Semantic Sovereignty

Premise

Computers are not appliances.

An appliance performs a bounded physical task. A computer is a general-purpose symbolic machine. It stores, transforms, filters, produces, and transmits meaning.

Because modern life is increasingly mediated by computation, software does more than help us perform tasks. It structures the environment in which we think, communicate, remember, create, associate, work, and understand the world.

Software is therefore not merely a product.

It is part of culture.

It is part of infrastructure.

It is part of the human environment.

The central principle

A person should remain meaningfully sovereign over the computational systems through which they experience and act upon the world.

Sovereignty does not mean that everybody must be a programmer.

It means that computational systems should preserve the user's ability to understand them, leave them, modify them, replace them, connect them to other systems, and decide what role they are permitted to play in one's life.

A society in which people cannot meaningfully control their computational environment is not fully autonomous, regardless of how convenient that environment may be.

The sovereignty at issue is semantic because what is at stake is not the ownership of a device or the licence on a file. It is authority over the systems through which meaning reaches us, and through which our meaning reaches everyone else.

Why free software is not enough

This document owes almost everything to the free software movement. The four freedoms, to run a program, to study it, to modify it, and to share it, were the first serious claim that software is a political object and not only an industrial one. Nothing here replaces them.

But they were written for a world in which the program was a thing you had.

What it argues for

The four freedoms attach to an artifact. They describe your relationship to a copy of a program, sitting on a disk you own. That is no longer where the important software lives, and it is no longer where power over you is exercised.

The program runs on someone else's computer. You cannot study a binary you never receive, and modifying your client does not touch the system that actually decides.

The program is a set of weights, and source availability confers no legibility. You can hold every parameter of a model and still be unable to say why it did what it did. "Open" stops meaning "inspectable" at precisely the moment inspection begins to matter.

The program is free and you are still captive, because what holds you is not the code. It is your correspondents, your history, your identity, your accumulated context. Freedom over the artifact was never freedom from the dependency.

The program is free and you cannot run it, because your hardware attests to a remote party what it is executing, and the service refuses to speak to a machine under your control. Tivoization was answered in the licence. Attestation reintroduced it as a protocol, where no licence reaches.

The program is free and its defaults are hostile. Nothing in the four freedoms addresses the enormous, quiet power of the setting that nobody changes.

Each of these is a case in which every freedom is formally satisfied and the person is not free. That is the signature of a framework whose unit of analysis has drifted away from the thing it meant to protect.

How it pursues its goals

The movement's instrument is the copyright licence. Copyleft turned copyright against itself, which was a brilliant manoeuvre and remains one. But a licence can only bind what copyright reaches, and it can only fire on the act copyright governs, which is distribution.

Software is no longer principally distributed. It is served. The dominant delivery mode of the era falls outside the trigger entirely, and no clause can be drafted to reach it, because the company genuinely is not giving you a copy.

The licensing war itself was won by permissiveness. The result is an ecosystem in which the largest proprietary platforms ever built are assembled out of freely given work. The code is free. The stack is not. The freedom of the parts did not propagate to the whole.

What leverage copyleft retained is now being withdrawn by the people holding the copyrights. Contributor agreements collect the rights, and the licence is changed once the userbase is captured.

Underneath all of it, the commons is maintained by unpaid labour that the wealthiest firms on earth consume as a free input, until one exhausted maintainer accepts help from a stranger and ships a backdoor to the world.

Open source made this worse on purpose. It was, by its own account, free software with the ethics removed so that it could be sold to business. It succeeded completely. It also disarmed the movement of the only vocabulary, freedom and power and autonomy, adequate to the fight that came next.

None of this is a failure of nerve. It is a category error that only became visible at scale.

Free software is a licensing movement, and licensing is no longer where power lives.

Power lives in protocols, formats, identity, defaults, attestation, and the terms on which one system is permitted to speak to another. Almost none of that is copyright.

What follows is the same argument with the unit of analysis moved: from the artifact to the relationship, from the licence to the architecture, and from a question about software to a question about power.

Five principles

1. Legibility

Systems that govern important parts of our lives should be understandable.

A user need not understand every line of code, just as a citizen need not understand every detail of civil engineering. But the rules of the system should be discoverable, describable, and open to independent examination.

Invisible mechanisms produce invisible power.

2. Agency

The computer should act for the user, rather than the user acting for the computer.

Users should be able to determine how their machines behave instead of being forced into workflows chosen by vendors, platforms, advertisers, or remote administrators.

Defaults are inevitable.

Captivity is not.

3. Reversibility

A technological decision should not unnecessarily become a permanent dependency.

