Computable Authority

After this lesson, the reader should understand:

• 01State the claim of computable authority precisely.
• 02Identify what the discipline deliberately leaves to human judgement.
• 03Position machine-readable policy as one domain among future others.

What computability does not claim

Does computable authority argue that machines should decide?

No. It argues that when a decision is made under institutional authority — by a person or a machine — the authority applied should be identifiable, the reasoning reproducible, and the outcome verifiable. Judgement remains where the institution places it. What changes is that the placement is now explicit rather than discovered after the fact. Consider a decision where the placement matters most: an emergency child-protection intervention. An artifact can check the procedural preconditions the law sets — that the required authorisation was obtained, that the family was notified within the statutory window, that the case was reviewed by the deadline — and refuse to record the action as compliant if any is missing. It does not decide whether the child is at risk. That judgement stays with the social worker, and the artifact records it as what it is^[1]: a discretionary state, routed to a named person, with the facts they had. The machine is strict about the procedure precisely so that the person's judgement is the only unexplained element left in the record. The strongest objection is that explicit placement is not neutral. Whatever becomes computable tends to get computed: once procedural checks are automated, the pressure to automate the next layer grows, and people who review machine output tend to defer to it^[2]. And formalisation shifts power toward whoever writes the code — the old observation that, in a digital world, code functions as law^[3]. Both critiques are well founded. The answer to the first is to design the discretionary states as authored outcomes that cannot be removed without an act of authority, and to measure deference rather than assume it away. The answer to the second turns the observation around: code already acts as law inside every automated system, usually invisibly. An artifact whose rules are public, versioned, and contestable is code-as-law brought into view, where it can be argued with. The last limit is the deepest. Verifiability is not legitimacy^[4]. A rule can be identifiable, reproducible, and independently verified, and still be unjust or issued by an authority that should not hold it. Legal-rational authority rests on a belief in the legality of the rules and of those who issue them^[5]; computable authority can make that legality inspectable, but it cannot manufacture it. A perfectly verifiable unjust rule is verifiable injustice. The claim can be stated precisely. A verdict V on an action is Eval_A(a, F, t)^[6]: the institution's sealed authority A, applied to the exact action a and the proven facts F, at the moment of action t, with four outcomes — authorised, denied, escalated, or incomplete. Two absences in that expression carry the argument. The agent is not an input: who presents an action changes nothing about whether it is authorised, which is why an agent cannot argue its way to a different answer and gains nothing by disguise. And context is not a separate input: it is facts, and a fact is either proven or it does not enter the evaluation. The expression itself is not new; attribute-based access control and trust-management systems have computed decisions from several inputs for decades^[7]. What the discipline adds is the type of each input: an authority that a stranger can verify and that is the version in force at t, facts that arrive through channels the authority declares, an action bound so that the verdict cannot be reused for another, and a computation performed by the system the action would change. Escalated and incomplete are not failures in that list. They are the correct execution of a limit on machine authority, and they make a precise definition of autonomy possible: not the ability to act without asking, but the ability to act independently within a defined authority boundary, which includes recognising its edge.

• Which decisions should never be delegated to an artifact?
• Does explicitness reduce or relocate discretion?
• What legitimacy risks does computability introduce?
• Once the procedural parts of a decision are automated, what keeps the substantive judgement from following?
• Can a verifiable rule be illegitimate — and if so, what does verifiability add?

No responses yet. Be the first to answer one of the questions above.

Write the claim in your own words.

1. 01State what computable authority asserts, in three sentences.
2. 02State three things it does not assert.
3. 03Test both against a decision in your organisation.
4. 04Note where you disagree with the framing.
5. 05For one decision, mark which parts are procedure a machine could check and which are judgement that must stay with a person.