52 comments

[ 4.0 ms ] story [ 99.2 ms ] thread
- Australia's proposed law criminalizes both creating and sharing (without consent) sexually explicit images that have been digitally created using artificial intelligence or other forms of technology.

- Sharing, whether by email or personal message to an individual or to a mass audience on a private or open platform.

There are already also laws against it in England and Wales, and ten US states (CA,FL,GA,HI,IL,MN,NY,SD,TX,VA). [0][1]

[0]: https://www.usnews.com/news/best-states/articles/2024-01-30/...

[1]: https://en.wikipedia.org/wiki/Deepfake_pornography#Combattin...

>or other forms of technology.

Sooooo not JUST AI ?

Is this another chink in the wall of "fair use"/"Parody" ?

> Is this another chink in the wall of "fair use"/"parody" ?

This is not the United States. There concepts of "fair use" and "Parody" with respect to free speech and copyright laws may not apply.

Also, armor has chinks, not walls.

Do we really need the ability to create sexually explicit pictures of people? No right is absolute, and the reason I think this new law is needed is previously it was extremely difficult to produce photorealistic pictures of people, while now it’s trivial.
So what if it is possible to generate photorealistic pictures? It’s still just art. An algorithm doesn’t change that.
People are acting as if it wasn't possible to create realistic looking fake pornography featuring real people before diffusion models.

Photoshop's been around for a while.

There is a pretty obvious difference between a) schoolchildren (esp. minors) cyberbullying their classmates/ex-BFs/ex-GFs/relatives/teachers with still images vs b) AI apps/porn sites/ advertisers with $$$ budgets for AI creating lifelike deepfake porn videos of celebrities or user submissions, for profit, intentionally, as a business model.

You'd expect the full weight of the proposed Australian law would be used against the latter not the former. Depends how they implement it.

Here [0] is a US Politico article on both the explosion of this happening in US schools, and the very fragmented state-level legislative and police response to US cases ("20 states have laws... who gets [/doesn't get] disciplined, how minors are treated and who is responsible for taking images to the police — varies widely depending on the state.")

Also, there are obvious civil-liberties privacy considerations (just like the long-running CSAM debate) about who [in Australia] will be mandated to scan users' images: telecomms companies, tech companies, the police or all of these?

[0]: Politico: "AI is shockingly good at making fake nudes — and causing havoc in schools" https://www.politico.com/news/2024/05/28/ai-deepfake-nudes-s...

that's generally the thrust of elvis laws, yes - I'm not sure they're coherent as a concept though.

why is elvis impersonation "traditional" in a way that "dolly parton impersonation" is not? and let's not pretend like there wasn't always a significant commercial aspect, the elvis impersonator officiating your wedding wants to get paid.

I don't understand the downvotes and you're changing the topic from deepfake AI porn (whether by individuals (schoolchildren, adults, revenge porn) or AI apps/ advertisers/ social media) to Elvis laws (prohibitions against unauthorized commercial use of AI voice/image/name, e.g. musician soundalike, as passed by Tennessee in 2024).

I'm not defending either cyberbullying or any particular set of laws (and the Australian one hasn't even had its wording published yet), I'm just pointing out each country or state will act(/not act) differently. For Australia, I wonder about their claim they had 2.7x as much deepfake-porn cyberbullying as other countries. (Is Australia using NGOs to drum up support for invasive govt powers which would be used more widely? Will govts worldwide use alarm about deepfake porn as a pretext for sweeping power grabs?)

>2% of TN's economy is music industry, so it's inevitable the state legislature will attempt to protect that from AI job loss as is clearly happening in other audiovisual industries; courts and time will tell.

> why is elvis impersonation "traditional" in a way that "dolly parton impersonation" is not?

I never said that was the case, I don't even know that that's specifically the case, but given you say it is, why do you think that is the case, and how is that related to this discussion? (Does it depend on how permissive/aggressive the rightsholder historically was on enforcing copyright? or else what?) Anyway, Tennessee rushed out a law (which doesn't even go into effect until July 2024) which on the face of it is not intended to target (human) amateur tribute acts, but AI impersonation. Is it overkill, like some suggest the Australian proposed criminal law could turn out to be? Will it be unenforceable/ appealed/ overturned in the SC...? Will other music-industry states like FL, NY, GA pass similar legislation?

