I'm Laurens Hof. I write connected places. It's a newsletter about Admosphere, Ed Proto, Bluesky, Fediverse, and the ecosystems, the open social ecosystems. I'm very interested in like figuring out like what are these systems actually. Like there's a lot of cool narratives about what these systems can create, but I'm I want to understand what are they. And recently Paul Vizy wrote an interesting article, practical decentralization, where he made two specific points. The point of decentralization, he wrote, is to guarantee the rights of individuals and communities on the internet. And secondly, he wrote we use protocols to structure what what who can do what.
Protocol designs are often about the how, but the consequence of the how is an authority model. And I think that's super powerful because that really sells what's actually at stake here. Like building protocols is cool, but what we're actually building is authority models of like how social interaction happens on the internet. Like it's easy, like many people have been in getting interest in open protocols as a sort of response to big tech bad and well, obviously big tech bad. But the reason big tech bad is because of the governance of the big tech platforms. So that means very much like if we want to build different platforms, new ecosystems, this is very much about what sort of governance and authority models do these new ecosystems actually allow.
And for that, I'm always very drawn to the statement by uh the cybernetician uh Stafford Beer with its great one-liner. The purpose of a system is what it does. And I think that's both simple and powerful because it lets you cut through like all these normative statements and intentional language about what you want to be happening, but instead you can look at what is the system, what should be happening. Like what not what should be happening, sorry, but what does it actually produce. And for that, it's also worthwhile to look at a little bit of the history of the internet.
Like internet runs on email, the internet runs on open protocols, and the most famous one for emails is the SMTP protocol. And that protocol just said very little about purpose or governance or anything at all. It just says the goal of this protocol is to transfer mail reliably and efficiently. And it turns out we've been getting extremely good at transferring email reliably and efficiently, and it also turns out that spam loves to be trans uh uh transferred reliably and efficiently. So once you build an open protocol and like you have this statement of like, hey, we want to do this thing, and then otherwise, like you have a technical statement, but you don't say anything about the governance or the power that you create with this new system.
You can have a decentralized open protocol all you want, but these gaps for power they get filled by economic models. And that's exactly what we see we have seen with email. Like spam filtering turns out to be a huge problem. Google got very good at that, and now email is basically duopoly with Google and Microsoft. So open protocols still duopoly, still centralized power because the protocol itself did not account for how it generates power. In general, that means power and value accumulates in the layer of a protocol that doesn't govern. We've seen the same with HTTP.
Like HTTP also the uh the spec of the protocol has like says nothing about power. Like uh Tim Berners-Lee made lots of blog posts around it about like great ideas of what you want the power to be. But because there wasn't defined in the spec anywhere, then um that allowed that value and the power to be accumulated in the layers around it, and that's how we got in this system of like well, big tech bad, because these plan these companies could accumulate this power in the layers around HTTPS. And we've seen like we've been doing open protocols for a long time, like since the early 80s.
And now we've seen that um in the last decade or so. We've seen new protocols arise, we've seen ActivityPub, we've seen Nostre, we see Matrix and now atproto as well. And these protocols start to incorporate normative language in the protocol documentation. They say stuff about governance, they say stuff about what what should this be doing, what should this system create. And I I got uh three quotes from the atproto documentation here, and I I do think it's very interesting to see the normative statements pop up here. Like speech and reads should be two separate layers.
Also the very normative, everybody should have everybody has a voice. And users are free to select their aggregators. Protocol design sort of tradition sort of assumes a sequence. First you build an architecture, the design specification, then you grow an ecosystem, and then you add governance to that ecosystem. But it turns out like growing that ecosystem is the same as adding governance to this ecosystem. These are because if you do not actually specifically specify like the governance of the ecosystem, once you grow it, like the implicit forms of governance, which are often like if it's not specified, it's just the most powerful economic actor who is there first, turns out to be become the governance layer.
And the reason for that is that because if you grow an ecosystem or open protocol, you create shared resources. And these shared resources are themselves governance structures because they compete, they involve competing interest. For example, here on atproto, like if you build if you run a relay, you immediately need to make governance decisions. You need to decide am I gonna which spam I'm gonna filter out? Am I gonna filter the abuse out? Am I gonna filter copyright material out? Or am I gonna transfer these like am I just gonna say I'm a neutral neutral layer that just transfers all of this.
