Capability Is Not Authority
Synthocracy Institute · Method / Working Paper · A normative argument (register B — argued, not empirical) · by Martin Novak · July 2026
Thesis. Capability is not authority. A system being accurate, efficient, or statistically right does not give it — or those who wield it — the right to decide. Being able to make a decision well is not the same as having the standing to make it. Synthocracy is what happens when we forget the difference.
Key points
- Capability is a fact about a system. Authority is a relationship with the people a decision affects.
- “Being right” and “having the right” are different in kind: no amount of accuracy crosses, by itself, from what is to what ought to be done.
- The confusion is natural and documented: people readily confer authority on systems that seem to perform well — a tendency researchers now call algority.
- What actually confers authority is legitimacy: authorisation by the governed, constraint by law, and answerability to those affected.
- Almost every other problem of synthocracy — the ceremonial human, platform power, the missing off switch — is a downstream effect of this one confusion.
The claim
We increasingly let AI systems make, or all-but-make, consequential decisions because they are capable: fast, consistent, and often more accurate than the humans they replace. That capability is real. But it answers a different question than the one that matters.
Capability answers: can this decision be made well? Authority answers: who has the right to make it?
These come apart completely. A weather model can predict a storm with great accuracy and has no authority to order an evacuation. A scoring system can rank loan applicants precisely and has no standing to decide who deserves credit. The competence is one thing; the right to decide is another. Synthocracy advances precisely by letting the first quietly stand in for the second.
Being right is not the same as having the right
The deepest reason is old. David Hume observed that you cannot derive an ought from an is: a description of how the world is does not, by itself, establish what should be done about it. A model is, at its core, a very sophisticated description of what is likely. The leap from “this is what will probably happen” to “this is therefore what we should do” is not a computation. It is a judgment — and judgments about what should happen require authority, not just accuracy.
This is clearest wherever a decision involves a trade-off. How much should a system weigh a false positive against a false negative — a wrongly denied benefit against a wrongly granted one? How should it balance efficiency against mercy, or the average case against the vulnerable exception? These are not facts the model discovers. They are value choices about whom to protect and whom to risk — and a system optimising a metric has no standing to make them, however well it optimises.
A machine can be right. Only a person — or an institution answerable to those it affects — can have the right.
This is not a new idea — and that matters
Honesty is part of the method, so we name the lineage. The distinction between capability and authority runs through centuries of thought, and a serious contemporary literature is now applying it to AI.
- Legitimacy, not competence, is the ground of authority. Max Weber distinguished the power to compel from the legitimate authority to rule. Joseph Raz argued that legitimate authority must be justified by how well it serves those subject to it — not by its raw ability to command. Neither is satisfied by being right.
- The technocratic error. The critique of technocracy — that expertise and efficiency are not a democratic mandate — is exactly the claim that capability is not authority, made about experts. AI inherits it, and sharpens it.
- Contemporary work. Philosopher Seth Lazar has pressed the question of whether AI labs are democratically authorised to build systems of such power, and whether inscrutable systems can meet basic criteria of legitimacy at all. Legal scholars argue that the legitimacy of algorithmic power depends on whether it is authorised by the governed, constrained by law, and answerable to those it affects. And empirical researchers have named algority — the human propensity to confer epistemic authority on algorithms where we would otherwise defer to human experts.
Our contribution is not to discover the distinction. It is to give it a hard, portable name — capability is not authority — and to show that it is the pillar under everything else the Institute studies.
Where the confusion comes from
If the distinction is so basic, why is it so easily lost? Three reasons.
Performance reads as permission. When a system is usually right, deferring to it feels not just easy but responsible. Accuracy quietly launders into authority. Algority is the name for this reflex.
Institutions borrow legitimacy from accuracy. An agency or a company can promote a decision as “data-driven” and “objective,” using the system’s competence to place the outcome beyond argument. The machine’s capability becomes a shield against contestation.
Opacity removes the check. Traditional expertise can, in principle, be questioned and cross-examined. Many AI systems cannot: their workings are inscrutable, often proprietary. So we are asked to treat as authoritative an “expert” we are not even allowed to interrogate.
What actually confers authority
If not capability, then what? Authority is not a property a decider has; it is a relationship a decider stands in with the people affected. It is conferred, not computed. Its sources are familiar:
- Authorisation — the power to decide has been granted, ultimately, by those subject to it, or by institutions accountable to them.
- Constraint by law — the decision operates within rules that bind the decider, not only the governed.
- Answerability — someone can be asked to justify the decision, and can be held to account if it is wrong.
- Contestability — those affected can challenge it and be heard.
A system can be enormously capable and possess none of these. When it does not, it has no authority — only the appearance of it, borrowed from its competence. Some law is beginning to insist on the distinction directly: Texas’s 2025 healthcare AI rules, for instance, require human oversight that preserves decision-making authority, not merely human involvement.
Why this is the pillar under synthocracy
Nearly every problem the Institute studies is this confusion, downstream.
The ceremonial human is what remains when a capable system has effectively taken the authority, leaving a person to hold only the accountability. Platform power is capability at the infrastructure level being treated as a right to set the rules. Admissibility is, in the end, the discipline of refusing to let capability alone stand in for authority — of insisting that a system’s right to act be established, on a record, before its ability to act is deployed.
Synthocracy, at its root, is a single substitution repeated everywhere: because a system can decide well, we let it decide — and quietly stop asking whether it should.
The test
To catch the substitution in a specific case, ask:
- Is this system being trusted to decide, or only to inform?
- If it decides, what authorised it to — beyond the fact that it performs well?
- Who, exactly, can be held to account for the outcome?
- Can those affected contest it and reach someone with the standing to change it?
- Are value trade-offs (whom to favour, whom to risk) being made by the system as if they were facts?
Where authority rests on capability alone, it rests on nothing — and that is the finding.
What follows
None of this is an argument against capable systems. Accuracy is genuinely valuable, and a capable model has earned a real place: at the table, informing the decision. What it has not earned is the gavel. The task is not to reject capability but to keep authority where it belongs — with humans and institutions who can be authorised, constrained, and held to account.
→ See What Is Synthocracy?, The Ceremonial Human, and Admissibility. To test one system, use The Ten Questions.
FAQ
What does “capability is not authority” mean? It means that a system being accurate or effective does not give it the right to decide. Capability is the ability to make a decision well; authority is the legitimate standing to make it. They are different in kind.
Why isn’t being right the same as having the right? Because you cannot derive what ought to be done from what is the case (Hume’s is/ought gap). Decisions involve value trade-offs that are not facts a model discovers, so accuracy alone cannot authorise them.
What actually gives a decision-maker authority? Legitimacy: authorisation by those affected or by accountable institutions, constraint by law, answerability, and contestability. A system can be highly capable and have none of these.
Is this an argument against using AI? No. Capable systems can rightly inform decisions. The argument is that authority to decide must remain with humans and institutions that can be held to account — capability earns a seat at the table, not the gavel.
How does this relate to synthocracy? Synthocracy is largely this one confusion repeated everywhere: because a system can decide well, we let it decide, and stop asking whether it should. Capability quietly stands in for authority.
Martin Novak is the founder of the Synthocracy Institute, an independent research institute studying how decision-making power shifts through AI systems. Warsaw, operating internationally. This argument draws on Hume, Weber, and Raz, and on contemporary work by Seth Lazar and others on the legitimacy and authorisation of algorithmic power.
