Administered by:
Decentralization and erasure: Blacksky, Bluesky, and the ATmosphere
https://privacy.thenexus.today/decentralization-and-erasure-blacksky-bluesky-and-the-atmosphere-2/
There's been a lot of discussion about whether or not Bluesky and the ATmosphere (the ecosystem using the AT protocol) are decentralized. Blacksky runs three feed generators, a moderation service, and a work-in-progress personal data store (PDS) as well as providing a starter pack. And the vision for Blacksky "extends beyond any single platform".
That sounds pretty decentralized to me!
But as far as I can tell, nobody else in the discussion is talking about Blacksky as an actually-existing example of decentralization. What's with that?
The Appendix of Decentralization and erasure: Blacksky, Bluesky, and the ATmosphere is a roundup of various articles and posts on the question of whether or not Bluesky and the ATmosphere are decentralized and/or federated. There are lots of interesting perspectives here, including from @laurenshof on @fediversereport, @cyrus, @cwebber @bnewbold, @rysiek, @jonny, @possibledog, @oblomov, @rwg, and @Kye. Every single one of those posts was worth reading, and I really appreciate the time everybody's put into it.
That said, it's still very strange to me that as far as I can tell none of you mentioned what seems to me an actually-existing example of decentralization on Bluesky today.
> Blacksky could easily get their own up and running – by themselves, or working with some of the communities Fraser is already hosting.
The "easily" is doing a *lot* of work here. Roughly 16TiB of NVMe storage, based on available information – and growing fast.
This is not decentralized, the same way Google Search is not decentralized. Yes, one can spin up their own web search engine, but the cost is prohibitive.
Using multiple siloes is not decentralization.
@thenexusofprivacy and the thing is there is no reason for Bluesky and people promoting Bluesky to insist on using the term "decentralization."
Clearly, users don't care. And that's fine.
As many have written already, a fast-scaling alternative to Xitter is needed, and Bluesky provides it. Great! Godspeed!
What I do not understand, what is truly beyond me, is why the insistence on calling "decentralized" something that decentralized (in all the important, power-dynamics-ways) is not.
@rysiek @thenexusofprivacy I think what a lot of people may be getting hung up on is that ATproto is decentralized, but the thing Bluesky uses it for is not. Very similar overlap to how ActivityPub is decentralized, but Mastodon is a network of centralized instances that federate. What you see as a Mastodon user is all pushed to you from a central point, and is therefore "centralized," in basically exactly the same way that Bluesky's ATProto aggregator is. It's just that Mastodon currently has more than one of those while Bluesky does not 
@nyquildotorg the problem with this is that if one Mastodon instance dies, the rest of the Fediverse continues to happily federate.
If one Bluesky relay dies, as it had in the past, the whole of Bluesky is dead in the water, as it had in the past. Along with all the other AppViews that use it.
There is a single point of control and failure in the center of the whole Bluesky social network, in a way that there is no such single central point of control and failure in fedi.
@nyquildotorg and again, this is an architectural choice that is understandable in the context of what Bluesky's team set out to build. And no, it is not fixable by, say, running more relays, as that's just not how the ATproto system works.
And that's fine and if people want to use it, great, go for it. Moving from Xitter to Bluesky is definitely a good step to take.
But decentralized Bluesky is not, in ways that the Fediverse actually is.
@rysiek @thenexusofprivacy my only point is that "decentralized" means a lot of different things.
If your mastodon instance goes down, you're dead in the water, too, unless you planned ahead with recent backups, and even then you're going to lose followers.
Even the in-depth architectural apologists aren't calling Bluesky "decentralized," in their big "this is how ATProto actually works" posts, they're calling ATProto that, and they're not any more wrong than the people who say Mastodon is.
Indeed there is only one ATProto "instance" currently, but that was once true of ActivityPub too. I don't have a horse in this fight, and I'm not interested in Bluesky no matter what changes they make; I'm just pointing out that a lot of people are creating their own debate here based on how other people are interpreting "decentralized."
@nyquildotorg my point is that words have meaning. "Decentralized" has meaning. And Bluesky-the-social-network simply does not meet the criteria to be called that.