And like there is not a neutral way to transfer spam or a neutral way to spam how to um transfer abuse in your relay. As soon as you operate a relay, you have made governance decisions because these shared these are the shared resources. This is what I like this article I wrote recently about the purpose of protocols, and uh I got two responses from uh Paul on that that I think both are worth highlighting. First, you said like my belief and hope is that we've opened the opportunities for other forms of communal resource governance to be implemented.
And like I very strongly agree with that. Like, this is why I care about this open ecosystems, and this is why I care about like this conference, because like there's so many people around here that like that can we can build these systems now. We have created this infrastructure, we can create this communal resource government. Like we have this open open relays, we can build the governance structures for it. But Paul also says I'm almost ready to say that the economics are determinative of the governance. Well, I am actually ready to say that the economics are determinative, but if they are determinative, what do they actually determine.
For that, let's look a little bit more in detail on what approach actually does. It basically it creates two layers a speech layer. Speech is just like it's cheap, it's permissive, everybody can publish, it's very easy to set up your own PDS. And then there's the reads layer. The reads is like that's where you get the feed algorithms, that's where you get the app views, that's where you get the indexes, that's where the people actually interact with your app. And the reads layer, that's actually the expensive part of operating the infrastructure, and thus that's where the economics are actually determinative.
And that's actually what you now see in the Atmosphere in practice. Like Bluesky, the company is like three orders of magnitude larger than any of the other operators, because the economics are determinative, and that's the uh they operate the expensive layers of the stack. I want to zoom in even more. I'm gonna specifically zoom in on like the algorithmic choice here. We do like proto allows for algorithmic choice, allows for everybody to run their own feed operators, etc. But feed providers need users, users want engaging content, and the thing is what counts as engaging content is what everybody else is engaging with.
Like our preferences for what we want to see, what we what we think is engaging is not neutral, it's not defined by the individual, it's defined by the people around this. Because we are like our preferences are shaped by other preferences, and our preferences are intrinsically mimetic. And for that, I want to zoom in on the for you feed by the independent developer Space Cowboy. SpaceCowboy it's an amazing feat, and um it's a uh the it has actually a fairly straightforward mechanism. It looks at the post you liked, like then it suits at what other people also like that post and what other people they like, and they suggest then that post to you.
This even like this very explicit sense of like how how we made it visible of like how our preferences mimic others' people's preferences. That is exactly in its most pure form what the for U feed does. And actually, if you spend any time here in the ecosystem, the atproto developer ecosystem, you know that people actually really like the for U feed. It is a very great feed, and people recommend it all the time. It's almost kind of a meme. Especially if you compare it to the other popular uh disc algorithmic feed, the discovery feed by uh Bluesky itself.
And that's also a bit of a meme to almost hate on like the discover feed. Shout out to Dan for that. Like the the reason for that is that uh what sorry what's actually interesting about the discover feed, uh Blue Sky's discovery feed is that it's an algorithmic feed, but it's not purely engagement-based. Because uh Bluesky is also trying to solve for other problems than just pure engagement recommendation. They also solve for a problem of hey, there's a new person signing up. They um we don't know any better anything about that new person, but we don't want to show them an empty feed.
So how are we gonna show them some cont some interesting content via the discover feed? And the discovery feed is very good at that. It has multiple purposes and it it's good at those purposes. But what it reveals and what you can see in how the AdPro the developer community engages with both feeds is that the reveal preference is that people strongly prefer pure engagement-driven feeds over generically algorithmic feeds that also provide other features. So now we start to see there there becomes a sort of a loop. And users engage with what other people engage with.
And what atproto does, it creates this single shared data space. Like this is the the nature of this open protocol that everybody has a PDS, like and the relays, all the relays, all the app views connect to all this single data that data space, like this means there's this singular shared data space. So users engage with what other people engage with, and that gets captured by that shared data space, like that gets captured by the likes on the lexicons, etcetera, etc. And these feed providers then s they look at this entire data space, pick out the most engaging content from that data space, because that's what the feed, that's what the feed providers want because that's what the users want.
