[DISCUSSION] Forming an AI generated content detection team

Summary:

With the rise of AI comes a force to equal it - people with free time

page 1 of 212next »

[DISCUSSION] Forming an AI generated content detection team

OriTiefling 06 Jun 2024 15:37

O4 link: https://scp-wiki.wikidot.com/forum/t-16859006/discussion-forming-an-ai-generated-content-detection-team

As we've seen in recent moths, AI generated content posted to the wiki continues to be an issue. We currently have no formalized process for identifying AI generated content, nor do we have specific personnel tasked with handling this. As AI models become more sophisticated, we have an increasing need for vigilance on our end. So-

Proposal: Form a team dedicated to identifying AI generated content.

This team's duty would be simple. It would be a clear group of individuals who are tasked with identifying AI generated content using common, key tells (that we would establish in policy). This team would do the investigation, compiled evidence, and submit to Disc/Admins so the user in question can be revoked/banned in accordance to existing policy regarding AI generated materials.

Benefits: Having a clear group of individuals tasked with this will make reporting AI generated materials easier for both staff and users- they now know specifically who to contact. This will also free up staff spaces from current reporting practices that can clog chat, these reports are now handled in a separate space by the team in question. Reports no longer need to be a disc issue, and revokes/bans can be logged in a single master thread on O5 rather than individual disc/nondisc threads. Again, freeing staff spaces from clutter.

Downsides: Right now the innate downside is the fact that there will never be 100% solid evidence that a user's content is AI generated, at least in the case of text. As a result internal policy/guidelines for content identification runs the risk of false positives. Likewise I can see potential concerns for this team being moved out of general staffchat into an existing/their own team server- the investigations become less visible to staff as a result (even if said spaces would remain public to staffers.)

What I need: One, I need bodies. I need people interested in joining this hypothetical team who are willing to participate in this team's potential workload (i.e. reporting, investigation, discussion). Two, I need assistance forming policy related to this team and its operations.

Policy questions: I'll admit I've never been a policy guy, so please offer further policy questions that I miss.

1. Should this team stand as its own entity, or become a subteam of an existing group (MAST, Crit, Disc, etc.)

2. How should this team be structured? Is a simple Captain/Vice Captain structure sufficient?

3. How should this team formalize reporting of AI generated content, and what expectations should be held regarding investigation and compiling evidence?

4. How should final reports be made? Should they be made directly to Disc, or simply an Admin so the user may be revoked/banned?

5. How should finished cases be logged?

These are the questions I'm currently able to pose. Again, I'm not a policy wizard so while I do have thoughts on all the above I would like to see discussion from interested staffers first. I am also willing to serve as this hypothetical team's captain/section head.

This discussion will be open for one week. There is also an ongoing thread in the public channel of the official staff discord server.

Last edited on 06 Jun 2024 15:38 by OriTiefling Show more

Re: [DISCUSSION] Forming an AI generated content detection team

AriadnesThread 06 Jun 2024 15:49

I am very pleased to see this team being formed, because it's sucking the life out of everyone trying to stay on top of it, but ironically it's also forced us to get a pretty streamlined process for the constant issue!

I will say that I think having it as its own entity makes sense, although putting it under either MAST or Disc or Lisc would also make sense logically. However, I do want it to be okay for JS to join it, which I believe some of those teams currently restrict. Thank you for taking the lead on this, Ori!

Re: [DISCUSSION] Forming an AI generated content detection team

DianaBerry 06 Jun 2024 15:57

I’ve been waiting for this for awhile. I’ve already mentioned my interest in it, but I won’t get much into that here.

Now to answer the questions.

Should this team stand as its own entity, or become a subteam of an existing group (MAST, Crit, Disc, etc.)

How should this team be structured? Is a simple Captain/Vice Captain structure sufficient?

I think captain vice captain and then members works fine for me.

Personally, I think it should be under disc or licensing.