It's quite likely many countries' or states' AI-related legislation (or lack of) will be incoherent, this sector is changing too fast.

Agreed. But is 7 years of jail for a joke the right punishment?
I don't see how it is a joke.
See: "Who's Nailin' Paylin?"
It has nothing to do with this law.
I dunno have you seen it? I was pretty convinced that was actually Sarah Palin. Even had the same glasses.
Non-consensual deep fake porn is no joke. Although 7 years jail seems a bit ridiculous; this seems like it'd be more appropriate for the punishment to be a fine under most circumstances. I'm not sure what jail time is supposed to achieve here. This is probably going to pick up a lot of idiot teenagers.
Real non-consensual revenge/rape porn is even less of a joke and PornHub distributed it for years and earned from it in ads, allegedly refusing to take it down after victims requested it. Let's check how many of their execs spent even a month in jail, let alone 7 years...

Not saying PH is Australia's business but if it was would they be behind bars for longer amount of time? If yes then at least it can be somewhat fair.

not saying either way, just that Australia has a habit of turning innocuous laws into clubs to beat people with.
Australia has fair use copyright provisions, as well as parody. They are not protected via the same mechanism or via the same wording as the US.

Now this said, no, this is not another chink in the armour. There is little to no fair use or parody that would withstand prosecution when it comes to deep fake pornography. One could try to make the argument that, for example, a media company may create a deep fake pornographic video for a skit. However, this is both in bad taste, flies afoul of content deceny laws, and the 'artwork' can readily be taken out of context and distributed as pornography in its own right.

Australia has been quite good, if a bit slow, at enacting considered legislation to protect against exploitation of the individual. Our conservative facebook and sov cit crowd constantly bellyache about freedom of speech, forgetting that Australia does not have freedom of speech provisions, but rather a set of legal precedents, case law and various legal interpretations which provide similar, but with an ability for interpretation to enact various protections. They're very difficult to circumvent, but are actively used for things like this by governments of the day. Also, unlike the US, our justices are not elected by politicians, and they must not be aligned to a political party. So parliament could go rogue, but the laws would be challenged and shot down if in the opinion of the court they were illegal or eroded other legally decided precedent or interpretations.

> Also, unlike the US, our justices are not elected by politicians

I'm pretty sure in practice they are elected by politicians. How do you think they are they appointed? They aren't elected directly and the alternatives I can come up with to political appointment are outrageous affronts to democracy.

In Australian states and in Australian federal courts the process is similar; to become a judge at level {X} there's a merit bar to pass (prior skills | qualifications) and prospective candidates apply for the position when vacancies appear (like other jobs).

The applicants are assessed by state or federal peer committees who give advice to state or federal attorney generals (head lawyer) which is a qualified position held by someone who is elected (elected to represent a district, not elected to be attorney general).

The Attorney General appoints Judges who apply and are qualified, those appointments can be appealed on various grounds.

Eg: In the State of Western Australia:

https://www.wa.gov.au/government/media-statements/McGowan-La...

2 x District Court judges appointed by Quiggles : https://en.wikipedia.org/wiki/John_Quigley_(politician) who's quite the chap, he was a lawyer for the police for a fair while until he turned and gave evidence on police corruption.

In the Federal sphere: https://www.ag.gov.au/legal-system/courts/judicial-appointme...

Sure, but that is over-complicating it a bit for the purposes of what anakaine wrote. An A.G. is a politician with the current executive government and consults with same group of people who organise the whips for legislative votes when making decisions.

That isn't differentiating the process from the US except in a cosmetic fashion. If a judge could get the whips' support in the US for a vote then they can get the AG's support in Australia through the same channels. If 51% of the ALP party room disagrees with Quigley's choices he is going to make different choices. Although there is also an argument that it'd be easier to get partisan results because only the AG has to be shameless and he is already selected as someone deeply embedded in politics. If there is a difference I assume it would be who is allowed to be in the candidate pool.

This would only affect sexually explicit parody / fair use.
Parody means imitation for comedic purposes.

There is nothing funny nor fair about creating deep fakes of women and distributing it to fellow students, colleagues etc.

