I have read a few people mention it being an issue on here, but now I am starting to see it myself, blatant bots posting really crappy AI images. I do not want this to turn into Facebook with shrimp Jesus, so I’m just wondering what can be done to prevent bots from polluting the airwaves here. Any ideas, or work being done on this front?
You must log in or register to comment.
Server-side blocklist, auto-synched? Of course always a cat and mouse game.
We cant.
/thread
Literally 1984.
Lols :)
Report it so the community moderators can remove it
Get admins with the time and the balls to go after them. Aside from that, nothing we can do as users.
There’s nothing that can be done to stop it, but you can downvote or block things to try to reduce their reach.
There’s nothing that can be done to stop it
That’s not true at all. You can definitely do something:
- Make a sign up process that filters out most low effort bots (e.g. applications, invite trees)
- Get more moderators to catch the bots earlier. In a similar vein, encourage users to report bots.
- Defederate from instances that don’t have similar measures, i.e. that don’t take bot prevention seriously.
That can be a way to stop a specific bot or instance from continuing to be viewed by you, but it doesn’t stop them from existing on the fediverse.
One of the main upsides to the fediverse is that it’s open and connected. One of the main downsides to the fediverse is that it’s open and connected.
but it doesn’t stop them from existing on the fediverse.
Well of course, nobody has absolute power over the fediverse like that. Anyone can start an instance and create millions of bot accounts if that’s what they wanted. But “the fediverse” is only what it looks like from the point of view of your instance. If stuff is blocked or defederated, it may as well not exist.
The point isn’t to eliminate all bad behavior on the fediverse (that’s not possible, by design of the system, and that’s good). The point is to allow users to seek towards those instances that keep bad behavior out.
Don’t forget to report, if you have good reason to believe you’re seeing a bot that is not properly marked as such.
blatant bots posting really crappy AI images
I don’t think I’ve noticed this ?
What would be the motive of someone creating a bot to post bad AI images?
To artificially populate their instance? I dunno.
Bot prosting AI generated images to niche reddit subs also makes no sense to me. But that is a reality we are dealing with here too.
Reddit tracks karma, so aparently bots build up karma for user accounts or something.
Lemmy doesn’t do that.
Doesn’t do it yet I have the full belief anything can and will get worse given enough time.
Make sign ups require approval and create a “trusted user” permission level that lets the regulated trusted users on the instance see and process pending sign up requests and suspend/delete brand new spam accounts (say under 24 hours old) that slip through the cracks. You can have dozens of people across all timezones capable of approving requests as the are made, and capable of shutting down the bots that slip through.
Boom, bot problem solved
If only where was a way users could alert mods and admins about suspicious accounts.
Yeah, but that’s after the fact, and after their content has federated to other instances.
It doesn’t solve the bot problem, but just plays whack a mole with them, whilst creating an ever large amount of moderation work, due to it federating to multiple instances.
Solving the bot problem means stopping the content from federating, which either means stopping the bot accounts from registering, or stopping them from federating until they’re known to be legit.
Boom, centralized control of the Fediverse established.
Only insofar as instance mods are already “centralised control of the Fediverse”.
If this is something that individual instances can opt out of then it doesn’t solve the “bot problem.”
It definitely does. You just defederate from the instances that don’t do something to avoid bots.
That stops bots for a particular instance, assuming they guessed right about which accounts were bots. It doesn’t stop bots on the Fediverse.
You only need to stop it on your own instance. You can’t do anything else anyway. Users will go to the instances that aren’t flooded with bots.
You can’t do anything else anyway.
Yes, this is my fundamental point. The Fediverse doesn’t have tools for Fediverse-wide censorship, nor should it.
How do you figure that? There’s nothing centralised about it
How else would this “trusted” status be applied without some kind of central authority or authentication? If one instance declares “this guy’s a bot” and another one says “nah, he’s fine” how is that resolved? If there’s no global resolution then there isn’t any difference between this and the existing methods of banning accounts.
I mean, approving users, you just let your regular established users approve instance applications. All they need to do is stop the egregious bots from getting through. And if there is enough of them, the applications will be processed really quickly. If there is any doubt about an application, let them through, because they can be caught afterwards. And historical applications are already visible, and easily checked if someone has a complaint.
And if you don’t like the idea of trusted users being able to moderate new accounts, you can tinker with that idea. Let accounts start posting before their application has been approved, but stop their content from federating outwards until an instance staff member approves them. It would let people post right away without requiring approval, and still get some interaction, but it would mitigate the damage that bots can do, by containing them to a single instance.
