AI-hunting on the boards

[SoManyTeeth]

I know we're all aware that AI-generated text (and other content) has started appearing all over the place, and understandably people on MN (as elsewhere) are concerned that they're having their time wasted by AI-written posts masquerading as genuine human contributions. But IMO, while AI-hunting is an understandable response, it's becoming a more widespread problem on MN. I feel it might be worth thinking about explicitly including it in the guidelines on troll-hunting in general.

Examples of AI-hunting on this thread: https://www.mumsnet.com/talk/am_i_being_unreasonable/5272283-fleecing-tourists-in-summer-with-high-prices — this is the one that prompted me to post this afternoon, but it's something I see a lot of now.

I understand the urge for posters to demonstrate that they haven't been "taken in", or to warn others against investing in or wasting time on a potentially bad-faith poster. But the same would apply to other types of trolling and bad-faith posting, where posters are advised to simply report rather than engage in troll-hunting. AI posts are presumably an attempt to drive engagement, like much other bad-faith posting. Troll-hunting and AI-hunting can increase engagement with bad-faith posts and threads, disrupt and derail genuine threads, and hurt those falsely accused of posting in bad faith.

There's also a disability aspect that I feel needs to be taken into account here. It's started to be recognised that texts written by autistic people are more likely to be misidentified as AI-generated. Autistic people are being routinely accused and discredited because of natural features of their writing style that they can't help. It's unfortunate that this also plays into old stereotypes and prejudices about autistic people being robotic, computer-like, or otherwise less than fully human, because it means that the false accusations can hurt more when they happen. I'm autistic, and my posts on MN have been dismissed and mocked as AI on MN more than once.

Is this something that could be considered as an amendment to the Talk Guidelines subsection on troll-hunting?

I also don't see the point of reporting a harmless post or thread just because it seems like AI. I can't see what it solves if it adds something to the thread even if it's "basic". Perhaps because I can often relate to being either too "basic" myself or too serious depending on how I'm feeling (I don't choose it and can't switch it off/on, it just happens) and don't often notice any 'weirdness' with these posts because I can be (seen) that way too.

The point is, as I wrote earlier, because we don't know the motivations for AI generated posts.

If someone generates some information using AI which they think is a valid contribution then IMO posting it is ok if, and only if, it is explicitly stated that an AI was used (and ideally which one). It may not matter on a gardening thread if an AI hallucinates misinformation but it might be important on others.

MNHQ definitely needs to write some clear guidelines.

MN would be the first platform to regulate content in this way. Behind each "AI" post is a human currently; AI can't just go and post on MN on it's own. I don't see anything wrong with using AI to help draft a post as long as it's not spreading misinformation or used for trolling. Poorly written, incoherent or "wind-up" posts are much more likely to be written by humans than AI.

Yes, I agree. That's part of what I'm saying too. I don't see the harm in people using AI to construct their post. AI isn't a robot that independently writes and posts on mumsnet. I believe AI has helped some people to be able to write properly or construct writings in the way they'd prefer but couldn't do before. It doesn't matter the method or tool used to write, what matters is the content of the post. If it fits the thread and is harmless, there's no need to expect people to reference that AI was used to write it anymore than people should be expected to reference using spellchecker or grammarly or google translate, etc to write.

No scientific studies yet unfortunately, as far as I know (though I have seen a study which found that non-native speakers are more likely to be falsely flagged as AI by detection software). But it is something that people are talking about, both automated detection and accusations from other people.

A quick Google search with the terms autism accused AI gives me:

Bloomberg article, talking about automated detection:
https://www.bloomberg.com/news/features/2024-10-18/do-ai-detectors-work-students-face-false-cheating-accusations?
https://archive.ph/kU77Q
Olmsted submitted a written assignment in a required class—one of three reading summaries she had to do each week. Soon after, she received her grade: zero. When she approached her professor, Olmsted said she was told that an AI detection tool had determined her work was likely generated by artificial intelligence. In fact, the teacher said, her writing had been flagged at least once before.
[…]
Olmsted disputed the accusation to her teacher and a student coordinator, stressing that she has autism spectrum disorder and writes in a formulaic manner that might be mistakenly seen as AI-generated, according to emails viewed by Bloomberg Businessweek.
[…]
The students most susceptible to inaccurate accusations are likely those who write in a more generic manner, either because they’re neurodivergent like Olmsted, speak English as a second language (ESL) or simply learned to use more straightforward vocabulary and a mechanical style, according to students, academics and AI developers.

Guardian article, also mostly about automated detection, briefly mentions the autism link:
https://www.theguardian.com/technology/2024/dec/15/i-received-a-first-but-it-felt-tainted-and-undeserved-inside-the-university-ai-cheating-crisis
Neurodivergent students, as well as those who write using simpler language and syntax, appear to be disproportionately affected by these systems.

