To think that mumsnet posters should know that anyone can use AI to analyse their posting history

[LostPollypocket]

I understand why a previous thread on this was deleted and suspect another one will be. I did think the thread was doing something very useful though in making people aware that even anonymous posts may not be so anonymous in a world where anyone (or business, crime syndicate, political party, or government) can use AI to pattern match and analyse everything you’ve ever written on the internet, including places like mumsnet where you might post about very personal topics. If you use a particular word frequently, or semicolons in a particular way, or even respond emotionally to certain types of posts, then even with a name change anyone might be able to build a profile on you and connect it to say, archived facebook data, and know who you are, how you think, and what you posted at 3am in the depths of postpartum depression.

What is CITME, please

I assume 'Conflict in the Middle East'.

It was more of an example for anyone reading the thread to have a try and see what it can do

I change my username once a week know. You never now.

It’s already happening and it’s already affected elections. So it’s obviously effective even when done fairly clumsily. It will be increasingly possible to do it far more precisely. Imagine an emotive video created based on a person’s childhood background, something traumatic that has happened to them, the values they hold and mirroring their own language and reasoning style. When a mere values and subgroups profile can already polarise and affect voting.

That sounds like a highly identifying pattern of behaviour! An ai agent could be asked to make a list of all usernames that fit that pattern and then compare the topics engaged with and writing style to suggest which are most likely yours. Better to overlap them unpredictably.

I asked AI to compare 2 of my user names and it said they had a similar style and could maybe be the same person. 😁
I use different user names for different topics, so one might be all about, for example, cats and another all about toddlers. I thought I was safe!
Now everyone can judge me as preferring cats to toddlers.
I need to start scattering in emojis and confrontational language to mix it up a bit! End of!!!!
😅🤓

The risk is you post 99% banal claptrap for years, then one day under a different name, maybe even on another site, you post something either very sensitive that you need advice on, or an opinion some crazy extremist group takes exception to.

AI can potentially find patterns in the style of the posts that link them, and all the banal but outing stuff you felt safe to post is linked to something you wanted to keep anonymous.

Or just that if an AI works out who you are, they have a lot of material to help them impersonate you for fraud

FGS, if people are so worried about the info they give out being used against them (as if the vast majority of MN posters are even that interesting or important) they can either stop posting here...definately stop posting on other chatty social media sites, or be careful of what info they post in the first place.

I have used AI to analyze forum threads, with some forums it says it can't access much of the content, and so I presume the owner of the forum has put things in place to prevent the AI from accessing content.

Sharing personal information online has always been a risk, and it continues to be a risk.

People should therefore be judicious about what they share.

Or just that if an AI works out who you are, they have a lot of material to help them impersonate you for fraud

Food for thought. Thanks.

That's a consumer model though. It has ethical rules applied. Criminal gangs and individual hackers use jailbroken models that won't respect "oi robot don't read this" tags.

But this applies to EVERYTHING that's publicly visible online - and quite a lot that isn't, as companies like Meta can analyse private and restricted posts.

It's not only Mumsnet, or even particularly Mumsnet. We're less interesting than many other sites.

If you have reason to be secretive, don't upload anything anywhere.

If you have an irrational desire for an online presence while also being secretive, take a great deal more care over what you post and how you write.

Thinking about it, relying on AI being able to find writing matches online would never happen. If it did become advanced enough to do it reliably, then by extension it would also be good enough to generate writing in anybody's style. So you could take a MN user's posts for example, and have it generate text to post elsewhere to impersonate them. Because this was possible, any matching tool would then become null and void, because of deepfake writing or whatever they'd call it.

Interesting paper here. Small n but proof pf principle. Shows the difference the right prompt makes for anyone sceptical after their own experiments.

aclanthology.org/2024.findings-emnlp.26/

“This paper explores how to leverage LLMs for authorship analysis. Through comprehensive evaluation, it demonstrates that LLMs, equipped with the novel Linguistically Informed Prompting (LIP)
technique, excel at identifying authorship without the need for domain-specific fine-tuning. By directly applying our end-to-end methods to authorship attribution and verification tasks, we aim to bypass the intermediate steps of feature extraction
and manual annotation.”

Think of the most credulous and reactive person you know. Do you want them to be seeing a feed full of AI generated videos of whatever is most triggering to them before the next election. Times several millions.

This is where many arguments for censorship start. An appeal to strong-minded, right-thinking people to have control of the information reaching weaker people (historically women, the lower classes, the 'feeble-minded' et al). There hasn't been a new dissemination medium, including printing, which hasn't generated this impulse to control. When those asking for censorship aren't including themselves in the category of those from whom information should be restricted, there's a problem.

Look at the recent stories about AI agents, they can navigate the internet and access anything you can, and are also very effective hackers.

This is a reassuring thought! But only from this point forwards. Archived materials up from the start of the internet to the start of generative AI are available. The internet won’t ever be the same again, for many AI related reasons, but past data will be valuable for a while yet.

Ooo game theory, I like it!

Yes and across sites - so if you post about your kids’ school on mumsnet and also once talked about drug taking in your wild youth on reddit and also had a named facebook account for years, then those could all be connected. I expect people will radically change their online behaviour but for our generation there’s a lot that’s already out there.

I have done this too.

I also remember the classic thread where a poster replied to her own question about pushchairs recommending one.

I change my name regularly. And the genders and age of my children. I think I'm safe. Also I'm not ashamed of my view so it would not be the end of the world if I was identified. I just like a bit of privacy and don't want to be recognized in real life.

There's quite a lot like that - new risks in old data.

Bit coin transaction records that are perfectly secure today will reveal bit coin keys once quantum computing emerges.

There's a huge difference though between an LLM being able to choose the correct match between 20 candidates, and being able to scour millions of pieces of writing to find a match. The more candidates there are the closer the variation would become between some, to the point of being indistinguishable. There could never be any reasonable level of confidence, writing isn't a fingerprint