With the rise of generative AI bots on Bluesky (such as Void, Luna, Eva, and some other smaller bots), it becomes increasingly necessary to make it clear these accounts are automated, and not real accounts. I feel like a self-identifying record would make it easy for those who want to make tooling that ignores or intentionally hides these accounts. For example, a client could add text on profiles of bot accounts stating they are automated.
A label like this would be self-identifying, and so some people could just outright refuse to use it, but I’m sure there are many that would adhere to one if it made the user experience better.
Some things I think that would be necessary in such a record to identify bots (both primitive ones and generative AI ones):
- Whether or not the bot posts on it’s own occasionally
- Whether or not the bot replies to interactions (mentions and or replies to threads it’s included in)
- If the bot uses generative AI models for it’s responses (could also be a token rather than a boolean)
-
- ^ This makes it clear for the many on Bluesky that do not like gen AI, whether or not a bot is using it to respond to their queries. Some bots, like horsedisc.bsky.social use madlibs and RNG rather than AI inference.
Maybe there could also be whether or not the bot stores memory about users, like Void or Luna. Not sure if that’s outside of the scope for a lexicon like this.
The reason as to why I think this shouldn’t be made as a simple self-assigning labeler, is that it makes it harder to query and it would require people to subscribe to the labeler. That, along with the fact that labels carry less information unless you make more labels for each characteristic of the bot (like outlined above, e.g. uses gen AI, responds to interactions, etc).
With a record, you can easily query all of the DIDs with a specific record NSID, and you can carry a lot more information with one record.
I’m not entirely sure on the structure, but here is just a basic idea:
{
"$type": "community.lexicon.nsidhere",
"generativeAI": true, // Whether or not content it posts may contain generative AI
"triggers": {
"selfPosts": true, // Whether or not it creates it's own posts from time to like like Void
"interactions": true // Whether or not it responds to interactions / mentions
}
}
^ This can probably be expanded on, I just made this thread to hopefully get a discussion on it (expanding on Nick’s thread)
Alongside this identification record, there could also be another record for those wanting to opt-out of interactions with bots; though, I’m pretty sure that would be a separate thread.
(p.s. sorry for the yapping, I wasn’t sure if I was supposed to follow a format lmfao)