This is a subfile of the primary AI4Communities post, and won't make any sense unless you read its parent first. It explores how the AI4communities idea would look in decentralised social networks powered by ATProto, the Bluesky protocol.
(Notes: This is an early draft. As explained in this newsletter edition, I am publishing these early versions as I develop my thoughts in the hope that constructive comments will help me finish the post. More version control in the footer.)
I knew Bluesky existed but not much else when I published my manifesto posts of 1/1/23, so early versions of the AI4Communities post was Fediverse-oriented. I created this post in early November 2024 to explore how AI4Communities would look in an ATProto-powered world.
By then I had annotated 11 resources tagged Bluesky (there'll probably be more by the time you read this) which I then synthesised for my November 2024 newsletter (which contains a lot of content I need to summarise and add here, so check it out). By then I had concluded that in many ways Bluesky already supports AI4communities, but not in the Fediverse-friendly "cozyweb village" paradigm I had been using until then.
After all, individual Bluesky users can already subscribe to custom feeds (ie Content discovery services) and moderation tools (ie Content moderation services, known as labellers in Bluesky) provided by 3rd parties. These can range from human-powered labelling through to AI-powered, although most currently are custom feeds driven by pretty simple search expressions.
The first custom feed I installed, for example, was “Science”, developed by Dani cRabaiotti, a zoologist and data scientist working at the Zoological Society in London. This is possible because Bluesky wants to give users Algorithmic choice. Instead of suffering the problems created by a one-size-fits-all algorithm, as seen on platforms like X and Facebook, Bluesky users can choose from an open and diverse "marketplace of algorithms". In this approach, algorithms "act as aggregator services, similar to search engines" - users add algorithms to their clients seamlessly, with Bluesky allowing them "to swipe between favorite algorithms or view a multi-algorithm feed".
Bluesky doesn't even control feed creation. Instead, they provide "APIs for feed generation... allowing for custom feed and moderation systems to be created as independent services ... [and] a feed selection system that enables users to explore third-party feeds and access them as effortlessly as their home timeline... we’re using a similar approach to address reputation, misinformation labeling, and moderation."
To test this I created some Bluesky feeds for the Brussels Bubble, to see whether that will help them off their nasty Xitter habit, and added them to the Bluesky Brussels Bubble Starter Pack. I tested two 3rd party services, neither are AI-driven - results in my November newsletter.
Although I'm pretty happy with Bluesky's direction of travel, concerns remain - for example:
While it's already easy to imagine how a community could form around their own AI-powered labellers and custom feeds, these would be very porous communities - there is currently only one ATProto network, not a connected archipelago of servers, and it's very open: "Anyone who knows how to code can write an app or tool that can read practically any data about anyone, without having to ask anyone for permission" (A complete guide to Bluesky).
Is this a feature or a bug? Bluesky's openness, and the fact that the entire network can be searched, makes AI4communities more feasible, but it's a light version - there will be no separate villages, or private groups for collaboration.
My metaphor is therefore upended: we won't see a community collectively choosing some algorithms, sharing their data with them, and monetising the result, as in AI4Communities on the Fediverse. Instead, each member will go to the store individually, and anyone who subscribes to the same algorithm will automatically join the club.
One possibility of making such a community less open: the store puts the algorithm out of reach of non-members, which would mean that the store becomes the community's gatehouse.
to be done
This is one of this wiki's pages managed with the permanent versions pattern described in Two wiki authors and a blogger walk into a bar…