How are folks doing? This is gonna be uh this is gonna start. I just want to preface something. This is gonna start as a little bit of a turn in the punch bowl kind of presentation. I want to promise you it's not gonna end there. Um this subconference is called At Science, and so I'm doing a little bit of at social science today because I'm a social scientist, I'm a PhD student in communications at the University of Pennsylvania. Um is really good for a couple things for social scientists. First off, there's everything that Sophie just talked about, which is fantastic, and I forgot to actually mention here, but in terms of the experiments that enables you to run.
Um if you have spoken to any social scientist who is studying media effects um on social media, they will talk about that exact same complaint, which is we have absolutely no control over what is going on, and so we are just try left to uh go through behavioral data. But even within that, it's gotten a lot harder recently because we're living in what some folks call the post-API age, where following a bunch of stuff in the last five to ten years, a bunch of APIs on most like big social media platforms were either made private, or if you are allowed to access them in the case of X, the rates are really exorbitant.
Or you have to get your, I think in the case of TikTok, you have to get like every specific part of your project approved by TikTok, which does have a sort of chilling effect, right? If you want to ask these really kind of critical questions, and you have to go to TikTok and say, hey, I want to ask really critical questions about you, will you give me permission? It's kind of an awkward process. Or you have to use something like Meta's content library, which is the official way you can look at like Facebook and Instagram information, and the problem with the platform like that is that it does you have to do everything through a custom browser that they have because they're really worried in the wake of Cambridge Analytica about how folks use their data, and that is a good concern to have, but the way their solution ends up working, ends up working really limits the amount of data analysis you're able to do.
Um so it's not great. And one of the really nice benefits of ad protocol, um permission spaces uh notwithstanding, is that because everything is public, you kind of avoid all of those concerns about pricey, hard to use or hard to access data. But the second thing, and this is what I think is more interesting, uh, is that it qualitatively changes the type of question you're able to ask. So I'm coming from the political economy of communications where we ask questions about how does the ownership of different media systems affects the affect the type of media that is produced.
Um they've asked a lot of really interesting questions in the past. Um there's a lot of stuff, uh, there's a whole literature that I won't go into here looking into like how does uh commercial media funding uh affect the nature of journalism, how what sort of effects does it have on a democratic society, all these really important questions. And folks have done really good thinking about what that means for social media, but there's kind of never been an A-B comparison you can make. Right? If you look at the if you look at uh journalism, you can say, look, there are some newspapers that are funded by advertisers, there are some that are there are some that are more publicly funded.
Um but on social media, past in the last 20 years, it's mostly been the same handful of dominant players, all of whom are operating as privately owned corporations where they have the same incentive set. Um and there's very little you can do. Um with Bluesky existing though, you can start to ask these questions about like what would a different model look like. And so a great example of this is um some work looking at uh looking at the different third-party services offered on Bluesky. Um it's kind of a an ethnography or a digital ethnography and a little bit of a uh a qualitative project, but their main finding is that if you look at uh feeds, starter packs, moderation lists, and labels, uh most there is not a lot of financial monetization.
I don't think this will be a huge surprise to surprise to most of the folks in this room, but most of these things are built because hey, I'm part of a community, I care about this community, I want to go build this tool for them. I think many of the presentations that we've heard today, that's how they start is I wanted this thing to exist and I went and build it. And that is great, and that is not necessarily a bad thing, but it does change kind of qualitatively the type of network that we're working with in.
Um there's also this point about supply often outstripping demand. A lot of folks will go and build a specific tool for themselves. They will ever I I know many folks here have gone and built a feed because they wanted to see one thing specific. Um the side effect of that though is that most if you look at where users are actually spending their time, a lot of them are just using whatever defaults are promoted to them through Bluesky or using a handful of larger actors. Uh, which brings me to my question, which is how decentralized is Bluesky really?
Like we're we're decentralized network, right? That's what we're here for. We're decentralized, that's what we talk about. Um, I want to say I'm building this work. Um Balduff et al. who wrote one of the first academic papers I could find on Bluesky, uh did a similar measurement of um handles in 2024. Um, and then uh a guy named Kuda Suter, who I have not, Kuba Suter, who I've not met, I do not know, uh, but I learned that he did a lot of this data analysis similarly uh literally yesterday. So um a shout out to him, he has a live feed of a similar version of this data because I'm always this presentation is gonna be a lot of me getting nerd sniped.
Okay, there are a couple ways you can measure decentralization, and this is where I I leave a little asterisk there because this is out of date. I would like a more rigorous way to go through this, and so since there are out proto developers here, I thought if anyone would know how to do this in a better way, it would be you guys. Um the first is you go to the relay, you say, hey, can you give me your list of hosts, we go to the hosts, we say, hey, can you give us a list of uh repos you're storing, and you you measure that.