If my Mastodon instance goes down, I can set up an account on another one, and reconnect with folks. A bit frustrating, but doable.
If Bluesky's Relay goes down, it doesn't matter which PDS I am on, that social network is dead.
ATproto might be decentralized on the PDS level, but it is not on the relay level.
@nyquildotorg the problem with making analogies between fedi instances and ATproto instances is that these do not map one to another. I don't know what you mean when you say "ATproto instance."
Do you mean "a PDS"? Then there are many of them already.
Do you mean "an actually functional, usable service" the way a single Mastodon instance is? Then you have to include the Relay in that, but additional relays will not make Bluesky any more decentralized than it is. 
@nyquildotorg my "skin in the game" is that we cannot communicate effectively, if we allow terms to be diluted to a point of meaninglessness.
If we want to call Bluesky-the-social-network "decentralized" because some layer of ATproto features it to some extent (namely, PDSes), then we'd have to call Google Search decentralized because we can still self-host websites.
Problem is, a PDS is less directly useful to a person without a Relay than a website is without Google.
@rysiek @thenexusofprivacy I think in a way we're saying the same thing, kind of.
I've really struggled the last couple years with the descriptive nature of language, how words evolve the way people use them, making the popular usage of them "correct" despite it absolutely being incorrect.
I guess I'm saying that "decentralized" has already become meaningless, and did well before Bluesky showed up, and that it having done so is actively contributing to debates like this one. But, like with "crypto," me shaking my fist at the sky just makes me feel bad, and — for me at least — it's probably healthier to just stop using it rather than getting mad at it repeatedly.
I'm sorry I chose your thread to finally join in on in order to work out this point. I wasn't trying to argue with you, and agree with your sentiment about diluting meaning. But, because I was having "wait a minute, this isn't really decentralized," discussions about the fediverse, on the fediverse, in 2017, seeing happen again with Bluesky has been interesting to see.
Again, sorry I sounded argumentative rather than trying to contribute to your argument. I'm trying to work on that lol.
@rysiek@mstdn.social@nyquildotorg all good, debate is what we're here for.
> I guess I'm saying that "decentralized" has already become meaningless
I don't believe this is the case. And even if it were, I don't believe we should agree to that. In fact, I strongly believe we need to fight for words to have meaning, to not be diluted and made meaningless. Otherwise we cannot have meaningful conversations.
I don't think decentralized is meaningless, I think it has different meanings and interpretations.
Does decentralized refer to network topology, power dynamics, or both?
If we're talking about network topology, does it matter if different nodes in the network are owned by different entities?
If an architecture is in principle decentralized, but the current implementation has one or ore single points of failure, is it in fact decentralized?
If one layer of the system is decentralized (web, PDSs) but power is heavily concentrated in another layer (search engines, relays) is it a decentralized network?
If we're talking about equitable distribution of power, what kinds of power are we talking about, and how equitable does it have to be to be considered decentralized?
etc etc etc
@thenexusofprivacy @rysiek where I keep getting hung up is "who controls my identity and content?"
On the fediverse, my identity and content are absolutely controlled by my instance, which makes that a huge obvious centralization point.
Sure I can "migrate" my identity, but that just causes (many of) my followers to automatically follow the new account, which comes with its own new identity. Now there are two identities out there. Unless my instance goes down, in which case I cannot "migrate" my identity.
But then, topologically speaking, even if I host my own instance on my own domain, I can't ever point that domain to a different instance without causing federation problems that require all the other instances on the fediverse to take action to get federation going again.
Those two things make my experience on the fediverse incredibly centralized.
@thenexusofprivacy @rysiek I do think "has different meanings and interpretations" makes "decentralized" unsuitable for debating whether or not the fediverse or Bluesky is decentralized; "many meanings" doesn't mean "no meaning", but I think it does when you're trying to use it to compare different aspects of different things with different people because it's almost impossible to be on the same page.
"Meaningless" is just shorthand for "no common ground to argue from."
@nyquildotorg whenever I see something like that claimed, I get suspicious. Who benefits from making such terms so blurry?
We've seen this time and again with "open source". Microsoft and others repeatedly tried to claim that licenses that do not meet basic criteria of being open-source are somehow "open-source".
