The Policy Problem

Modern institutions rely on policies as the primary mechanism for expressing authority. However, policies were designed for human interpretation rather than machine execution. As artificial intelligence systems increasingly participate in institutional decisions, a fundamental challenge emerges: how can authority be represented in a form that machines can execute and independent parties can verify?

After this lesson, the reader should understand:

• 01Why traditional policy documents depend on human interpretation.
• 02Why AI systems create a new requirement for machine-readable authority.
• 03The difference between information and executable authority.
• 04The foundations of computable authority.

The EU AI Act

If an AI system must determine whether an organization complies with the EU AI Act, what representation of the regulation does the machine require?

The published Act is a 140-page instrument written for lawyers and regulators. A machine cannot act on it: the risk classifications are defined in prose, obligations are scattered across articles and annexes, and the conditions that trigger them are expressed as legal tests rather than evaluable predicates. What the machine requires is a representation in which each classification is a named definition, each obligation is a rule bound to that definition, and each trigger is a condition that can be evaluated against recorded facts — issued under a version, so a determination made today can be replayed unchanged in five years.

• Can a policy ever be fully objective?
• Which parts of authority require human judgement?
• Should every regulation have a machine-readable representation?
• Who validates the machine interpretation?

Analyse an existing policy from your own organisation.

1. 01Identify the rules — the operative must, may, and must-not statements.
2. 02Identify the definitions the rules depend on.
3. 03Identify the exceptions and the conditions that activate them.
4. 04Identify the decision points where an outcome is produced.
5. 05Identify the unresolved ambiguity, and note who would resolve it.

• Legal theory — the interpretive turn and the limits of literal reading.
• AI governance literature — accountability under automated decision-making.
• Formal methods — specification languages and deterministic evaluation.
• Knowledge representation — ontologies, defeasible rules, and legal reasoning.
• Digital government research — rules as code programmes and their findings.

The document is the dominant carrier of institutional authority, and it is a carrier with no defined behaviour. This lecture examines what a document can and cannot do as a governance instrument, and why format modernisation — PDF to HTML to XML — does not by itself close the gap.

After this lesson, the reader should understand:

• 01Distinguish a document's content from its behaviour.
• 02Explain why digitisation is not the same as computability.
• 03Identify the properties a governance artifact needs that a document lacks.

A procurement threshold

A procurement policy states that contracts above a threshold require competitive tender. Why can two officers reading the same clause reach different outcomes?

The clause omits the inputs. It does not state whether the threshold applies per contract, per supplier, or per financial year; whether the figure is gross or net; or what happens to an option to extend. Each officer supplies the missing inputs from experience, and each supplies them differently. The disagreement is not about the rule — it is about the unstated interface to the rule, which a document has no way to declare.

• Is a well-drafted document simply an under-specified program?
• What is lost when prose is replaced with structure?
• Where should the authoritative text live once an artifact exists?

Take one clause and specify its interface.

1. 01Write the clause verbatim.
2. 02List every input it silently assumes.
3. 03State the type and source of each input.
4. 04Note which assumptions the drafter would need to confirm.

• Document engineering and legal XML standards.
• Rules as code — early government pilots and their limits.
• Specification theory — completeness and under-determination.

Machine-readability is often treated as a downgrade of legal text. This lecture argues the opposite: the machine-readable form is a second, disciplined expression of the same authority, and the two forms must be governed together.

After this lesson, the reader should understand:

• 01Separate the authority itself from its human and machine expressions.
• 02Explain why parallel expressions require an explicit precedence rule.
• 03Describe how ambiguity surfaces when a policy is expressed twice.

Tax guidance and its calculator

When an official calculator and the published guidance disagree, which one is the authority?

Most institutions have never answered this question, which is why calculators are published with disclaimers. Under computable authority the question is answered before release: either the artifact is authoritative and the prose is explanatory, or the prose is authoritative and the artifact is a compiled representation whose divergence is a defect to be corrected under version control. Silence on precedence is itself a governance failure.

• Should the artifact ever be the primary legal instrument?
• What review process should a compiled artifact pass before issue?
• How is divergence between the two forms detected?

Express one rule in both forms.

1. 01Write the rule as an institution would publish it.
2. 02Write the same rule as conditions and an outcome.
3. 03List every difference in meaning between the two.
4. 04Decide which form governs, and record why.

• Legisprudence and drafting conventions.
• Bidirectional traceability in requirements engineering.
• Case law on reliance upon official automated tools.

Interpretation has always been unrecorded, and institutions absorbed the resulting variance because human throughput was low. Autonomous systems remove that slack. This lecture examines why the authority layer becomes load-bearing precisely when execution becomes machine-speed.

After this lesson, the reader should understand:

• 01Explain why unrecorded interpretation scales into systemic risk.
• 02Distinguish model alignment from authority binding.
• 03State what an institution must be able to prove after an automated decision.

An agent acting under delegated authority

An AI agent approves ten thousand cases in a week. What must the institution be able to show a regulator?

Not the model, and not the prompt. The institution must show which authority governed each approval, which version of it applied, which facts were evaluated, and that the same evaluation reproduces the same outcome today. A model can be aligned to a policy and still leave nothing to inspect afterwards. Authority binding is a separate requirement from model behaviour, and only the former is verifiable by a party that did not run the system.

• Is prompting a policy into a model a form of policy compilation?
• Who is accountable when an agent misapplies an artifact correctly compiled?
• What is the minimum record an automated decision must leave?

Draft the evidence pack for one automated decision.

1. 01Name the authority and its version.
2. 02List the facts evaluated and their source.
3. 03State the outcome and the path that produced it.
4. 04Identify who could verify this without access to your systems.

• AI governance — auditability and the right to explanation.
• Safety engineering — the role of the operating envelope.
• Provenance research — verifiable records of computation.