And then you see again, get back to the top, users engage with what other people's most engaged with. So here we get this very interesting dynamic. What it means that if you have algorithmic choice in a single shared data space that produces convergence. And that means like you have what I mean with that is that because of this loop, we get like we converge on this single centralizing point of the most engaging content. And like this doesn't mean there's room for like niche content and niche feeds. You can still always create the MOS feed all you want.
But this is by definition niche. And there's like what you get is like this main set of feed that are like highly engagement driven because the underlying data space is like reflects this engagement-driven optimization, and you have a very long tail of a whole lot of niches of algorithmic choices. You can actually already see this in Blue Sky's culture, where like all the data is open, you can look at what content is the most liked on any day, and that consistently converges on American left liberal politics. And that's exactly what we would expect in this virtuous feedback loop of like the that the this pro uh diversity of feed providers selects for the most engaging content.
So the open system is competing on engagement, and that's how you get convergence. So you have an architecture that says pluralism, but the economics of it says no, you actually have convergence. So we have a protocol like the protocols that focuses very much on the visual rights, but these rights only matter as much as that you your choices to make to actually also are meaningful. And that that's how this openness produces convergence because this governance is entirely absent in the in the protocol layer. And that's exactly what we've seen in earlier dynamics for open for open protocols for SMTP that also produce Gmail.
Like the mechanism is different, like that was spam filtering, and here it's in instead of like engagement optimization. Like a protocol, the mechanism is that a protocol creates an open space, but because the governance of this open space is undefined, you get pushed towards economics, and the economics pushes towards concentration and not towards diversification. And now we finally get to the point where obviously you all want to do uh waiting for, we can finally talk about K-pop. Because like K pop are like the the fandomps are some of the most largest fandoms and the most coordinated engagement optimized driven comma optimization driven communities.
And that actually works in both directions. Like you have the companies who produce the music who are strongly commercially driven and deliberately optimized for impact and for engagement because well they're companies, they want to make money, and it's much more important to them than the artistic impression itself. Like I'm not shitting on K-pop here. I love K pop, like I fucking love Jenny. Like we should all be more well like Jenny, but like the companies maximize engagement. And the same actually is also true for the fandoms themselves. And I think it's actually interesting to look at one YouTube comment section on a new K-pop song, like to see what are the fans actually talking about.
And you see that they're very explicitly talking about listening to the song on repeat to make sure that on multiple voices often to make sure that the songs dominate the new charts because like they're they're deliberately participating in engagement hacking because the engagement is a validation of their fandom and of their idol. Like I'm guessing you understand that this isn't actually all about Kpop. Like this is really what happens when you have like an ungoverned reads layer on an open protocol, and like well, when there's well the governance is there, and but the governance turns out to be economics, and the economics then and the and the economics means that the um that the community doesn't match the there is governance on an open protocol, but it turns out that because the governance is not defined, the governance is economics, and the economics determines the outcome.
And that means that communities whose pattern doesn't match the dominant engagement pattern get structurally disadvantaged over the communities that do participate in like this engagement optimization, regardless of irregardless of that the protocol guarantees your rights. And the rights to choose is of you're to choose your own algorithm is only meaningful is like if the if the algorithms can actually if there's worth choosing in between them, but because they all point to this single data space, like they all converge on each other. And there is still that doesn't mean like there's still space, uh space for institutional governance, and BlackSky actually gives a very good illustration of what that can look like.
These are two recent posts by Rudy, like he um he showed trending on what was trending on um on the BlackSky feeds, the blending uh black side trending topics versus what's trending on um on Bluesky on the wider network. And I think that gives a very good insight of like okay, what does it actually mean to do create this sorts of divergence in topics and conversations because that means setting very clear boundaries and very clearly defined communities, and those communities can have their own governance and their own cultures that create it. But what we see is that when you have the implicit governance choices, they're like all the spaces that do not get um share uh uh governed by a specific culture that do not have this clearly defined boundaries, that's where you get economics determine the outcome.
And when economics determines the outcome, culture that is optimized for engagement, like K-pop dominates. Thank you. A few questions? Sure. Questions. So what do we do next? Listen to K pop. Okay, like it's very easy. Thank you. Great, we're done now. You've written a great uh post about this. Uh I think uh a lot more about governance going forward generally, and as you said, collecting funds for other entities in the network. Absolutely. Yeah, thank you. Thank you very much. Thank you.