We've seen this with encryption, for example in the context of Telegram. Telegram insists it is "encrypted", even though it very much is not.
@rysiek @thenexusofprivacy I don't think anyone is "making" decentralized blurry, I think it's always been a poor marketing term applied to whatever people are trying to promote the usage of.
@nyquildotorg if it is such a poor marketing term, why are Bluesky developers defending it? Why can't they just say "sure, Bluesky is not decentralized, so what?"
Users would shrug anyway. So what gives?
Why are they investing so much time and effort into fighting for that hill?
I know why *I* am defending the stricter meaning of the term, and why *I* consider it important. But what is *their* motivation?
@rysiek @thenexusofprivacy well, two things:
1) Bluesky devs are defending "decentralized" with ATProto, not Bluesky.
2) they're just doing so on reply to high profile people on the fediverse claiming that what they're doing isn't decentralized.
Both sides have fallen into the trap I'm trying to illustrate: it's not a meaningful discussion because "decentralized" means different things to each side of the argument.
@nyquildotorg just to be clear: I am not claiming they are doing this maliciously and with premeditated ulterior motives.
But there surely is a reason why they invest time and effort into writing blogposts and taking part in conversations about Bluesky where they defend the claim that Bluesky is decentralized.
If it's such a bad marketing term, why do all that?
@nyquildotorg is e-mail a centralized or decentralized service, in your opinion?
What about XMPP?
Can you provide an example of a communication system that is, in your opinion, decentralized?
@rysiek @thenexusofprivacy email and xmpp are federated.
"Decentralized" doesn't apply, because in order to get and send your mail, you have to connect to a central point.
(Email is also now a bad example because so much of it really is centralized. Almost every email you send or receive talks to GMail servers.)
@rysiek @thenexusofprivacy p2p things like Gnutella, or BitTorrent (with dht) are closer to decentralized, but this is all kind of supporting my claim that "decentralized" is not a good term to use for discussing merits of platforms
@nyquildotorg there is a difference between a "decentralized" system, and a "distributed" system. Peer-to-peer systems are better described as "distributed" systems. Nobody is claiming fedi is a "distributed" social network.
@nyquildotorg you don't have to connect to a "central" point.
You have to connect to *a* point. There are *many* points that could be your entrances into the network.
The network will survive if one or several of such points disappear. Such network is a meaningfully more decentralized network than a network with a single global point of failure.
@rysiek @thenexusofprivacy with email, you do have to connect to the email server that (federated, not decentralized) DNS records point to. If your mail server goes down, email sent to you does not get to you set up an account on a new mail server and then configure DNS records and hope an eventual retry delivers that mail to the new central point that manages your mail.
XMPP works essentially the same. The servers federate, but are not decentralized. And even if you are ok with calling the server topology itself "decentralized," you still have to lean on DNS, which is absolutely not decentralized.
> And even if you are ok with calling the server topology itself "decentralized," you still have to lean on DNS, which is absolutely not decentralized.
Incidentally, this is 100% the case with all Bluesky identifiers currently.
@rysiek @thenexusofprivacy just to be clear, I've never claimed that Bluesky is decentralized. My claim is "neither is the fediverse," and all these discussions going around and around about it get us nowhere."
But also, the fediverse leans on DNS, too, so I guess this is a good time to pack in the discussion about "decentralized" lol.
@nyquildotorg sure.
Basically to me this boils down to some version of: could Musk buy out a given social network today and screw it all up if he throws enough money at it.
With Bluesky, that's a "yes". With fedi, that's a "no". That, to me, is a pretty important difference in the context of *why* people are migrating off of Xitter. 
@rysiek @thenexusofprivacy I think the best we can do is rank the "aspects of protocols that allow for aspects of decentralization" against each other weighed against what's important to us.
For me, where Bluesky fails is at "a for-profit entity currently doing all the work." It actually is possible for someone to create a non-Bluesky relay, which would put Relays and Instances on a more comparable level.
What I like about atproto is that it does actually provide a mechanism that divorces the identity and data and content from the "instance," as well as the ability to move instances without being just a new identity with a tentative pointer to the old one. I very much hope the fediverse evolves to implement either a version of this, or even just atproto's PDS.
