It should probably be illegal, but the rational here isn't great. Something shouldn't be made illegal just because it makes a detective's job more difficult.

This will stand until the first wave of child rapes and sexual assaults happen wherein the cops find huge amounts of the ai-created child porn in the perpetrators computers.

The prurient interest angle of the images an will then be unavoidable as an enticement and incitement to the assaults and it will be banned.

Unfortunately, the law has to see actionable harm FIRST from something before it can be banned.

Just give it time, as you notice the language the judge used - the "compelling interest" the government has - they know that it should ne illegal, but have to wait until the balance of content is ai-generated and the predators come out to play.

It absolutely sucks to have to unfold this way.

it doesn’t matter if it’s a real child or not

Real children who were abused to make CSAM say that they are re-traumatized when they come across the material or learn that it is being shared. I think most victims would say it does matter.

What I think you're trying to argue is that a generated facsimile of CSAM is also harmful.

There is currently no good evidence that consuming child porn leads to abusive behaviour - there is some contradictory evidence of the kind you mentioned.

As such I think it should be legal unless it is established that it actually does lead to harmful behaviour.

I don't think the model's training data matters that much here, because that's the same regardless of what the model is used for.

It's like how people instinctually want to give child rapists the death penalty: it's counterproductive.

What it leads to is that the child now gets murdered as well, because the guy will try to minimize his chances of getting caught.

I agree, it is a fine line. Another relevant inconvenient truth is that there are pedophiles who have never acted on their urges, never harmed a child, and have even sought appropriate therapy to manage their mental state. Yet at the end of the day, they are still literally pedophiles.

Japan seems to be the canary in the coal mine for this issue. They have long made it legal to draw manga "pornography" featuring apparently underage characters, as long as it doesn't depict an actual child in any way. This is a vice mini-doc on the subject from a few years ago.

People in that culture try to make the argument that it gives pedophiles a way to manage their feelings without harming any children. But in reality, it's seeming more like giving suicidal patients access to guns. The challenge seems to be that: philosophically we want universal free speech, but biologically the human brain isn't evolved to control itself in that environment. You're not staving off a pedophiles' urges, you end up nurturing them. You create a subculture that embraces this mental state as acceptable and normalized. It would be like....building Westworld robots for serial killers to abuse. Surely nothing good will come from creating a safe space in society for people with these urges.

I personally do not agree with any laws against speech, and also I don't think anyone should ever create "artificial" CSAM in any capacity. But I don't know how to square that circle. Because if you try to just make a blanket law against it, on the one hand people will find ways to skirt the line, slowly developing arguably even more depraved genres; and on the other hand you'll have rulings against perfectly above-board, consensual pornography/speech, just because a politician doesn't like it (as have been attempted many times throughout history, and remains a concerted goal of the right to this day).

I have to assume this is what it feels like to be an old-fashioned puritan, seeing a clear line between pornographic content and the erosion of society. But in this case, I think that's where I am. I just don't think it's ok for the future of society to normalize artificial CSAM.

It won't because it normalises it. Groomers often use media like this to encourage pedophilia and convince their victims that it's normal. Litteraly how the pedo side of the fediverse opperates.

Also, the CSAM has to be produced. I know people say AI can just approximate how a child would look doing sexual acts, but that's half the truth. Oftentimes, predators will use victims' names to generate it, and pictures/videos similar to the ones in their CSAM would appear. So now, according to this ruling, the produced media would be legal. That, and how sometimes loli/shota artist use CSAM for reference, the AI will most likely use existing CSAM as a reference as well simply because of the tags used in a prompt.

Who are the victims of AI generated images? This stuff is absolutely repulsive, but the reasoning for outlawing it (that it’s evidence of, and was produced during, the commission of a crime) falls apart when it’s fake. Now if it can be linked to the commission of future abuse, that starts to make sense in the way that reckless behavior is illegal even if a particular incident doesn’t hurt anybody.

That would mean less money to go around to track down the real monsters abusing children because everything spent to track down victims that don't exist is wasted.

Every law is, fundamentally, a restriction on someone's freedom. The key justification for why that's okay is that the law protects one or more people from harm in the process. So in this case the question is "who is being harmed when an AI generates this particular configuration of pixels? How are they being harmed?"

When you dig into that it's more complicated than the simple knee-jerk "that stuff is just plain bad!" Attitude that has been drummed into the general public.

From the article:

For example, researchers have found that "a major dataset used for training image-generating AI contained hundreds of CSAM images."

However, there is no source for this. I would really like to see one because that's a huge part of the issue.

But regardless, this court ruling said that only possession is protected by current laws, not production or distribution.

Any model able to generalize will necessarily learn how to combine "nude" and "child" in the same way that it would learn how to combine "rainbow" and "banana". That's the whole point of the model is to generalize. It isn't required that one particular combination of concepts is in the training data for the model to learn how to do it.

I'm an opponent of the death penalty. If there's a convicted murderer on death row I would oppose executing him.

Does that mean I'm definitely a murderer?