Now the problem is not every personal data server is gonna be uh is gonna give you the list of repos, not it's not always gonna be accurate, um, and it's not always gonna tell you if that user is active. And so the second way you do it is you go to PLC, you say, hey, can I scrape every action that has ever been taken? And then from those actions you say, hey, when is the last time each user updated uh where their data is stored, and you combine that, and the number you get is something like 99.7% of all accounts are owned by Bluesky.
That seems pretty seems pretty centralized. But okay, I I know what you're saying. I know, I know what every single one of you is saying. But Billy, my name's Billy by the way, hi. Uh, but Billy, it might be centralized in practice, but in practice, the people who are really spending time on the network, the folks who are posting every day, who are liking stuff, who are getting followed, those are all the those are us. We're we're the cool people, and we go and we host our own data. So clearly, if you look in practice, it's not gonna be the same.
That's a great point. And so another way you can do it is you can look at the fire hose. You say, okay, let's let's look at the fire hose, let's look at you know, sample over the course of an hour or a day or a week, and let's see how much activity is coming in, and let's see for that activity uh who is who is being followed, who is being liked, and who is posting. And does that activity, is that more decentralized? Um you do that, you resolve the idea, you look at the files, it's not great.
99.9% of follows are going to uh accounts hosted on Blue Sky's PDS. Same with likes. The promising one is that if you look at posts, just over 1% of posts are made by folks who are self-hosting or using another place to host their data. This is the turd in the punch bowl part of the presentation. Now, um I want to talk about a few other things, which is in talking about the Atmosphere and talking about Bluesky, we often talk about a credible exit or as I like to think about it, like what would happen if 2022 Twitter happened tomorrow.
Um there are a couple other concerns. This is the continuation of the third. Uh the first one is a lack of adoption and a challenge with funding. Um I think we are all aware of this. Uh Paul Frazy uh in his recent February 2026, he wrote a little bit about this, how I think it is well understood that most of this stuff is still being owned and hosted by Bluesky, and that does present a certain level of problem. And that gets back to the paper I was referencing earlier, where oftentimes we are really we are building these economies, but these economies are based off of subcultural capital, they're based on gifts, and although I think there are really promising uh moves in projects like Grays, the funding of these systems is all is not clear.
It's also not clear for a user what bat benefit they get out of going out of their way to host their data somewhere else. Um that's not an easy solution, and I think it also reminds us that the folks in this room are gonna be amongst the most technical users of Bluesky. By way of example, a week ago I was visiting my grandparents. Um boy, I tried explaining what I was presenting this weekend to them. I don't think a single part of it computed. I love them dearly, but if we want to have this as a protocol that serves everybody, we have to think about what incentives folks have to be actually decentralized.
Um gets to the broader point of the what the limits of what a protocol can do. Um a protocol is only as powerful as our willingness to follow what the protocol says. So if you think about it, if tomorrow uh Bluesky Social, let's say it went worse to the worst, you know, Elon Musk took it over tomorrow, and they said, we're actually not gonna interface. You can tell us that you want to migrate your data, tough nuggets, we're not gonna let you do it. Um, there's nothing really we could do. They could choose to hold on to your data.
They could choose to make the platform no longer interoperable with anyone who's not hosted on Bluesky Social directly. It wouldn't be easy for them, but if they choose to not follow the protocol, they don't have to. And this brings me, if anyone here is a lawyer, I know what you're I know what the non the zero lawyers in this room are saying. But Billy, Bluesky is a public benefit corporation. It's a great word. It means that you are legally in the US allowed to have a responsibility that is not purely financial. First off, it is crazy that that needs to be specified.
But second of all, it is a really nice thing. It's about a decade and a half old. I think 2013, Delaware where most uh is the state where most uh American corporations, most American businesses are incorporated is in Delaware, um, they start allowing it. The challenge is in 2019 there was a change in the law, stick with me, uh that changed it from if you want to go from being a public benefit corporation, meaning you have some mission to not one of those anymore, it only requires 50% of your of your um stock hold of your shareholders to vote against that.
The consequence of that is we are now in a place where if 50% of the folks who own Bluesky decided to no longer have a decentralized mission, they could do that tomorrow, and there would be nothing you could do to stop them. Which brings me to another point, and I'm gonna say this really quietly because of the sponsors on the wall. We don't know who owns Bluesky. That information is not public. We know the list of people, but we do not know how that is distributed amongst those different owners. So it's a concern because if they wanted to tomorrow, the entire protocol for everything valuable that is built, there's nothing keeping that Bluesky information from being re-siloed.