Several articles about an autistic American university researcher who tweeted about being misidentified as AI from the tone of their emails; here's the NY Post one:
https://nypost.com/2023/07/21/autistic-purdue-professor-accused-of-being-ai-for-lacking-warmth-in-email/

Many blog posts, message board posts, social media posts etc. from people who've been falsely accused either by automated detection software or other people — just a random selection:
https://andrewggibson.com/2024/10/27/false-ai-detection-human-writing/
https://www.reddit.com/r/autism/comments/1eemf3e/accused_of_using_ai_but_im_just_autistic/
https://www.threads.net/@fork.ca/post/DF6dFLvsctR
https://writerethink.medium.com/please-stop-accusing-people-of-not-being-human-403f15476837

None of this is conclusive scientific evidence, but I think it's enough for initial suspicions that something may be going on there.

My own anecdotal experience is of having been accused several times (by other people, not automated detection), and while I don't think most people have that experience, I've come across several other autistic people who've also said it's happened to them.

There are also autistic people who say they haven't had it happen to them, and presumably it does also happen to non-autistic people, but it seems there are features that are more common in autistic people's writing that get picked up on.

It's also possible,I suppose, that autistic people are no more likely to be accused than anyone else, but because of historical preconceptions about autism, quite reasonably wonder whether the perceptions of their writing as computer-like are linked to those stereotypes. They also might be more likely to find the accusations hurtful and to remember them.

Apologies if any of those links don't work — can't remember if archive links are allowed.

My personal way of handling it is - if I see a post that I think I can identify as AI, I just don't reply to it. If people don't engage, instead of calling troll, this is the best way of killing traffic to such posts.

Yeah I could do that. But that won't change anything, it'll just remove those particular posts. Threads will still get derailed, people will still be piled on by posters picking apart their posting style and talking about them like they're a machine, and AI-generated posts will still get loads of engagement from people accusing and analysing them. AI-hunting has a very similar impact to other troll-hunting, but unlike troll-hunting it isn't widely known among MN users to be against Talk Guidelines.

I can't remember whether it was in my OP or my first reply, but I mentioned that I didn't actually know whether MN considered AI-hunting to be against the rules or not. It's not explicitly mentioned in the guidelines, and I see lots of perfectly well-regarded posters doing it, so there was nothing for me to go on there. When nobody knows about a rule it doesn't get followed, and unless you've got very heavy moderation that rule might as well not exist. So the type of disruption that anti-troll-hunting rules were intended to prevent goes on happening in this guise, because nobody knows they're not supposed to and nobody knows it's reportable.

Behind each "AI" post is a human currently; AI can't just go and post on MN on its own.

Of course there's a human behind them. The human may of course be able to set up a 'bot' which posts AI generated content.

what matters is the content of the post

Yes... and currently AIs are good at creating authoritative sounding content that's riddled with inaccuracies. The human doing the posting needs to check for accuracy.

But the motivation matters too. We know 'bad actors' set up 'sleeper accounts' - ai content may be being used to create their 'legend'.

It's naive to assume AIs aren't being used nefariously.

I think the impact on autistic people is unfortunate, but I don’t think it’s reason enough to stop people pushing back against AI generated content taking over the internet.

I didn't say that it was.

And I didn't say that we shouldn't be trying to combat AI-generated content masquerading as genuine human interaction.

I'm saying that, for the same reasons MN has troll-hunting rules, it should have AI-hunting rules. Turns out it does, but nobody knows about them so, in a reactively moderated forum, they might as well not exist.

I don't like trolls, but while I can see that there are problems and flaws with the MN troll-hunting rules, I can also see that those rules are aimed at limiting the disruption trolls are able to cause and reducing general toxicity on the boards. Engaging with bad-faith posters, in any form, has never been a good way of combatting them.

Well-communicated rules against troll-hunting, combined with high-quality, responsive moderation, are part of the pushback. The second part, though — the high-quality moderation — is absolutely necessary, if posters are to trust that suspected bad-faith posts will be dealt with quickly and fairly, and not feel the need to make accusations on-thread. I do feel MN could do better on that front, especially WRT the lack of (UK) nighttime moderation.

The potential for AI-hunting to have a disproportionate impact on autistic people is an additional factor, not the main one. Even if that were proven not to be an issue, all the other issues around disruption, engagement, and unfair suspicion would still stand, just as they do for general troll-hunting.

For those who are objecting to my request for clarification of the rules on this:

Is your problem with the existence of the troll-hunting rules in general? If so, fine, and I'm sure there are plenty of arguments you can make for that position, but that's not really what this thread was intended to be about.

Or do you think that troll-hunting should still be banned, but AI-hunting should be allowed? If so, please could you explain why you think those rules are the right way to deal with suspected traditional trolls, but not the right way to deal with suspected fake posts where the text has been generated using a computer program?

I think that AI posts are a concern because as pp said, we don’t know the motivations of the poster. While I feel sorry for posters who are genuine and use AI for reasons like English is their second language, MN posts are often reposted on other media platforms so you could have journalists creating shocking content for fun or trolls who get a kick out of people engaging with their bullshit. It shocks me how many weird trolls there are and how their fetish isn’t the expected sort of topics.