And here's an Australian article from [0], quoting some stats and an Australian academic. I couldn't easily find any legal expert analysis, or the text of the proposed bill. Much Australian media paywalls US search engines, so go read the full article:

> Professor Nicola Henry, a socio-legal scholar at RMIT University in Melbourne... ran a survey last year of people who have had fake images shared and found that Australia had far higher rates of victims (3.7%) compared with the global average of 2.2%.

Interestingly, her research also found that men were more likely to report being perpetrators as well as victims of deepfake pornography.

“Unfortunately, there’s not a huge amount of research that’s been done to examine the extent or the impacts of deepfakes or AI-generated image-based abuse,” Professor Henry added.

The government will also bring forward legislation in early August to outlaw the release of private information online with an intent to cause harm, known as doxxing.

[0]: https://www.news.com.au/national/politics/six-years-jail-tim...

(comment deleted)
Reminder that deepfake porn is just free speech. Someone expressing things that most people have in their heads, like fantasizing about some celebrity, shouldn’t be illegal. Claiming it’s real should be illegal - but that’s already covered by defamation law.
As stated in the title this is about distributing content.

You are free to produce this content in the comfort of your own home.

At what point does content distribution happen? If someone distributes a seed and prompt that reliably generate a deepfake of a celebrity using open source models and tools, is that covered?

This is ultimately about compression, and the famous DeCSS t-shirt and illegal prime number[1].

Not taking a moral position here, but regulating distribution independently of creation for personal consumption seems difficult.

[1] https://en.m.wikipedia.org/wiki/Illegal_number#Illegal_prime...

If so, then maybe it should be defended on accessibility grounds? After all, it could be presented as helping people with aphantasia approximate what people with working visualization ability can literally do in their head. "Undressing someone with your eyes" is, apparently, not a metaphor for some (most?) people.
> is, apparently, not a metaphor for some (most?) people.

I take it you have aphantasia then? I wouldn't say it's "not a metaphor," as it's clearly not literal, but yes, most of the population can see at least faint images in their minds with some effort or when unfocused (daydreaming). The detail of the image and effort required to contruct and maintain it exists on a spectrum. But yes, people with visual imaginations have all "seen" images that have never been physically produced.

> I take it you have aphantasia then?

Yes, I do. Which is why I am always fascinated by the many things I could never understand, or thought were metaphorical, that turn out to be literal for those who can visualize. Like, how my parents and teachers couldn't comprehend why I was bored out of my mind by books with lots of scenery descriptions (which I had to skip to stay with the book) - they said they were the beautiful, important, most enjoyable bits, that made the world vivid, etc. Turns out they are... if you have the ability to visualize stuff. Had I learned about aphantasia as a kid, instead of adult, it would've saved me a lot of heartache growing up.

It might be free speech in the USA but Australia is a sovereign nation with its own legal system so nobody cares what people in the US think.
We are not debating sovereignty. We are debating the absurdity of Australian law, which is increasingly authoritarian, and has no principled protection of free speech.
US ideals of free speech are not required for a functional democracy and misunderstandings about the universality of US laws leads to misunderstandings online and does need to be corrected unfortunately.

IMO Australia desperately needs better protections of political speech but that has been debated for years and some people have the view that political speech is best protected at the ballot box.

Only a small minority of Australians would believe freedom of political speech should be extended to protect sexual abusers.

Australia is not close to being authoritarian. It is a liberal democracy with very fair elections. There is an unhealthy media concentration and our lopsided economy creates an environment where certain industries have too much political influence but that isn't unusual amongst our peers either and something we need to work on.

>Reminder that deepfake porn is just free speech.

That "reminder" is incorrect

In America, it is. See Ashcroft v Free Speech Coalition (2002).
It is literally speech. If pixels are generated, it’s art. If it is falsely claimed to be real, then it is defamation, which is already illegal.
My tinfoil hat theory is that: A big chunk of the real reason these laws are being enacted is to protect the porn and OF model (and not to protect real victims of sexual violence, that's secondary in their minds). With AGI, automation, etc, a lot of people's "value" is in simply being attractive / unique and one-of-a-kind. So these laws are being put out to protect that value from being destroyed and commoditized. I think "TPTB" see that what little human value is going to be left soon is in people (mainly women) selling their own bodies and their goal is in protecting feminist "empowerment" ideologies from being shattered (like men's labor was).