Users should be able to export their data, change software, undo updates, remove unwanted features, and migrate between services without losing their digital lives.

A system that is easy to enter and difficult to leave is not merely convenient software. It is an institution exercising power.

4. Interoperability

Computers should form an open technological commons rather than a collection of private kingdoms.

People should be able to communicate across software boundaries, move information between tools, implement compatible clients, and create new systems without requiring permission from an incumbent platform.

Protocols should matter more than platforms.

5. Possibility

The machine should remain a machine of possibilities.

General-purpose computers are historically extraordinary because they can become things their designers never anticipated.

A healthy computing culture preserves this openness.

A computer whose legitimate functions are restricted to those approved by its manufacturer is becoming something less than a computer.

Freedom without programmer exceptionalism

Semantic Sovereignty does not demand that everyone compile their own kernel.

Most people do not repair their own cars, build their own houses, or manufacture their own medicine.

The important question is whether expertise exists within an open social structure.

You may not personally inspect the software you use. Someone else must nevertheless be permitted to inspect it.

You may not modify it. Someone else must be permitted to modify it for you.

You may never leave a platform. You should nevertheless possess the ability to leave it.

Freedom is valuable even when it is delegated.

This is also where the older movement mistook its own constituency. A freedom that only a programmer can exercise will, in the end, only be defended by programmers, and there are not enough of them to defend it.

Convenience is not the enemy

The goal is not to make technology difficult.

Good systems should be convenient, beautiful, reliable, and accessible.

The false choice is between convenience and freedom.

The real distinction is between convenience that serves the user and convenience that purchases dependence.

The ideal computer requires little knowledge from someone who wants simplicity, while imposing few restrictions on someone who wants control.

Architecture is politics

Every computational system contains assumptions about how humans should behave.

A social network decides what constitutes a relationship.

A search engine decides what knowledge becomes visible.

A recommendation system decides what culture reaches us.

A navigation system changes how we perceive geography.

A productivity platform changes how organizations conceive of work.

A smartphone operating system determines what forms of software are permissible.

These decisions eventually cease to feel like decisions. They become reality. A choice made by a product manager in one year is, a decade later, simply how the world is, and there is no longer anybody who remembers that it was chosen.

Power today is exercised not only through laws and ownership, but through interfaces, protocols, algorithms, file formats, permissions, defaults, APIs, and standards. These are not neutral technical details. They are governing instruments that happen to be written in a technical vocabulary, and that vocabulary is most of what protects them from scrutiny.

Technological freedom should therefore be treated as a component of personal and civic freedom.

The question to ask of a system is not merely:

Does it work?

but:

What kind of person does this system encourage me to become, what forms of action does it make possible, and who ultimately has authority over it?

What this requires

Principles that demand nothing are decoration. Each of the following follows from the five above, and each is addressed to somebody in particular.

1. Consequential systems must be describable. Anyone subject to an automated decision that materially affects them should be able to obtain an account of the rule that produced it, in terms they can act on.

2. Hardware you own must run software you choose. And no service may condition access on proof that you have declined to exercise that right. Attestation used to enforce a vendor's preferences against the owner of the machine is not a security measure. It is a leash with a certificate.

3. Exit must be an operation, not a policy. A right to export is worth nothing unless the export is complete, documented, and in a form a competitor could actually import. Anything less is a compliance artifact.

4. Systems that have become infrastructure must be speakable-to. Where a platform has taken on the role of public infrastructure, third parties must be free to implement compatible clients and to interoperate with it without asking, and the law should protect them when they do rather than prosecute them for it.

5. General-purpose computing must remain lawful and practical. No manufacturer, platform, or state should hold the authority to decide which programs a person may run on a machine they own.

What is to be done

Individual choice is not sufficient here, and its insufficiency is part of the design. A person cannot consume their way out of an architecture. Saying so plainly is more useful than pretending otherwise.

Still, some things follow.

Build on protocols rather than platforms, and publish the protocol before the client.

Prefer systems you could leave, and leave one occasionally, to confirm that you still can.

Pay for the commons you depend on, in money or in maintenance, because volunteer exhaustion is now a security posture for the entire world.

Refuse the framing that this is a matter of taste. It is a question about who governs the conditions of thought, and it should be argued in those terms, in public, by people who are not programmers.

The aspiration

We want computers that are quiet when ignored, powerful when examined, comprehensible when questioned, modifiable when necessary, and replaceable when something better appears.

Technology should expand the space of human possibility rather than quietly defining its boundaries.

The computer should remain a tool for constructing worlds, not merely a terminal through which someone else's world is delivered.