Blocking AI crawlers on the fediverse (fedia.io)

submitted 8 months ago by cecep@fedia.io to c/fediverse@lemmy.world

33 comments fedilink hide all child comments

Given how Reddit now makes money by selling its data to AI companies, I was wondering how the situation is for the fediverse. Typically you can block AI crawlers using robot.txt (Verge reported about it recently: https://www.theverge.com/24067997/robots-txt-ai-text-file-web-crawlers-spiders). But this only works per domain/server, and the fediverse is about many different servers interacting with each other.

So if my kbin/lemmy or Mastodon server blocks OpenAI's crawler via robot.txt, what does that even mean when people on other servers that don't block this crawler are boosting me on Mastodon, or if I reply to their posts. I suspect unless all the servers I interact with block the same AI crawlers, I cannot prevent my posts from being used as AI training data?

top 33 comments

sorted by: hot top controversial new old

[-] FaceDeer@kbin.social 63 points 8 months ago* (last edited 8 months ago)

We're sick of closed walled-garden monoliths like Reddit! Let's move to an open federated protocol where anyone can participate and the APIs can't be locked down!

...wait, not like that!

Yeah. This is what you signed up for when you joined the Fediverse, the ActivityPub protocol broadcasts your content to any other servers that ask for it. And just generally, that's how the Internet works. You're putting up a public billboard and expecting to be able to control who gets to look at it. That's not going to work. Even robots.txt is just a gentleman's agreement, it's not enforceable.

If you really want to prevent AI from training on your content with any degree of certainty you're probably looking for a private forum of some kind that's run by someone you trust.

[-] cecep@fedia.io 24 points 8 months ago

I don't expect anything, I was merely asking a question to clarify this

[-] FaceDeer@kbin.social 25 points 8 months ago

Well, I hope my answer clarifies it. You can't prevent LLMs from being trained on your public posts.

[-] pop@lemmy.ml 0 points 8 months ago

We're sick of closed walled-garden monoliths like Reddit! Let's move to an open federated protocol where anyone can participate and the APIs can't be locked down!

Can you point to where the fediverse collectively said that? Speak for yourself and don't act like fediverse was designed to suit your definition of freedom. The fediverse is open and federated as in, there are multiple instances and owners without a centralized administration and the owners who hosts those instances decide what to lock down.

[-] FaceDeer@kbin.social 3 points 8 months ago

And some of those hosts can decide to serve up their content to AI trainers. Some of those hosts can be run by AI trainers, specifically to gather data for training. If one was to try to prevent that then one would be attacking the open nature of the fediverse.

There have been many people raging about their content being used to train AIs without permission or compensation. I'm speaking to those people, not the "fediverse collectively". As you suggest, the fediverse can't say anything collectively.

[-] CameronDev@programming.dev 36 points 8 months ago

But robots.txt is not a legal document — and 30 years after its creation, it still relies on the good will of all parties involved

You can ask nicely, they can (and will) ignore it.

[-] sukhmel@programming.dev 13 points 8 months ago

Also, I've already seen complaints about AI companies scraping everything ignoring robots.txt

And we would block the obedient and useful crawlers while doing no harm to malicious

[-] Tomato666@lemmy.sdf.org 22 points 8 months ago

Surely the AI crawler company can set up their own node. They post nothing but collect everything going forward from the time they go live?

[-] cecep@fedia.io 11 points 8 months ago

After reading your comment I was disappointed openai.social doesn't exist

[-] andyburke@fedia.io 9 points 8 months ago

They don't want AI to hate itself, so they don't want our training data, thankfully.

[-] mozz@mbin.grits.dev 16 points 8 months ago

You are correct. Some of the largest instances block bot traffic, but most don't, meaning your posts have been seen by AI crawlers and will continue to be so.

Short of not participating in federation and only discussing things within a private non-federated community on a personal instance or something, I don't think there's a way to prevent it.

[-] cecep@fedia.io 5 points 8 months ago

Thanks for confirming. It's unfortunate that people who are outraged about Reddit selling their data to AI companies don't really have an alternative in the fediverse.