Doing this will still rely on the central failure point of DNS, of course but that's already a central point of the entire Internet.
Nomadic identity fulfills (for me) an important part of the "decentralized" dream that a lot of people have bought into (and want to defend to the death), and I hope to have it some day. But I'm not willing to move from a centralized-bit-noncommeecial platform to a centralized commercial platform to get it, because the non-commercial part fulfills more of the dream of "decentralization" (for me) than the nomadic identity part does. But both are possible, and that's what I'd like to see. I just can't bring myself to talk about those things as "decentralized," and my hope is that critical mass will abandon "decentralized" as a descriptive term.
@rysiek @thenexusofprivacy I absolutely agree. But Musk buying Bluesky is actually the primary thing ATProto was designed to prevent, and part of the reason I think they've joined in on using (erroneously , I believe) "decentralized."
(Whether their VC funding will allow that to remain true is something I am absolutely not holding my breath for.)
@nyquildotorg @rysiek @thenexusofprivacy I ask how is it practically different than Musk buying the two instances by Mastodon nonprofit which make up 30% of the MAU. People notoriously don’t like change, there’s no easy way for them to keep their content, their social graphs would be severely impacted. No it wouldn’t shut down the fedi but it would be catastrophic
@damon the difference is that *if* people from these instances decide to move to other instances, their contacts from the remaining ~30k instances will still be there, accessible to them.
Plus, those largest instances are not managed by a *company*, so "buying them out" is a bit more difficult.
And finally, I wrote about the large instance problem on fedi a while ago:
https://rys.io/en/168.html
@rysiek @nyquildotorg @thenexusofprivacy you can’t say that for sure. If it was Elon Musk people would respond in kind by blocking those instances so fast and if people never downloaded the .csv those people are screwed
@damon there is a massive, massive difference between "all of the network and every single person on it is no longer accessible to you" and "a large part of the network is no longer accessible to you, but a large part of it remains accessible to you".
I will insist that such difference *is* meaningful.
Because it *does*, in very real way, undermine the network effects holding people hostage to *actually* centralized services.
@damon and also note how many "ifs" you had put in there:
- *if* it was Elon and other instances blocked…
- *if* people were not able to download .csv…
These two "ifs" do not even exist in Bluesky. No CSVs to download, no independent Bluesky-social-network instances to migrate to, no migration facility…
And even then "these people" are not "screwed". They can set up accounts on other instances and reconnect with people already there.
I totally agree about the large instance problem on fedi. One way to look at Bluesky is as a very large instance with internal structure attempting to avoid moderation and monoculture problems, tbd how well it succeeds on that front. And Bluesky PBC currently is far more dominant in the ATmosphere than Mastodon gGmbH is in the fediverse, so the consequences of somebody buying it are more significant than somebody buying Mastodon gGmbH (which could certainly happen although it doesn't strike me as particularly likely).
In terms of data portability, if somebody gets kicked off of .social they lose their posting history and social graph unless they've backed it up. If they have backed it up (or if they migrate before getting kicked off) they can largely recreate their social graph someplace else (although .social can still block them from communicating with anybody on .social, and there might be other instance-blocking issues, plus migraton often loses some followers). In principle the Bluesky situation is somewhat better: you can import your posting history into a new PDS, and if you're running your own PDS you don't need even the separate import/export step. In practice though right now there isn't a meaningful "someplace else" (and it's not clear if and when that will change) so the reality's not as good -- if you're blocked from the Bluesky AppView you basically can't communicate with anybody.
That said, Bluesky's story is much better story than fedi's "we didn't design for this and haven't made any progress in the last N years but since BLuesky started giving us a hard time about it there's now a SWICG working group that's got a draft spec that nobody's implementing!"
It's frustrating because this absolutely could be solved in an AP framework but nobody with resources has any incentive to solve it. In Cory's piece on Bluesky he said something like "Any system where users can leave without pain is a system whose owners have high switching costs and whose users have none", hmm I wonder why Mastodon gGmbH hasn't tried to reduce the pain of people being able to move?
> In practice though right now there isn't a meaningful "someplace else" (and it's not clear if and when that will change) so the reality's not as good -- if you're blocked from the Bluesky AppView you basically can't communicate with anybody.