How should this team formalize reporting of AI generated content, and what expectations should be held regarding investigation and compiling evidence?

I think Zoob has a pretty good template for what we do already. Run it through 0GPT, mention the hallmarks of AI. Personally if this becomes a team it should either have a server, or a channel in the server for the team it becomes part of. And possibly in the staff server so it’s not just done in general.

How should final reports be made? Should they be made directly to Disc, or simply an Admin so the user may be revoked/banned?

It should be made to disc. And if it’s a non staff member, it should be made to a member of the subteam.

How should finished cases be logged?

I admit I’m not sure on this one yet.

Overall I think this AI subteam is a very good idea as it’s very prominent on the wiki.

Re: [DISCUSSION] Forming an AI generated content detection team

Prismal 06 Jun 2024 16:15

Glad this is getting discussed! Been a long time coming.

1. To me it seems apparent that this should be an arm of the Disc Team. Ultimately this AI detection team's goal is to sus out AI usage for the purpose of taking punitive action against people, at least to my understanding. I don't think under our current structure and charter that it would make sense to do otherwise. You could have it stand alone but then you'd have appeals and punishment for certain disciplinary action go through a new team, which seems like it might get confusing for users and in terms of purview in some scenarios. IMO it's just all around easier to make this part of the Disc Team

2. Seems to make the most sense to me. I will mention that the Disciplinary Team is still without either Captain or Vice-Captain, which I strongly think should be fixed anyway, but especially if another team/subteam gets added to the mix.

3. This I don't really know. I will say that clearly we've already informally sorted this out since we do dole out action for AI usage, based on people's investigation. I suggest looking to that and attempting to formalize that. I can't be more detailed than that since I've not really followed AI detection and disciplinary action.

4. I'd say same way Disc or AHT does it as needed, ping an admin, tell them what you need them to do and have them do it.

5. One thing that I think should be discussed here is how transparent the team should be about the evidence collected/methodology used. In most other situations I'd advocate for maximum transparency insofar as how the investigation was conducted and what evidence was used to determine AI usage, however, if this is done then we risk having people remove the hallmarks of the AI generation that we use to detect the text. All the same, you have to leave it open to allowing people to appeal, so I think there's a balance to be considered there, but I'd avoid tipping people off to the AI detection methodology

6.

Re: [DISCUSSION] Forming an AI generated content detection team

Mew-ltiverse 06 Jun 2024 16:21

1. I’d say it should be a subset of crit or licensing
2. I’d say Captain/vice captain structure should suffice
3. From what I’ve seen, Zion’s format is pretty good. Check 0GPT, and what us as writers know are AI staples.
4. I think final reports to disc works.
5. Unsure

Glad this team is coming to fruition, I hate the uprise of AI on the wiki, and it’s sad that it’s had to come to this, but it’s unfortunately necessary :p

Re: [DISCUSSION] Forming an AI generated content detection team

EstrellaYoshte 06 Jun 2024 16:32

1. Should this team stand as its own entity, or become a subteam of an existing group (MAST, Crit, Disc, etc.)

I'd say Disc subteam, sine it's the easiest to structure, but otherwise it's fine enough to act mostly standalone - All Disc members can weigh in on this subteam's proceeding, but whether members of this subteam can decide on more general non-AI cases is TBD.

3. How should this team formalize reporting of AI generated content, and what expectations should be held regarding investigation and compiling evidence?

Beyond reasonable doubt I guess.

4. How should final reports be made? Should they be made directly to Disc, or simply an Admin so the user may be revoked/banned?

Re: [DISCUSSION] Forming an AI generated content detection team

DrLeviStarkin 06 Jun 2024 16:51

I love this idea, and I would be very interested in joining this or helping out in any way possible. My main question is if this will be its own team, or a subteam under crit/mast/etc. From what I saw in other comments, there hasn't been a clear answer, but maybe I'm wrong.

Edit: Found the answer.

