AI Powered Mod Tools for Nostr Communities on Satellite

Hey everyone, I want to tell you about something I've been working on β€” I wasn't planning to post about this today because it's not ready quite yet, but I noticed that nostr:npub1l2vyh47mk2p0qlsku7hg0vn29faehy9hy34ygaclpn66ukqp3afqutajft had created a new nostr community n/AI this morning β€” and it just fits too perfectly.

So here goes: I'm working on a feature that will allow community founders to "spawn" an AI moderator to help detect and defend against spam, organize their community, and eventually do a bunch of other things.

I believe this capability may become an essential tool for nostr to survive the coming onslaught of AI-powered spam. It seems to me that if we really want to have a realistic chance of maintaining public, human-led social spaces, we have to match AI with AI. It's not something we can afford to leave on the table.

On a technical level, nostr is a perfect fit. The shared, permissionless, data architecture means that each community's AI can "live" on Satellite's server (or any server, actually) and, after having been nominated as a moderator, proactively sign events with its own pubkey just like a human.

The AI will be given a simple task: *Enforce this community's rules as written*.

So the human admin of the community can essentially set policy to be implemented by the AI. For example, if the AI decides to remove a post, it can be instructed to explain why it removed the post, and to specifically cite the "legal basis" for its action. Human admins will be free to tweak the rule, overrule the AI's decisions case-by-case, or remove the AI from its position of authority.

It's very interesting to consider where this leads, and what other things an AI that "lives" in a community may be useful for. For now I'm focused on getting an MVP with this mod stuff deployed on Satellite asap. I'm genuinely curious about what it will take to make this work well in practice.

If you want to open your own thread on the AI/Nostr intersection, here's the link to the community https://satellite.earth/n/AI/npub1l2vyh47mk2p0qlsku7hg0vn29faehy9hy34ygaclpn66ukqp3afqutajft

I'll be writing a lot more about this.

Reply to this note

Please Login to reply.

Discussion

Super interesting 🧐

Can’t seem to reply on mobile yet, will check out desktop

I’m pushing an update today that fixes a bug on mobile

Cool, looking forward to see how well this works!

As am I!

Makes me think of GAN

What's GAN?

Generative adversarial networks?

Generative Adversarial Network

Interesting! Yeah it's starting to look likely that the future of social media will be AI vs AI. We need a governance layer that allows humans to have our own priorities.

super cool! lol couple days ago I just post this note nostr:note1nqujw8jnhrn47zskun9z3nnp72sknq6la50k4x0txccty7wpkjts9tlk42 I guess that's the solution now!

If we don't find a way to apply AI to maintain decentralized spaces where free speech is possible, we're just leaving it to be used by centralized entities to further corral and censor people.

yes that is the point!

Yeah I was agreeing with you!

Woah. I like these ideas

This sounds very interesting!

Very good use case for decentrilised AI algos amd a very interesting problem to solve. How do you hope to counter a centralized AI service like chatGPT which can geneate an infinite content noise?

I am asking this as I expect that no community would have resources to run anything close to an LLM, which could counteract other LLM generated content.

We are kind of in a david vs goliat situation. Maybe a way to think about this is to imagine bunch of davids fighting a goliat: simpler algos for individual communities colaborating to filter out content from large scale AI. How do we achive that?

Don't you think a WoT is a much more scalable approach to this type of attack?

World of Tanks? Just kidding I do not know what WoT is πŸ˜… I am assuming here that AI content generators would depend on some platform like ChatGPT or whatever, as they should be able to provide the most realistic sounding-looking content.

web of trust

Thanks πŸ˜… I do not know enogh about to give an answer. Any sources on how WoT is implemented?