Okay, we're gonna take a breath. Let's talk about some good stuff. Um first off, the move towards in this independently spinning off uh the PLC directory is a really good one. Um that will at least allow so that your identity can't be shared elsewhere. I think that is a really promising first step. The second thing is these data backups. Um I uh met Fig briefly earlier in the elevator today, uh, very nice person, uh, like Hubble, um, are really valuable, right? Because we want to be able to say, listen, not only can I move my identity over, but I can also move the data with that over.
So having these public backups is a really useful way of doing that, especially for users who a priori might not think to go out and make a copy of their data. I also want to give a shout out to the folks in this room who are working on building um other sort forms of PDSs. I'm thinking of BlackSky, I'm thinking of Eurosky, uh North Sky, there are any number of these, um, but people really think doing the hard work of thinking about how do we build these other communities, that is incredibly valuable, and God willing we can reach a decentralization through those efforts where this whole presentation is a moot point.
Um finally, all of you wonderful folks. I want to be clear when I'm saying all this. I'm just trying to be honest about the situation that we find ourselves in. I do not think this is a foregone conclusion, nor do I think these are in unsolvable problems. Um and so I'm really I've been really motivated over the last couple days speaking to everybody and seeing how much everyone cares. Um of course I said I got nerd sniped, and nine days before this presentation, Lawrence Hoff wrote a blog post about exactly this. So thanks for sitting with me while I repeated stuff that he mentioned.
Um I really like this quote of him and I'm gonna abuse uh Scott, your metaphor about the party. Uh not only do we have a good party set up, I think we have a great party going, but the problem is everyone's stuck in the kitchen, it's getting a little cramping here, and we'd really love it if some people move to the living room, some people move to the balconies. You just got a little bit more spaced out. That's my presentation. Thanks, guys. Thanks. Uh any questions?
Hey, I'm Daniel. I do protocol stuff at Bluesky. Don't worry, you're not offending any of us by saying this. We basically agree with you. Uh and I'm talking about very similar stuff in my talk on Sunday. One of my questions is like there's sort of this logical impossibility, like we want the network to decentralize more. There's sort of this logical impossibility for Bluesky decentralizing it. Like you almost need other parties to step in and take power and authority and control in the network. So what do you think can encourage people to do that more?
That's first off, uh, it's really nice to meet you. I've read your stuff, I've read a lot of stuff in this room. Absolutely. Um I really was hoping I would end this presentation with a slide of here's all the things we can do. Um, I think part of but what I've thought about it more, again, I'm coming from a social science background. One of the things I want to push us towards thinking about just as a community is all of the, you know, we're talking about so-so technical systems, all the social ways that these networks are built up.
And so I think my point here is not that there is an easy solution, and I really wish there was. I can kind of spitball ideas against the wall. My point is more so to say that I think there is just to make the point that there's a limit to what we can do with the protocol is really valuable and there's a lot of stuff that can be done, but that is insufficient to do it on its own. Now, I could start daydreaming with you. Um there's really good uh proposals um with folks I've worked with and folks I've seen uh looking at like local anchor institutions, so could we have uh you know public funding towards uh like local hosting of this.
I mean, this is an ideal world where the government landscape looks different than it currently is and the funding landscape is different. Um there are questions about kind of the it's that kind of civil society question of can we rethink models and funding of social media and that goes again in this long history of debates about um like public funding of journalism, for example. Those are some of my more optimism optimistic takes, and I I want I would love to talk more after this. But generally I do think we run into this issue where there's just a limit to what we can do technically.
Anyone else? That was awesome, thank you. Uh I am Jeremy Miller. I'm actually the original board member on Bluesky. And I can tell you just in words, right? I can't promise it, but I would put my reputation behind everybody on the board, everybody involved is fully behind this decentralized distributed mission. So as much as all the things you said are true, uh we are fully behind this and we believe in it. I want to be thank you so much. Um I want to be very clear, none of this is attacking anyone. I'm just trying to think through what is the worst case scenario if like you know asteroid came clam clam cla Jesus, asteroid came crashing into Earth tomorrow, or Elon Musk came in and said, Hey, how much money to have Bluesky 2?
Just thinking through that stuff. Absolutely. Yep, I am. Last questions. All right, let's thank Billy again.
These are also great things to think about in the UN conference, uh, which will be starting in about 10 minutes, I guess. So um if you have yeah, ideas, you can put post-it notes on the boards in the back. And when we meet in about 10 minutes, we'll come and cluster around topics that we're interested in, and probably in this room, um maybe like six groups or so. So yeah, feel free to take a break, be back in 10 minutes. And yeah, last session, last session of AT Science.