I.e. what's the point of OF models or porn actors if any "stunt double" or AI can do the deed, and you realistically put someone else's face on, or any face to your choosing/liking. Industry disrupted, value ruined, and yet again automation and AI taking over more of humanity's value.

Australian law and courts have sovereignty in Australia, not US law.

Second, what you wrote is out-of-date even if talking just about the (20) US states which do have deepfake laws, Politico article [0] discusses that there is very fragmented [US] state-level legislative and police response to US cases ("who gets [/doesn't get] disciplined, how minors are treated and who is responsible for taking images to the police — varies widely depending on the state.")

Third, it's quite likely that more US states (plus say NZ, Canada, EU countries et al.) will rush to legislate explicitly against deepfakes, cyberbullying, and possibly doxxing, given how fast technology is advancing.

Fourth, the tech industry, social-media companies, telecomm companies et al. will presumably try to come up with some common harmonized guidelines, because they operate in multiple countries. (But that gets us back to the ongoing CSAM - online privacy debate). Australian legislators will take a different approach to US ones on govt powers to scan people's images and messages, esp. without a warrant.

[0]: Politico: "AI is shockingly good at making fake nudes — and causing havoc in schools" https://www.politico.com/news/2024/05/28/ai-deepfake-nudes-s...

Those state level laws wouldn’t hold up to a challenge. They nevertheless have a chilling effect on speech and set dangerous precedents.

> Third, it's quite likely that more US states (plus say NZ, Canada, EU countries et al.) will rush to legislate explicitly against deepfakes, cyberbullying, and possibly doxxing, given how fast technology is advancing.

This is just not going to happen without constitutional amendments, at least not in a way that sticks.

I made multiple predictions about multiple countries plus some US states.

> "...at least not in a way that sticks."

The situation is changing very fast, faster than legislators can move. Even a bad law can give victims some power of legal discovery to prevent deepfake porn of them being made or published to public forums with the intent of harassing them. Equally they can also have a a chilling effect on (non-deepfake-porn) speech and set dangerous precedents.

I'd expect NZ, Canada and western EU countries will (or already are) rush out some legislation. In the US it will depend on the individual state; 20 already have some legislation, but per the article their laws vary hugely. The Canadian federal Privacy Commissioner is scrutinizing Pornhub (Aylo Holdings) for previously allowing material without adequately verifying consent.

All this area will surely be an active matter in policy-making, legislation and courts, ongoing: non-consensual deepfake porn and related issues (online identity rights, e.g. the Scarlett Johansson soundalike controversy), and the counterbalance on privacy and government intrusion.

A standard might emerge for digitally signing explicit photos/videos by multiple parties, allowing people to add or remove consent.

> new criminal offence of sharing, without consent, sexually explicit images that have been digitally created using artificial intelligence or other forms of technology.

It will be interesting to see how precisely the law is written. If the definition is too broad, it may limit use of soft pornography in advertisement and popular culture.

I do not want to see yet another digitally enhanced six pack in movie. On actor that looks like me... Sharing this stuff makes me deeply uncomfortable. (sarcasm, but not really)

I find it worrying when news outlets publish articles that simply repeat politicians talking points. We all know of laws that say one thing on the tin but are much more draconian in practice.

This article fails to analyse the wording of the law, or cover any dissenting opinions. It’s taking the Australian government at face value.

Of course the wording isn't given:

    The attorney general, Mark Dreyfus, is expected to introduce legislation on Wednesday to create a new criminal offence of sharing, without consent, sexually explicit images that have been digitally created using artificial intelligence or other forms of technology.
it will be available to all (reporters and the general public alike) once tabled to be considered on Wednesday.

After that it will be debated, modified, and have to pass through two houses of government before it becomes "Law" and even then it will still have to pass through the "live testing" of actual charges and a non trivial court case on a real world application.

They can be held liable or accountable for interpreting the details of the law incorrectly. This way the only thing they can be wrong for is misquoting what someone else said.
What's the definition? If you generate porn with Stable Diffusion, which generates generic faces, is that a criminal "deep fake"? This is similar to a "right of publicity" law. For those, there has to be an identifiable person who is the victim. Is the law that narrowly drawn?