Censorship on Bluesky

Bluesky has been in the news. They’ve opened to the public, removing their invitation system. But I want to talk about censorship. Specifically: Bluesky actually has two censorship regimes, but they only talk about one. The second censorship regime arrogates decisions about speech to a handful of Americans, away from democratic government. I don’t want to fall into that trap, again.

Bluesky Federation

Here is the federation architecture as described in a Bluesky blog post.

It’s much more complicated than the usual Fediverse structure. The equivalent in the Fediverse is just PDS’s (Personal Data Servers) that talk to each other, full stop. The problem is, this doesn’t scale. If an article on a small server goes viral, that server will be instantly overloaded.

Bluesky solves this with “relays” (formerly called “big graph servers” or “BGS”). These are large servers that hold effectively all content from the entire social network. That’s far too much data for a hobbyist, or even charitable organisation, to manage reliably. Bluesky envisions that there would only be a handful of these relays at most.

This means Bluesky has undone the chief benefit of federation! It seems the major motivation of people moving from Twitter to federated social media is the censorship policies. Interestingly, while open source development has roots in a libertarian anything-goes ideology, most people were turned off Twitter because it started allowing more content, specifically far-right content. But there’s no contradiction here. People want the freedom to establish their own regimes, to control their own experience.

The Official Censorship: Labelers

Bluesky has an answer to this: Labelers. The appropriately-named Labelers only attach labels to content. These labels of course include classifying abusive or offensive content. They can do so in a detailed way, so you can choose whether you want to be shielded from sexual content, or violent content, or both. They can do more than just censor, for example helping you keep political discussion separate from self-help.

This is how Bluesky gives you the freedom to curate your own feed to your tastes. The Labelers broadcast their decisions, and the Feed Generators use that to filter their output. This is the content you eventually see in your app. Crucially, maintaining a Labeler is much easier than maintaining a Relay. So if you disagree with a particular censorship regime, it should be possible to establish your own.

There’s a problem here: the bad content is still out there. There are clearly categories of material, such as CSAM, that must not be stored in the system in any way, and certainly should not be available to bad actors who wish to access it. This architecture doesn’t address that. After all, if I’m free to add my own Labelers and Feed Generators, those systems must have access to the unfiltered content. But if I am a bad actor actively trying to access that bad content, the system doesn’t prevent me doing so.

The Hidden Censorship: Relays

So sooner or later, the system must have another component, a filter that prevents bad content making it to the Relay in the first place.

The original architecture diagram is coherent enough. When there is a popular outcry that extremist opinions are being expressed online, Bluesky can simply pass responsibility to their network of Labelers. Hopefully the Labelers will do their job, and there won’t be anything offensive to fuel the outcry in the first place. Some isolated voices may complain that the bad content is still there. In that case Bluesky can point to the architecture and lament that unfortunately we can’t prevent people attempting to publish this stuff, but rest assured it does get filtered out before people see it.

But once you add a Filter before content gets to the Relay, that argument is undermined. It manifestly is possible to do that for CSAM, so why not do the same for Islamist terrorism? Why not for anti-democratic rabble-rousing? Why not protect children from gender ideology? Suddenly everything is really messy.

That’s fine. Censorship is indeed messy. You have a public discussion, a compromise is reached, reform is regularly proposed. We’ve been through this for broadcast media many times before. This is the kind of question that democracy can handle.

But the question I want to ask is: whose democracy?

The Peculiar Bias of Anglo Censorship

There is a narrow-minded view, especially in the English-speaking world, that censorship regimes fall along a single continuum. The question is not what is censored, but how much is censored. It is assumed that any work censored in a liberal regime would also be filtered in a strict regime. In this implicit but rarely-expressed view, CSAM lies on the far end of the spectrum and is always censored. Pornography is next in line, and falls in various degrees from being hidden to being banned outright. Profanity follows, and these days is freely allowed except in children’s programming. Only illiberal regimes begin censoring political opinions.

I believe these assumptions are partly due to the English puritan movement, which of course was heavily represented among the original American settlers and influenced that culture. This movement had peculiar concerns. They believed that good Christians had a responsibility to concentrate exclusively on the spiritual realm, and any contact with the profane degraded the soul. So they were obsessed with eliminating sexual temptation of all kinds. However, since their views were considered heretical by the highly profane kings of the time, they demanded the right to express their spiritual views freely. So they wound up believing in a strange censorship regime that is obsessed with policing sex, but treats even the most outrageous political opinion as sacrosanct. In America, sex was finally allowed through the censorship system only via the argument that sex itself is also a form of political expression.

