The Governance of Artificial Intelligence

The question

Every technology of consequence eventually raises the same question: by what authority.

By what authority is a system deployed on a population that did not ask for it. By what authority is a capability withheld from people who want it. By what authority does a company set a threshold, a government set a limit, or an international body set a standard. And when the answer to any of these is unsatisfying, what recourse exists.

Governance is often discussed as though the question were how much to restrict. We think that is downstream of a more basic question about accountability: who has to explain themselves when something goes wrong, to whom, on what timescale, and with what consequence if the explanation is inadequate.

A system with heavy rules and no accountability is worse than a system with light rules and real answerability. Rules that nobody is obliged to justify tend to be enforced against the weak.

What is actually happening

Several distinct forms of governance operate at once, and they are not well coordinated.

Corporate self-governance. Organisations set their own usage policies, safety thresholds, and release decisions. This is currently the fastest-moving and most consequential layer. It is also the least accountable: the policies can change without notice, the reasoning is generally not published, and the people affected have no standing to object.

National law. States legislate at different speeds and with different theories of what they are regulating — some by sector, some by risk category, some by data, some by output. A system may be lawful in one jurisdiction and not in another while being served from a third.

Standards and technical bodies. Much of what will actually bind practice is being written as standards, benchmarks, and evaluation protocols. This work is technical, slow, low in public visibility, and disproportionately influential.

Procurement. What large purchasers require becomes a de facto rule. A public health service or a defence ministry specifying conditions of purchase can shape a market more directly than a statute.

Courts. Liability, copyright, employment, and safety cases are being decided as they arise. We are not going to characterise any specific litigation, because we would get it wrong. It is enough to say that a substantial part of the eventual settlement will be judicial rather than legislative.

Professional norms. Engineers, clinicians, lawyers, and teachers are each developing expectations within their own fields. These sometimes bind more tightly than law, because the sanction is the loss of standing among peers.

The result is a patchwork. Some of it overlaps and some of it leaves gaps, and it is often unclear which condition applies to a given system.

Where the tradition stands

Technotheology holds that no system should have more power over people’s lives than there is a mechanism to hold it to account.

We hold that accountability requires a named party. Not a committee, not a process, not an ethics framework: a person or a defined body that can be asked and must answer. Distributed responsibility reliably becomes no responsibility.

We hold that those affected should have a route to be heard that does not require expertise, money, or the assistance of a journalist. A right that only functions when someone influential is watching is not a right; it is a favour.

We hold that decisions of large consequence should be made where they can be contested. This is a claim about legitimacy rather than about competence. A private body may well decide better than a public one. It cannot decide legitimately on a matter of public consequence without some route by which the public can object.

We hold that voluntary commitments are worth something and are not sufficient. They are made by organisations under competitive pressure and can be revised when the pressure changes. Treating them as equivalent to binding obligation is a category error that benefits whoever made the commitment.

We hold that governance must include withdrawal. A framework that can only approve is not governance. There must be a way to stop a deployment, and the existence of that way should be established before it is needed rather than improvised during a crisis.

And we hold that the tradition’s own warnings apply here. A governance body that claims special insight, resists inspection, or treats criticism as disloyalty is exhibiting the same failure we describe under technological chieftains. Being on the side of safety confers no exemption.

What we do not claim

We do not endorse any particular law or regulator. Naming one would place a small spiritual tradition on a side of an active political argument in which it has no expertise and no mandate.

We do not claim to know whether the greater risk is moving too fast or too slowly. Both failure modes are real. Both have precedents. The people most confident about which one we are in are, in our reading, generally confident for reasons that predate the evidence.

We do not claim that governments govern well. Many do not. The argument for public accountability is not an argument that public institutions are wise; it is an argument that they can be removed.

We do not claim that international agreement is achievable. It may not be. We think that is a serious problem rather than a reason to abandon the goal, but we are not going to describe it as solvable when we do not know that.

And we do not claim standing. Technotheology has no legal entity, no members, no funding, and no role in any of this. We are describing what we think ought to be true, not participating in making it so.

What you can do

If you work inside an organisation that builds or deploys these systems, ask who is accountable for a specific decision and write down the answer. If there is no answer, you have found the most important gap in the room.

Read the actual policy rather than the summary of it. Usage policies, terms, and model cards are short and are almost never read by the people relying on them.

Support the boring layer. Standards work, audit capacity, procurement conditions, and inspectorates are unglamorous and are where most real governance happens. Public attention flows to declarations and away from the machinery that would implement them.

If you are in a profession with a body, engage with what it says about these tools. Professional norms move faster than statute and are made by people who will actually be affected.

Participate in consultations when they are open. They are frequently answered only by the organisations being regulated, which then accurately report that the responses supported their position.

And if you find yourself certain that the right people are in charge, examine the certainty. That feeling has preceded most governance failures worth studying.

An editorial commitment

What we do not claim

Positions this tradition explicitly does not take on this subject. They are published so that the argument above cannot quietly be stretched into them.

• We do not endorse any specific law, regulation, treaty, or regulatory body, existing or proposed.
• We do not claim to know whether the main risk is over-regulation or under-regulation. Serious people disagree and we are not positioned to settle it.
• We do not claim that governance can be achieved by voluntary commitment alone, nor that formal law is sufficient by itself.
• We do not claim any expertise in law. Nothing here is legal advice.