I know that in some quarters of both the Bluesky and Fediverse side of the #OpenSocialWeb there is a justifiably allergic reaction to AI. I get it and share it. But I'm becoming more convinced than ever at this: if we in the open social web space don't build our own healthy, safe and fair versions of #AgenticWeb interactions here, it will be defined and done for and to us. The big platforms are already building agentic features: social enabled agents that read your feed, draft your… 🧵 1/3
Replies:
38
Boosts:
6
@tchambers I'm not absolutely against AI -- I've gotten real benefit personally from AI tools.
That said, the kind of agentic AI in social web I think you're talking about is for brands to build a presence & for influencers to extend their reach. The problem of having that "done to us" seems bigger in something like BS than in the fedi. People are already pretty negative towards anything hinting of brand building or marketing. And I think the fedi would even more solidly reject AI brand building
@tchambers This feels a bit like a solution in search of a problem post. Which is not to say that the fedi doesn't have problems. But I'm not clear how an agentic model of the fediverse improves what's already here, and the potential downsides are fairly obvious. What sorts of things do you envision folks building with these tools?
@prism Good questions for a future blog post …
@tchambers I don't feel like this is the right framing at all.
We counter Big AI's so-called "agentic web" by boycotting it entirely. We don't need a healthy, safe, and fair version of the Torment Nexus. We need it to not exist in the first place. It's an entirely flawed, cynical vision of UX which only serves wealth & power aggregators at the very top.
@jaredwhite @tchambers It also seems to presume that, if a 'good' implementation existed, that would somehow satisfy or resist the 'bad' version; despite the repeated and blatant demonstrations of how many people see the unpleasant aspects of this stuff as the most compelling feature.
Some compromises you don't make, full stop; but you definitely don't compromise without any reason to believe that it will do something beyond hasten the outcome you don't want.
@jaredwhite @tchambers You are well within your rights to boycott agentic systems, but the fact is they are being built on the web — the W3C is active in this area and Tim Berners-Lee mentions his vision for agents in his recent memoir (and is focused on it in his current work). It’s ok that you and others don’t want a bar of it Jared, but don’t ascribe bad intentions to those of us who are interested in this development.
@ricmac @jaredwhite @tchambers
Uh, no, Dick, I think I will ascribe every bad intention to you and people like you that push inevitability narratives around inherently exploitive products and methods of computation.
You are not absolved simply because you think your intentions are good.
I’m baffled what any of that means. What does “read my feed” mean? Why would I want it to draft replies in a format that allows only 500 characters?
I’m legit trying to understand a single feature that would be even theoretically useful.
@ThreeSigma @tchambers A lot of folks use agents to summarize things for them, so that’s one use-case for “read my feed.”
ActivityPub doesn’t have a 500 character limit. That’s a Mastodon limit. The mindset for drafting a post like that is to state casually what you want to say and have it reword it for you. From a business standpoint, it would draft it for brand voice, etc.
@tchambers @ramsey
Okay, but don’t pretend Mastodon isn’t the predominant ActivityPub application. And very few people use AP for business, even assuming that having an LLM draft for you is a “good” thing.
How is this any different from using a chatbot, other than changing windows? Why not let it be a client-level thing?
We also don’t have a blockchain. Should we be getting that too?
@ThreeSigma It’s not any different from a chatbot. I can’t speak for @tchambers, but I didn’t read his posts as meaning we needed to embed agentic experiences in Mastodon itself. I think it was more a call to lead in designing what those experiences look like at the network level, which is something we can do with or without Mastodon.
FYI, I only dabble in #Nostr so not sure what communities there think of the idea of the #agenticweb - I'll watch rectiions to this post there: https://njump.me/nevent1qqs8lxfk6vheah745fhjwrwkvvr6a78fktycn4cfhj2432z5pfm7hecpp4mhxue69uhkummn9ekx7mqzypg4tlu7du03q97v5qzyq9l3mvup8f55p4yec549e0ssrxqcpjnf52jvv47
They are not building tools. They sre building surveillance and extraction systems, squandering planet scale resources to do so fast enough before they can be stopped.
The njump arg normalizes this to mere large scale web activity and as Dana says, presumes there is value there.
The near universal revulsion at the l.l.m scammers is not at the technology orcthe ideas; its against the structured exploitation and ruination designed into its deployment by corporations that hsve abundantly proven to us over snd over *that the (mostly) men running these schemes are dangerous people*.