My Experience Of Censorship

Although I grew up in English-speaking countries, I now live in Germany. There is a very different censorship culture here. During the early 2000’s I worked for amazon.de. One of my duties was maintaining a system that filtered “Mein Kampf” and other works contrary to the Basic Law out of the catalogue. These works, and worse, are freely available in America. On the other hand I also helped launch Amazon’s magazine catalogue in Germany. Soft-porn magazines were a popular component of that catalogue. There is no way these would be allowed on the American site.

For myself, I have sympathy with revulsion at the systematic commercial exploitation of womens’ bodies, and I have sympathy with arguments that “Mein Kampf” must be studied in order to prevent a repetition of the Holocaust. Nevertheless, I’d have to say Germans’ priorities when building a censorship regime are the right way around. Whereas the American censorship regime is nuts.

And this is what bothers me about Bluesky. There is implicitly a censorship regime being implemented here, and it must necessarily be centralised. There must be a single set of decisions affecting everyone in the world. Those decisions are being made by an American company bound by American laws, which in turn are decided by the American people participating in American democracy.

That’s nice, but I don’t get a vote. I’m very happy with the censorship under which I live, and I intend to use my vote to preserve it. But Bluesky seems intent on undermining the effectiveness of my vote. Indeed, undermining the right of almost everyone in the world to decide not only what should be forbidden, but also what should be allowed.

The Role Of The USA

Part of the problem here is the “single continuum” assumption that underlies English-language discussion of censorship. Due to reforms in the 70s, America now finds itself right on the extreme liberal end of the imagined spectrum. All kinds of revolting pornography are allowed! The only exception is CSAM. So surely, if any nation feels there is a problem with social media content, all they need to do is add a Labeler and mandate its use in all Feed Generators. The only necessary action is ratcheting up the strictness. There’s nowhere to loosen it down to except CSAM, and we’re all opposed to that.

This argument is false. The best example is the demise of Switter. Again, I sympathise with the argument that sex work is inseparable from human trafficking. But it must be up to each democracy to decide this on its own. In Germany sex work is perfectly legal, while in America it is almost entirely illegal. This is at least one area where accepting American rules as a baseline would neutralise any ability for sovereign nations to implement their own reforms.

If this argument only concerned sex work, this case would be hard to make. But that’s only an example. I think the principle is clear enough that it is worth defending now. The threat of a second Trump presidency, sooner or later, should also concentrate minds.

Comparing Fediverse and Bluesky

The Fediverse approach addresses this problem very clearly. There is no single central store of data that would require a global censorship regime. Anyone is free to run their instance within their own country’s borders, and therefore subject only to their own country’s laws. Conflicts between laws can be dealt with by the usual process of blocking individuals or entire instances. And if posts go viral, sorry, servers will go down.

The Bluesky story here is much more difficult. One potential solution is to reassure everyone that there will be at least one Relay following national laws available in each country, or at least each democratic country. That doesn’t sound too outrageous, there are only about 200 countries.

But it’s not enough for the Bluesky company to simply establish all those servers itself. They are still an American company, and sooner or later would come under pressure to remove illegal content globally. Somehow there would need to be a Bluesky ecosystem robust enough to ensure the survival of all these Relays as independent services. Possible, but we’ve seen so much consolidation and centralisation in the IT industry, it’s hard to believe that the Bluesky ecosystem would be an exception.

And it doesn’t help at all if Bluesky won’t even acknowledge that this is part of their architecture.

The Conspiracy Theory

It’s clear that there is currently a global movement to wrest control from gigantic IT corporations back to national governments. But it’s also clear that the huge profits those corporations expect are just the manifestation of global monopoly power that comes from centralisation.

In that context, I’m tempted to wonder if there is a deliberate strategy by Bluesky to avoid confronting the issue of censorship head-on. It seems significant that the term “Big Graph Server”, which is an excellent description, was replaced by the far less threatening term “Relay”. Who could find anything sinister about a beefed-up Cisco router?

For now, I think malicious intent seems unlikely. Rather, I think Bluesky are subject to three cultural assumptions buried too deep in Californian IT culture to think past:

  1. The USA’s censorship regime is so liberal, it effectively counts as no censorship at all
  2. Censorship of political opinions is undemocratic and can always be opposed
  3. Centralisation of social media somewhere is architecturally inevitable

So it’s up to the rest of us to see past the optimistic vision being presented to us, and question the assumptions the designers don’t even know they’re making.

Leave a Reply

Your email address will not be published. Required fields are marked *