I guess the best hope is for new mechanisms to control AI crawlers to emerge, so they can be blocked per user rather than per domain. Maybe https://spawning.ai will come up with something. One can hope.

[-] FaceDeer@kbin.social 12 points 8 months ago

I really don't see how it would be physically possible to do that and still allow the content to be publicly seen by other humans.

[-] Cheradenine@sh.itjust.works 9 points 8 months ago

It is unfortunate, buy we are giving our data freely, as we did on Spezzit. IMHO it would be great to block efforts to monetize Lemmy by ai, but that is not what we signed up for.

Lemmy is neither private, nor closed. It's just the way it works.

Contributing in an open forum means the data will get harvested. If it closed there will be fewer views, open is what we have now.

Companies will train on what we post, we are not giving that (directly) to a centralized service though. To me that compromise is enough.

[-] TheOneCurly@lemm.ee 7 points 8 months ago

I don't see it as hypocritical at all. Public comments are, for me at least, put out for the public good. The same reason someone might license open source code with the MIT license. My issue with Reddit is that they restricted who can obtain the data and then privately sold them to only the highest bidder. They should be freely available to all who want to view them without restrictions on money or power.

[-] originalucifer@moist.catsweat.com 3 points 8 months ago

it really sounds like you really want a walled garden so you can control your.. .whatever. the fediverse is public by nature, so discussing how you can control public information is kinda.. weird.

[-] cecep@fedia.io 3 points 8 months ago

Is it? Reddit is technically "public" too in the sense that you can view all the content without an account, yet Google and others pay for the data anyway. And for many years, people made stuff public and could reasonably expect it won't show up in any major search engines because Google, MS and others respected robot.txt. I know it was never legally binding. I'm also not naive, I know I give up control when I post publicly and there won't ever be a perfect solution to the AI crawler situation. But a lot is changing right now, regulatory and technologically.

[-] FigMcLargeHuge@sh.itjust.works 4 points 8 months ago

I think you are mistaking publicly available with public. Just because reddit made everyone's posts publicly available doesn't mean they are public. Once you post something, they have the right to use that data in any way they choose, and you agreed to that when you signed up. Per their user agreement:

"You retain any ownership rights you have in Your Content, but you grant Reddit the following license to use that Content:

When Your Content is created with or submitted to the Services, you grant us a worldwide, royalty-free, perpetual, irrevocable, non-exclusive, transferable, and sublicensable license to use, copy, modify, adapt, prepare derivative works of, distribute, store, perform, and display Your Content and any name, username, voice, or likeness provided in connection with Your Content in all media formats and channels now known or later developed anywhere in the world. This license includes the right for us to make Your Content available for syndication, broadcast, distribution, or publication by other companies, organizations, or individuals who partner with Reddit. You also agree that we may remove metadata associated with Your Content, and you irrevocably waive any claims and assertions of moral rights or attribution with respect to Your Content."

Because they allow anyone to see the posts doesn't make it "public" data, it just means that they are allowing you access to the data they now have a license to. Now lets say you work for a state agency. Any work you do is property of said state and is public. I believe the same goes for some government agencies, like NASA. The work they produce is public. That's completely different than reddit allowing you to post on their platform and then allowing others to see your post. They can do whatever they want with the data, including turning it off one day and just sitting on it if they wanted. Expecting anything public from a private company, well good luck with that. Back to lemmy, well even if you blocked all AI from scraping from an instance, nothing would stop a company from just setting up their own instance, federating it, and just sucking up all the info as it comes in. Nothing you post on here will ever be private.

I think people are about to learn a hard lesson on the internet. Nothing is ever private if it is online.

[-] originalucifer@moist.catsweat.com 4 points 8 months ago

the fact that google has to pay for the data proves the walled garden you claim is public.

the fediverse is public, by default. it publicly distributes information to other publicly accessible servers.. by default.

its public information on publicly accessible servers that are opt-out by default. publicly.

im baffled how people can have some expectation of privacy in such a clearly defined public space.

