Marking things as adult content?

eblu

I’m sure there’s a lot of discussion about age-verification laws around here right now and for the sake of keeping things on topic I won’t really broach the subject here, but it has gotten me thinking that there really isn’t much that can be programmatically marked as adult content on the fediverse.

I haven’t dived too much into researching the subject, it looks like Lemmy lets you set posts as NSFW, but most activity is centered around microblogging and that appears to have coalesced around Mastodon’s approach of freeform content warnings. This seems like a disaster in the making if “don’t show adult content to minors” becomes something that has to be more strictly enforced; these content warnings can be used for everything from benign spoiler warnings to very obviously signposting sexually explicit fetish content. Computers can’t really understand this level of nuance unless you throw something that does natural language processing at it, and that will almost certainly come up with a lot of false positives and wasted energy in the process; I can’t imagine this going over well with anyone really.

So, I’ve been wondering, how difficult would it be to standardize a separate mature-content warning from the content warnings currently in place? This idea has clearly been floated before (see this issue on Mastodon’s GitHub and this blog post written by someone who was a minor and directly affected by this issue at the time) but I haven’t actually seen any work towards anything beyond paying lip service to the subject. Maybe it could be a boolean toggle, like how the former Cohost did it (on top of content warnings) or something closer to how Bluesky does it where you have a few set moderation labels that you can apply yourself (see below).

We could also consider moving this distinction beyond posts; the Mastodon issue that I linked above also mentioned applying this to users and even entire instances.

There are a few caveats here in that people historically don’t really appreciate being hidden/deboosted for posting adult works, and there is the potential for backlash if something gets marked as adult when it really isn’t. I’m not entirely sure how this could be addressed beyond leaving this to implementers and maybe leaving some strong advice to be understanding and not shove people in a corner because they draw kink art for example.

I’d definitely appreciate more thoughts on the subject, please let me know what you think.

Rimu

PieFed uses Lemmy’s sensitive attribute for nsfw content and has also added nsfl (with an L) for gore and combat footage.

Rimu

Also I am strongly considering adding gen_ai which is an integer, 0 - 100 indicating the percentage of AI-generated content in the post.

Emelia Smith

Hi! We’re actually working on a specification for content labels:

https://github.com/swicg/activitypub-trust-and-safety/issues/41

essentially a Note (or other object ) can have many labels associated with it, and these labels would exist as part of well known vocabularies, such that software can give users better choice over what they see and don't see.

Yes, that does mean software may provide methods of complying with age verification laws may mean certain categories of content are unavailable without some form of age verification (but that's between you, your server software, and you instance administrator as to what that is). Currently there are some tools for instance administrators, particularly of mastodon to completely filter certain content from their servers, making their servers somewhat explicitly child-friendly.

This would also allow for third-party labellers in the future if needed (through annotations), which allow for bluesky style labellers which can catch content not self-labelled.

I want to stress that the goal of content labels is not to moderate the adult content nor queerness from the fediverse, but rather to give creators and consumers of content more control over what they publish and who sees it or what they see.

It is unfortunate and terrible the way that age verification is being rolled out as a means to censorship and authoritarianism, and these laws should be fought in the courts and politically to be repealed or changed. Adults must be able to exist on the internet, not everything is for children.

Emelia

I should note: content labels don’t replace as:sensitive from ActivityPub Miscellany, but rather they compliment each other. Software may decide to automatically mark certain content labels as as:sensitive or hide them from display or place them behind content warnings, but what we envision is something more like how labels work on bluesky.

eblu

oh, I see this has already been halfway solved, cool! I really like that we’re moving towards atproto style content labels, it’s definitely the more flexible option. I’d say that maybe it would be a good idea to standardize a few important labels before things officially get recommended (so that we don’t have an issue where e.g. server A has a post labeled adult and server B has one labeled nsfw and neither of them are detected on either side because of the mismatch) but this might have already been floated somewhere, I didn’t see it in the github issues that were linked above. either way there seems to be some sort of coalescing here, good work rimu@piefed.social!

also, I really appreciate this clarification, thisismissem:
> I want to stress that the goal of content labels is not to moderate the adult content nor queerness from the fediverse, but rather to give creators and consumers of content more control over what they publish and who sees it or what they see.
>
> It is unfortunate and terrible the way that age verification is being rolled out as a means to censorship and authoritarianism, and these laws should be fought in the courts and politically to be repealed or changed. Adults must be able to exist on the internet, not everything is for children.

this has been worrying a lot of people and I’m glad to see someone acknowledge this!

Emelia

eblu yeah, I’m very sex worker positive, and I’m queer myself, so I definitely don’t want to see these labels used for harm. It’ll probably take some amount of work to come up with an initial set of labels, but we basically support things like “same as this other label”, “broader than this other label” and “narrower than this other label”, which pulling in a graph like that results in some rather powerful capabilities.

Labels also have full descriptions, and all properties are translatable, here’s how that looks in my FIRES project: https://hachyderm.io/[thisismissem](/user/thisismissem%40activitypub.space)/114755378976572686

Welcome to Caint!

Marking things as adult content?