI 'nudifying' tool banned

> Hopefully we very quickly get to the point that nobody will ever be able to know if a photo is real or fake and it’s just not considered an issue therefore.

The ongoing "fake news" crisis proves that we are, in general, bad at spotting fakes - that is the purpose of a fake - and will also vehemently disagree about what is fake. Especially when it can be used for political purposes. Expect the first world leader brought down by a deepfake in the next decade.

I'm reminded of https://www.theverge.com/2017/7/12/15961354/pakistan-calibri...

In America, the federal child pornography law applies only to depictions of an actual child (and you have to know it, for possession offenses, though that’s another matter). But the Justice Department has long taken the position that an image of a clothed child that’s altered to then make the child look nude—-they used to call these “morphed” images—-counts. I don’t think it’s ever been definitively resolved by the Supreme Court, and I don’t know what the courts of appeals have said, but tools like DeepSukebe have made that argument way more appealing. I’d bet that this is where regulation will begin: images of children. That has always been a domain where American courts have been extremely reluctant to intervene; for example, any visual depiction of a seventeen year old engaged in sex is proscribable without resort to the ordinary inquiry into whether the work as a whole is “obscene,” etc.

But under reigning American First Amendment law, it gets a lot harder to explain why a law like the one being proposed here would be acceptable. The Supreme Court has, for example, held that the distribution of animal-cruelty videos cannot be forbidden. And it’s not clear to me how one could proscribe the distribution of an imaginary visual depiction of an adult who was nude. You could call it defamatory, I suppose, but if it’s concededly fictional… I don’t know.

How effective could it be short of an international convention?

Reality is, unless most countries ban it, it's gonna be on the internet.

Note that this is already an offense in Scotland. https://www.legislation.gov.uk/asp/2016/22/section/2

> A person (“A”) commits an offense if—

> (a)A discloses, or threatens to disclose, a photograph or film which shows, or appears to show, another person (“B”) in an intimate situation,

> (b)by doing so, A intends to cause B fear, alarm or distress or A is reckless as to whether B will be caused fear, alarm or distress, and

> (c)the photograph or film has not previously been disclosed to the public at large, or any section of the public, by B or with B's consent

(my highlighting: "appears to show" would cover realistic fakes).

> It’s a shame the headline communicates the former when it seems the proposal is the latter.

Because a headline blubbing about AI-X-ray-specs grabs attention. Making people more likely to click the link.

You can see the effect in action right here: the link made it to the top of the front page in less than an hour of being submitted.

Besides that tough, the article does contain the typical misunderstanding of politicians about how software works ("software developers should be forced make $SOFTWARE so that it can't do $MY_SPECIAL_EXCEPTION_CASE"), which is problematic per se, unrelated to any machine learning aspect.