I think it is important to add: and that's not just because Bluesky's Relay *happens* to be the only one, but because for Bluesky-social-network, Bluesky's Relay will *never not be* the only one.
@rysiek @thenexusofprivacy @damon would you two mind removing me from future replies. My position remains the same: that everyone is arguing about different things here, that the important parts of "decentralization" are different to each participant, and every new reply just reinforces that and makes me mad lol
@nyquildotorg sure, will do.
@thenexusofprivacy as in, this is significantly different from early Mastodon days when there was only one Mastodon instance.
Adding instances on fedi meaningfully removes power from the few big instances and Mastodon gGmbH within fedi.
Adding ATproto relays *does not* meaningfully remove power from Bluesky Social, PBC and the Relay it runs and controls within Bluesky-the-social-network.
Bluesky-social-network, Bluesky's Relay will never not be the only one.
It depends on how you define Bluesky-the-social-network. If it's just in terms of Bluesky-the-AppView, then you may well be right. If it's defined as "the people currently using Bluesky", I think it's very likely we'll see alternate Relays and AppViews within the next year. Time will tell!
as in, this is significantly different from early Mastodon days when there was only one Mastodon instance.
Agreed. So we'll see how Bluesky and the ATmosphere evolve. My perspective is that Mastodon evolved in a way where Mastodon gGmbH had enough influence that aribtary decisions by Eugen led to innovation basically flatlining -- and Mastodon's dominance within the fediverse, combined with SWICG's inaction, has held everything back. So there's room for Bluesky and the ATmosphere to do better ... or to hit the wall in different ways. Once again, time will tell!
> If it's defined as "the people currently using Bluesky"
This does not make any sense to me at all other than "people currently using Bluesky-the-AppView". That is the only meaningful definition of "people using Bluesky".
Would I be able to connect and interact with these users while using my own PDS, my own Relay, and my own AppView?..
@thenexusofprivacy @rysiek @damon
> I think it's very likely we'll see alternate Relays and AppViews within the next year. Time will tell!
I'd be very interested to see if this actually happens; right now Bluesky (the company) is not profitable and has no real path to profitability (and are openly discussing ads,) relays are bananapants crazy expensive, and so anybody running one will have to be either incredibly well subsidized, or for-profit.
I remain unconvinced that a for-profit entity can ever truly be a reasonably safe space.
Right now Bluesky (the company) is the centralizing entity for both the technical and organizational sides of this coin. The technological is so far ok, so long as bsky continues to feed it money, but the organizational is essentially the same kinds of problems we see on fedi but with only one ultimate end to them. For example: what we've seen in the past few days about bluesky trust and safety demonstrating yet again that they are explicitly anti-trans.
From a technology standpoint it's very interesting, though despite trying to understand ATProto well enough to come up with a cogent explanation for how multiple relays would work, it would seem like to me that multiple relays would ultimately appear like two separate and non-interactive social media "streams" running on the same infra, like two protocols running on IP sharing a network. So unless I understand wrong (which is, of course, a real possibility) it can never be federated, or at least not in the sense that anyone using "federation" in good faith actually means. Certainly not like email, and definitely not like xmpp or Activitypub is.
From an organizational standpoint, it has all the same problems Twitter had pre-Musk. No surprise there.
We shall see.
Your point is mute then because you said if Elon comes to buy it. I’ve seen you move the needle so many times it’s ridiculous and I can’t take your opinion seriously. I don’t believe you engage in good faith. You have your opinion which as others have pointed out in terms of decentralisation is very flawed, you don’t budge. You point out my using ifs yet I’ve tracked all of the conversations you’ve had concerning this and yours are also laced with ifs and possibilities
@damon I guess it's best we don't continue this conversation then. 
@thenexusofprivacy yeah, these are all important distinctions.
Distinctions that are very often *not* made when somebody is claiming that Bluesky is "decentralized".
And I would argue that while fedi is far from perfect, it is decentralized in more meaningful and demonstrable ways on more of these levels than Bluesky.
And I would argue that calling Bluesky-the-social-network "decentralized" in any meaningful way is – currently – incorrect.