[-] cecep@fedia.io 3 points 8 months ago

You don't need to explain to me how the Fediverse works and I never said I have any expectation of privacy. But generally speaking, you're overlooking the fact that there always have been rules for what can, and cannot be done with information that is publicly available. Just because someone publicly posts his Facebook profile picture doesn't mean it's legal to use in an ad without permission, for example. People might break the rules, yes, but then they might face consequences, and that alone prevents many from breaking them in the first place. Not perfect, but better than nothing. And I'm saying we're in a process where rules are being renegotiated when it comes to using public information for AI training

[-] originalucifer@moist.catsweat.com 2 points 8 months ago

fair points, but i still posit that its a waste of time to attempt to regulate what can be viewed anonymously.

personally, i could not possibly care less about any of my data being ingested by 'ai'. not a battle i care to fight, or even find worthy of fighting.

[-] cecep@fedia.io 4 points 8 months ago

That's fair, but I think if AI companies would be legally required to disclose the sources of their training data and if you make some successor to robot.txt legally binding as well (both is being discussed in the EU for example), at least the "bigger players" in the AI industry would respect the rules. Better than nothing

[-] ptz@dubvee.org 5 points 8 months ago

Mine is but a wee instance, but our bot blocklist is large. For the ones that slip through, once identified as bot traffic, the firewalls go up in their direction.

[-] will_a113@lemmy.ml 6 points 8 months ago

I wonder if content should carry some license automatically. Like if you agree to the TOS of an instance, your comments are automatically all licensed as CC:BY or CC:O or the more restrictive license of choice of the instance owner.

[-] hollyberries@programming.dev 11 points 8 months ago

There's someone running around lemmy with a creative commons sharealike link as a signature. Quite funny to be honest. I can't remember the username though. They're bound to show up sooner or later :)

[-] Rentlar@lemmy.ca 7 points 8 months ago

Oh yeah it was @onlinepersona@programming.dev

You go champ! If an AI starts ending their posts with a CC BY-NC-SA license I know who to credit!

[-] onlinepersona@programming.dev 12 points 8 months ago

You're welcome

CC BY-NC-SA 4.0

[-] ArbitraryValue@sh.itjust.works 7 points 8 months ago

I don't think that would make much of a difference. Training AI on copyright-protected data appears to be fair use.

[-] FaceDeer@kbin.social 2 points 8 months ago

Yup. There are dumps of Reddit's entire archive of comments and posts available via torrent, I suspect the only reason Reddit's getting paid for that stuff right now is that it's a legal ass-covering that's comparatively cheap. Anyone who's a little daring could use it to train an LLM and if they prep the data well enough it'd be hard to even notice.

[-] ptz@dubvee.org 5 points 8 months ago* (last edited 8 months ago)

Really, there's only one way to prevent that, but it would offer no guarantees; the instance with the weakest security in the group would allow your posts to be crawled.

It would require an agreement among instances to block crawler bot traffic (by user-agent, known IPs, etc) and only federating, via allow lists, with instances that adhere to the agreement. At that point, it's more of a federated private forum, but there would still be some benefit I guess.

[-] vamp07@lemm.ee 4 points 8 months ago

I don’t object to my content being used for training. I do object to Reddit profiting from that data. It’s the reason I basically don’t participate on Reddit anymore. Anything I post in the fediverse I am aware I am offering it up for free to be crawled and used as seen fit as long as it is not monetized without my consent. I don’t consider model training to be monetization.

[-] cecep@fedia.io 1 points 8 months ago* (last edited 8 months ago)

Fair reason for not participating in Reddit. I would argue though that while model training is not monetization per se, with this "AI as a platform" rationale promoted by OpenAI, Google and others, there is very direct link between model training and monetization. Monetization without your consent - especially when these companies refuse to reveal the sources of their training data. Wouldn't be surprised if GPT-4 or Gemini have been trained on your Fediverse posts, or will be in the near future

[-] vamp07@lemm.ee 1 points 8 months ago

Agreed but it bugs me that I need to pay Reddit to not see ads and on top of that they get paid for the content we produce. The fediverse is a better model.