---

Levi Starkin

Last edited on 06 Jun 2024 16:54 by DrLeviStarkin Show more

Re: [DISCUSSION] Forming an AI generated content detection team

OriTiefling 06 Jun 2024 19:03

I don't mean for this to sound harsh, but in the future please make sure that you're reading posts in full before commenting on them. This was very much addressed in the initial post.

Re: [DISCUSSION] Forming an AI generated content detection team

Zoobeeny 06 Jun 2024 17:11

Should this team stand as its own entity, or become a subteam of an existing group (MAST, Crit, Disc, etc.)

Being a subteam of Disc makes the most sense to me. While there are elements of AI-detection that touch on the areas of the other groups, the two focuses of this group are "determining AI usage" and "determining if punishment is applicable". The entire latter half is solely a Disc action and would benefit from being part of Disc itself instead of having to port every single case to Disc for adjudication.

How should this team be structured? Is a simple Captain/Vice Captain structure sufficient?

Captain/Vice Captain structure would work. I don't see any reason any other structure would be necessary.

How should this team formalize reporting of AI generated content, and what expectations should be held regarding investigation and compiling evidence?

Investigation should be comprehensive so as to avoid false positives, including analysis of the article/work in question, comparisons to previous examples of works (both previous articles and sandboxes, if applicable), and revision histories. Specific examples of AI-generation in the article itself should be highlighted to the rest of the team. Oddly enough, despite what others have pointed out above, I don't personally use ZeroGPT all that often due to the unreliability of AI-detection software. I mainly go by what I see myself.

How should final reports be made? Should they be made directly to Disc, or simply an Admin so the user may be revoked/banned?

If the team is formed as a subteam of Disc, then reports can just be internal to Disc. If it's an adjunct or separate team, then final reports should be ported directly over to an Admin so as to not require Disc to have to further deliberate on the issue as decided upon by the Anti-AI team.

How should finished cases be logged?

I believe both in-progress and finished cases should be ported over to a new forum category on O5. Unless we want to do a more "ban-wave" approach so as to minimize the particular tells the team uses to determine AI usage.

Last edited on 06 Jun 2024 17:20 by Zoobeeny Show more

Re: [DISCUSSION] Forming an AI generated content detection team

AstersQuill 06 Jun 2024 19:54

Should this team stand as its own entity, or become a subteam of an existing group (MAST, Crit, Disc, etc.)

Subteam of disc or licensing works best. I don't think there's enough here to justify the creation of a new entity specifically for this team, especially given the overlap with existing teams.

How should this team be structured? Is a simple Captain/Vice Captain structure sufficient?

Cap/VC structure works, though I feel like this team would work best with a more horizontal heirarchy, similar to how discord crit functions.

How should this team formalize reporting of AI generated content, and what expectations should be held regarding investigation and compiling evidence?

Echoing Zoo's comment given their expertise on the topic

How should final reports be made? Should they be made directly to Disc, or simply an Admin so the user may be revoked/banned?

Reports should be made to both Disc and Admins

How should finished cases be logged?

O5 is most likely the best bet, similar to AHT or maturity bans in my opinion

Re: [DISCUSSION] Forming an AI generated content detection team

Uncle Nicolini 07 Jun 2024 00:30

1. Should this team stand as its own entity, or become a subteam of an existing group (MAST, Crit, Disc, etc.)

Seeing as this team has the power to get people kicked from the wiki for use of AI, this would have to be a Disc subteam.

2. How should this team be structured? Is a simple Captain/Vice Captain structure sufficient?

Yes.

3. How should this team formalize reporting of AI generated content, and what expectations should be held regarding investigation and compiling evidence?

Echoing Zoobeeny.

4. How should final reports be made? Should they be made directly to Disc, or simply an Admin so the user may be revoked/banned?

Reports should be made to Disc.

5. How should finished cases be logged?

O5, just like everything else.

page 1 of 212next »