A16Z Funded AI Platform Generated Images That Could Be Categorized as CSAM

As the quote in the original article doesn't mention CSAM but rather child pornography, I think it is relevant to talk about the difference between these terms.

The term of CSAM (Child sexual abuse material) was introduced to (IMHO reasonably) intentionally change the focus from pornography - linking it with legal forms of pornography - towards the key issue that this material is provided through horrific criminal abuse of children.

However, if we're talking about AI generated images then they indicate a difference. They could be properly described as child pornography, as they are pornographic images involving children, but they aren't really CSAM, as no actual sexual abuse was involved - so perhaps it's time to start treating those as separate terms, separate classes of material which share some similarities (and in some aspects should be treated the same) but have meaningful differences, and in some aspects should be treated differently.

I think the genie is out of the bottle on all content (be it fictional writing, fake photos, interactive conversation) for all forms of stimulation (entertainment, artistic, propaganda, pornographic, etc). If not available to everyone readily, it will be within 10 years.

The time for pearl-clutching is over. The stakes of AI are two orders of magnitude greater than this.

Down with this thinly veiled morality policing. Someone should make a read-only bootable USB flash drive OS, with all the AI models in it plus GPU drivers, etc. That way, people will be able to create anything they want with it and enforcement will be nearly impossible because everything runs locally and is lost on power off.

In this post,

https://mastodon.social/@jasonkoebler/111528504674470914

the following assertion is made, wrt generated images: "the thing I want to stress is that generative AI is hurting people today"

I've been discussing this today. TL;dr IMO this assertion is not true as written.

Whether and how synthetic content, harms real people in any sense (moral, legal, emotional, psychic, reptutationally, etc.) is appropriately a matter of societal debate.

One thing we can say is that historically, there has been great societal ambivalence about whether and how harm is done in cases where the nominal victim is unaware of the harm—especially when it's not simply a case of ignorance, e.g. while one's peers or circles or employer are privy to different information—but ones in which the simple fact of some potentially harmful document existing.

Or, as in this case, of there often being I assume no actual individuals involved, so harm such as it is, is inferred from class identification—to apply to all members of a class. (To put a point on that, is harm done to all children, or any child, should someone generate AI images which correspond to no single child?)

"What you don't know, can't/doesn't hurt you" is often the refuge of those acting in bad faith. This case is more complicated, in as much as there has been an argument made that generated images provide societal benefit by acting as a relief valve for fetishes or pathologies which cannot ever be expressed or engaged with in a non-harmful way.

I am not making that argument.

But that argument has been made. And it's not obviously false, any more than the author's assertion is obviously true.

We really should be clear about the way we reason as a society about these things. And the specific case of CSAM is not a good place to start, in as much as it is all but impossible to reason dispassionately about anything adjacent.

We should find some non-triggering analogous case, and sort out our societal consensus there—and only thereafter attempt to either apply it cases more toxic and radioactive only later.

In the meantime, statements like these are actively harmful to urgently needed (and non-trivial) debate.