(2023-07-19) Berjon Decent Imaginaries

Robin Berjon: Decent Imaginaries. Many of us, across the online, can feel the era burn and rot away. But this twilight of the gods tells us nothing of what renewal ought to look like when morning comes. And so the question: what now?

In preparing for change, many of us talk about decentralization (or "decent" for short) but that's an abstract word.

The "decent" project is about bringing democratic governance where it hasn't been before, and it should hopefully be obvious to all that this takes more than just technology. Some technologies have been necessary to the emergence of democratic forms of course, but never sufficient.

In the same way that ecologists of all stripes are not just finding technical solutions to save the world but also popularising beguiling ideas that few of us would have imagined without them — silvopasture, farmer-managed natural regeneration, carbon architecture, microgrids — us technologists need to offer clearer, concrete, bewitching futures for us all

One way to understand computer systems in large social settings is as automated bureaucracies.

What are its rules, how accountable is it, what are your pathways to changing it — that's the stuff of democracy.

people have been asking pointed questions about how trust & safety could work, or whether it can work at all, in a decent world.

We can immediately rule out reproducing the current system in which a tiny number of companies govern permissible speech and what safety means for billions of people

This is the Google Search Fallacy: that it's actually possible for a single, unified system to organise other people's information at any significant scale in a way that is equally respectful of all and that doesn't have undesirable sanitising side-effects on the world.

Importantly, you can't support trust & safety at the instance level in a federated system with each instance doing the work... if every instance has only on the order of a hundred users in order to be tractable, that means that we need millions of instances to serve all online people, and as an admin you need to decide which ones to federate with or not. (Mastodon, Fediverse)

What's more, you can't really defederate from large instances. Email is federated but 85% of it is owned by Google.

For a democratic form of trust & safety to function, we need to distribute and to reuse work, and to create room for cooperation.

Bluesky have helpfully described composable moderation and also support subscribing to mute lists

We can also imagine shared blocklists, both for content that shouldn't just be labelled and people who shouldn't just be muted.

The complement to composable moderation at the protocol layer is polycentric governance: a system with multiple decision centres, each of which has limited and autonomous responsibilities, and operates under its own rules

For example (staying light on details):

CSAM content blocklists. Produced by frontline experts, nonprofits, law enforcement and likely applied not by people but directly by each instance

Nazis, antivax, disinformation. Could be good candidates for blocklists (of people or content) maintained by dedicated nonprofits

Code of Conduct. Multiple instances decide to agree on the same code of conduct, the drafting and evolution of which they govern together

Gambling, addiction, child-safety. Various forms of self-labelling, which is sometimes legally mandated, and of automated warnings

Blockparty services. Collaborate with others you trust in order to maintain blocklists of people you don't want to hear about

And then, because the protocol support composability, you bring your selection together at the instance and personal levels

The important part is that the workload gets distributed and shared, and that there are enough moderation sources to choose from that you don't get a sanitised world out of it.

I also want to be clear that I'm skipping over a number of challenges and overhead — I'm only outlining the general shape of it to make the case that it's possible. That doesn't mean that it's not hard.

Decently real

For decent to prevail, it has to be better, it has to be better on people's own terms, and it has to show it's better, not tell

In order to support human agency, a protocol needs to achieve two things: it needs to prevent the accumulation of power imbalances between parties (maintaining equality) and it needs to make it easy for users to cooperate in building the the rules they want for how the protocol's operation affects them

Maintaining equality, liberating people is, to the extent possible, supported by self-certifying protocols. A non-technical way of understanding their value is that they provide freedom from authority.

Then people need to be able to get that freedom to work for them.

They need the freedom to organise, in a variety of polities, so that they can cooperate towards their goals and solve problems that they could not begin to touch alone.

A self-certifying protocol without a cooperation layer is (exactly) like git without GitHub

Cooperation is "the other invisible hand". There are good reasons to believe that it is better to build social software

on democratic principles.

There's great current work around cooperation.

Between a massively online population, new primitives like CRDTs and content-addressed databases (that can be safely written to in parallel by anyone), the emergence of post-Ponzi, low-emissions blockchains with diversified approaches to consensus, and all the work on automating the bureaucracy of governance with DAOs we can legitimately hope to see cooperation emerge as this decade's defining aspect online as well as offline.

Self-certifying protocols are also making great strides: IPFS, Peergos, ATProto, Nostr, and more.

The entire project of decent systems today is figuring out how to assemble these parts into a technosocial system that provides freedom from authoritarian software and that makes cooperation easy, even pleasant.

Technology cannot fix the fact that freedom is an endless meeting.