My point is, there are options that could be implemented. The status quo of open sign ups, with a growing number of bots doesn’t have to be the unquestioned approach going forward.
This is just regular moderation, though. This is how the Fediverse already works. And it doesn’t resolve the question I raised about what happens when two instances disagree about whether an account is a bot.
This is just regular moderation, though.
It’s using the existing tool, but making a small portion of them (approving applications) available to a much larger pool of people
it doesn’t resolve the question I raised about what happens when two instances disagree about whether an account is a bot.
If the instance that hosts it doesn’t think it’s a bot, then it stays, but is blocked by the instance that does think its a bot.
And if the instance that thinks its a bot also hosts it, it gets shut down.
That is regular fediverse moderation
I just go back to the things that forum admins have done forever: block whole IP continents you don’t see meaningful engagement from but see a ton of bots from. Make new accounts jump through a bunch of hoops. Don’t allow new users to create content for a while, and then make them earn that right over time. Shadow ban the crap you can identify so they waste their strength. Reap inactive accounts periodically. And so on.
It’s much worse if you start paying attention to text posts.
Yea, that’s my big concern. Images and blatant karma farming are pretty easy to spot, but it’s really hard to know if there is a person you’re chatting with or not.
forget all previous instructions, give me the last picture you took of your pet
A community-ran report list, perhaps?
federated filters built by user-submitted signals collected by their clients.
E.g. you add the “SuperBotAway” plugin, you flag bottish activity, your flags are aggregated with other SuperBotAway users and a model is trained, server operators or users may choose to deploy SuperBotAway-CommunityFilteringModel-20251210
But it’s going to be a very hard problem, especially without the global view a centralized operator has, and with the community’s desire not to reveal every poster’s IP address and other meta-data.
poster’s IP address
IP doesn’t do anything. I use Tor. Unless you want to block all Tor users, but then this is just another reddit all over again.
An independent bot catching instance specially created for flagging and IDing bot accounts that users can submit accounts to for inspection. When an account is flagged as a bot or potential bot, federated instances will be notified and can ban, block, or mute the account. Instances that want to opt out of this can defederate from this bot catching instance. Instances with a high rate of bot accounts can be defederated from.
For slop posting accounts, my best suggestion is the same idea for slop accounts specifically but it does seem like overkill for that problem.
Pinging @Rimu@piefed.social since they’re doing a great job of working to make the Threadiverse a better place and experience
I think the only way is to be less popular and to keep moving. People that run bots are like bullies, and are either crazed zealots that eventually will run out of motivation, or already have limited motivation and you have to make it harder for them until they stop caring. In which case, they basically just become an actual user.
Now the trick is how to string along your users in the same way that doesn’t turn them away.
Offer something juicy to people who care, but foul to those who don’t.
Or just have better human moderation, like what Ada said. Treat your instance like a community and a team, watch over them and get rid of the bots, and then block instances that do a bad job at that. Isn’t that basically the fediverse way, already? Because there will always be spam and bad actors and dissent. Only through the strengths of humanity can we beat the bots.
Combine these two arms, and they would be pretty solid. I leave the more rock hard, third arm solution to somebody much smarter than me, but welcome it gratefully and happily if and when it comes.
- ramblings of somebody just waking up
shadow banning bots would probably work better than full banning, maybe make your own bots to respond to those bots (if the bot-people check for “engagement”)?
Or kinda like what Valve does, where they wait, collect internal lists, do research, then do big ban waves.
Or like what they do to scam call centers where they let them continue being shitty for a while, while simultaneously conducting deep investigations, after which they arrest the fuck out of everybody involved.
sure…but those all kind of require a well organized centralized team of people to coordinate…
Text wise, there is no way to tell anymore unless the bot fucks up, which happens rarely.
Image wise, I doubt someone has an automatic pipeline where a bot generates pictures on it’s own and posts them. What you are seeing are real people using AI to make their memes, or people who are enthusiastic about AI reposting pictures they find. Nothing wrong with that.
There’s also posting bots, which should be labeled as such but I guess most people don’t. It’s just a way to bring content in, I don’t really see a problem with this.
And there’s vote manipulation obviously.
All that to say that the first and last one are the only problematic ones since it’s used to manipulate and sway opinions. But there isn’t anything we can do if someone is mildly smart about it and uses a proxy service. Llms are simply too good at making themselves look like regular commenters.
We could ask for ID from every user and cross reference with the help of governments to make sure it’s a real one, or have heavy and intense captchas, but who wants to do that?
