4 ms·

"Not harm society" will depend on what material the models are trained with.

roenxi 2y ago
Running some order of magnitudes, we've had people abusing children and taking photos of it since around 1850. By this point there is a large enough corpus of whatever horrors they want to look at that a model can be trained without doing any further harm. It is a total waste of taxpayer funds to be going after someone if all they've done is generate however many images with stable diffusion. I'm glad in this case he's at least done something objectionable although an arrest seems a bit of an overreach, this seems to be the sort of situation that could be resolved with a fine rather than jail time based on skimming the indictment. If that.
- Sporktacular 2y ago
  How would policing that work exactly? Downloading only old CSAM is acceptable? Technically any child abuse material available was produced in the past, so you could argue any pedophile downloading any of it isn't 'doing any further harm'. But the harm in it comes from generating further demand and normalising the sexualization of children. And than can happen with any material from before or after 1850. If a model is trained on it, it becomes a derivative work that should be treated no differently than a b&w version of a colour abuse image. Not sure what tax dollars have to do with this but IMO it's a small price to pay for an important service.
  - roenxi 2y ago
    > Technically any child abuse material available was produced in the past, so you could argue any pedophile downloading any of it isn't 'doing any further harm'. Sounds like I'd be happy to go with that. You're taking it to the logical conclusion and I generally agree. The problem to me is (1) people abusing children and (2) people paying other people to abuse children. > Not sure what tax dollars have to do with this but IMO it's a small price to pay for an important service. It isn't an important service. With the advent of generative AI models, going after people for possessing dodgy images is now more than likely persecuting people who haven't done any harm to anyone and aren't going to. We really should have learned this lesson after Turing. If people aren't harming anyone then it is OK for them to have bizarre sexual likes and dislikes.
    - Sporktacular 2y ago
      "it is OK for them to have bizarre sexual likes and dislikes." I agree, so long as they are not consuming something that harmed anyone. Just as we know banning ivory sales reduces elephant poaching, contributing to demand causes harm. Putting an AI step in between the CP and its consumption is the same as before, it doesn't change anything.
      - roenxi 2y ago
        It changes everything. It removes the part where harm needs to be done to satisfy demand. If someone says "I want a picture of X!" and an AI generates them the picture, no harm is done.
        Sporktacular 2y ago
        The harm was done. It was just done before the consumption step, just like buying an ivory shoehorn or making a photocopy of an existing image of child abuse. The only way around it is the model is not trained on any child abuse material, otherwise legally it's consuming a CP product.
s1artibartfast 2y ago
how is that? the AI companies aren't commissioning new sexual abuse to train their models on.

Running some order of magnitudes, we've had people abusing children and taking photos of it since around 1850. By this point there is a large enough corpus of whatever horrors they want to look at that a model can be trained without doing any further harm. It is a total waste of taxpayer funds to be going after someone if all they've done is generate however many images with stable diffusion. I'm glad in this case he's at least done something objectionable although an arrest seems a bit of an overreach, this seems to be the sort of situation that could be resolved with a fine rather than jail time based on skimming the indictment. If that.

Sporktacular 2y ago
How would policing that work exactly? Downloading only old CSAM is acceptable? Technically any child abuse material available was produced in the past, so you could argue any pedophile downloading any of it isn't 'doing any further harm'. But the harm in it comes from generating further demand and normalising the sexualization of children. And than can happen with any material from before or after 1850. If a model is trained on it, it becomes a derivative work that should be treated no differently than a b&w version of a colour abuse image. Not sure what tax dollars have to do with this but IMO it's a small price to pay for an important service.
- roenxi 2y ago
  > Technically any child abuse material available was produced in the past, so you could argue any pedophile downloading any of it isn't 'doing any further harm'. Sounds like I'd be happy to go with that. You're taking it to the logical conclusion and I generally agree. The problem to me is (1) people abusing children and (2) people paying other people to abuse children. > Not sure what tax dollars have to do with this but IMO it's a small price to pay for an important service. It isn't an important service. With the advent of generative AI models, going after people for possessing dodgy images is now more than likely persecuting people who haven't done any harm to anyone and aren't going to. We really should have learned this lesson after Turing. If people aren't harming anyone then it is OK for them to have bizarre sexual likes and dislikes.
  - Sporktacular 2y ago
    "it is OK for them to have bizarre sexual likes and dislikes." I agree, so long as they are not consuming something that harmed anyone. Just as we know banning ivory sales reduces elephant poaching, contributing to demand causes harm. Putting an AI step in between the CP and its consumption is the same as before, it doesn't change anything.
    - roenxi 2y ago
      It changes everything. It removes the part where harm needs to be done to satisfy demand. If someone says "I want a picture of X!" and an AI generates them the picture, no harm is done.
      - Sporktacular 2y ago
        The harm was done. It was just done before the consumption step, just like buying an ivory shoehorn or making a photocopy of an existing image of child abuse. The only way around it is the model is not trained on any child abuse material, otherwise legally it's consuming a CP product.

s1artibartfast 2y ago

how is that? the AI companies aren't commissioning new sexual abuse to train their models on.