(Lol if a i or l l m appear as words in the text my filters will block it. )
@tomjennings I share every single one of your concerns. At least as far as i can read them, come across a bit jumbled text in bits.
Its telling and interesting they've not produced any pocket sized tools we could use at home, on our own data sets. Even just inverted word indexes or whatever for search. Not that I'd necessarily trust them. But they've not contributed one thing at all, beyond letting people use chat, which is of as much or more benefit to them than any user.
@tomjennings there actaully are efforts at "small AI" focused on local user only-datasets - but you are right that work is not being done by the big tech giants.
The final sentence? Spaces to render my filter's eords illegible. Otherwise?
@tchambers my best guess as to the reaction here:
@tess We'll see, so far constructive.. and I tried to open with that I share the justifiable allergic reactions to AI. Those aren't crazy talk. But lots of work to do and if WE don't do it well and right and in open and private ways, it will be done in the oppositie ways and that ship will have sailed.
@tchambers *waves at ship*
@tchambers counterpoint: if we want a space for humans by humans, then perhaps it's fine to let that ship sail.
Or to put it a different way, *what exactly is on the ship we're missing*? What is the value being added? What is the technology being used for?
If you say, "there's work to do and if we don't do it someone else will [with AI]" - what is that work? Why is it valuable? What does it bring to the open web?
@tchambers @tess Let it sail! Bon voyage! Bye!
…replies, manage your presence. Those are shippping soon. Sooner than you think. And the open social web protocols are well, open by definition to these. They'll do it in ways that serve engagement metrics, not people. If the open social web cedes this ground, the definition of what "AI agent + social" means will be written by the same companies we built alternatives to escape. We have something they don't: open protocols, user-owned data, and communities that actually debate what's… 🧵 2/3
@tchambers what is the value of a so-called social web if it is going to be populated with agents?
@django Agreee: but if we don't build in some sense agentic web functions (and defense) that is humane and open and fair, it will be done TO us in way that aren't.
@tchambers how would it be done to us, and by whom?
I just don’t see Mastodon trying to turn the fedi into Teams?
@tchambers I have no interest in reading what your assistant (human or AI) has written for you. If other people want to have their agent read what someone else’s agent wrote … good luck to them, but count me out.
@__d I agree: enabling "robot butler inhuman posts" is not what I'm talking about here. I'm talking about how to have some sense of defense against that, but also enabling human posts better.
@tchambers Having an AI write your posts for you is appalling. We're supposed to be talking to each other, not firing slop at each other? (Also issues of where the training data came from, how much energy it uses etc)
…acceptable. That's exactly the right foundation for building agentic tools with consent, transparency, and human review at the center. The question isn't whether agents are coming to social; it's whether we shape them or they shape us. That ends my Ted Talk. 🧵 3/3
@tchambers dafuq? What about those of who don't want "social enabled agents that read your feed, draft your replies, manage your presence" on our network at all, in any shape or form? And what makes you think "build[ing] our own healthy, safe and fair versions of #AgenticWeb interactions" (whatever the f- that's supposed to be) would be more of an obstacle to bad actors than not building them at all?
@tchambers You know what AI could help with in the Fediverse? Writing #AltTxt that complies with accessibility standards.
@maximum_mew True that. @IceCubesApp specifically uses it for assisting in human creation of that.
I don’t want to contribute to the vitriol you’re certainly anticipating in response to this.
All I’ll say is that I think I get your point and where you’re coming from … but notwithstanding, this post has probably made me more anti-AI and even anti-tech.
Not because you’re wrong or have any intent I’d criticise, but because if we’re at this point, where this can make sense on the Fedi, I struggle to see the appeal of any of this tech shit, and increasingly wonder about the virtues of seriously throwing a lot of it in the bin.
A great thing can mutate into something repulsive and pernicious. And sometimes our only choices are to stay or leave.
@tchambers This made me think of what Project VRM was working towards a decade or so ago: https://projectvrm.org/
"The agentic Web" feels like a productive framing to me, in a way that something like "Web3" never did. Beyond software bots/agents specifically, agency -- who has it, and how -- seems like the core issue.
I'm no longer much of a technologist (if I ever was), but if you know of any folks/projects working in this area I'd love to see whether there are any ways I could help.
