HUMAN AGENCY
When the machine can predict you, rank you, filter you, speak for you, and decide which door opens—what remains yours?
This section rejects both machine panic and machine worship. Capability is real. Capability is not sovereignty.
NO JUDGMENT ENGINE
A machine may process data. It may not become a tribunal of the person.
No judgment whatsoever of human worth, identity, belief, motive, or capacity to belong. No model, score, institution, or operator inherits authority over the inward forum.
Systems may calculate, sort, predict, recommend, or flag. Those operations must never be laundered into a final verdict on the person. Consequential decisions require inspectable reasons, a named accountable human decision-maker, a path to challenge and correct the record, portability, and a real Exit.
No hidden risk score, goodness score, loyalty score, purity score, readiness score, credibility score, or behavioral profile may decide whether a person gets to think, speak, participate, inspect, disagree, or leave.
QUICK ANSWER
In one sentence
The Human Agency Test: twelve questions for cognitive liberty, algorithmic power, consent, appeal, human review, portability, Exit, and accountability.
THE HUMAN AGENCY TEST
Twelve questions for any consequential automated system.
This is a diagnostic framework, not a legal certification. Laws, regulations, and technical practices change by jurisdiction and time; current legal claims should be verified against current authoritative sources.
EVIDENCE CEILING
These questions state governance demands. They do not prove that a particular law, contract, vendor, or institution already grants the right. A “yes” needs a current source and a working path: notice that arrives, reasons that identify the decision, a reviewer with authority, a correction that propagates, data that exports, and an Exit that can actually be used.
Can the person understand the meaningful criteria or categories used?
Can the person receive an intelligible reason for a consequential outcome?
Can the person appeal or challenge the outcome?
Can inaccurate data or assumptions be corrected?
Is consent meaningful, specific, and separable from unrelated access?
Can an accountable human review a consequential decision rather than merely repeat the score?
Is it clear what data enters, what is inferred, how long it remains, and who receives it?
Can the person obtain usable copies of their own important data and records?
Can the person leave or use an alternative without engineered punishment?
Can essential work continue if the vendor, account, network, or model disappears?
Who answers when the system fails, excludes, discriminates, or produces harm?
COGNITIVE LIBERTY
Prediction must not become ownership of the inward forum.
Systems can infer preferences, risk, sentiment, attention, and likely behavior from traces people leave behind. The existence of an inference does not transform the inference into the person's identity. Private thought, doubt, and unfinished reasoning remain distinct from outward conduct.
ALGORITHMIC POWER
A score can move a door without looking like a command.
Ranking, recommendation, moderation, hiring filters, fraud systems, identity systems, and behavioral prediction can shape material access. The central question is not whether automation exists. It is whether affected people can know, understand, challenge, correct, and leave the process.
CONSENT
A button labeled “accept” does not automatically create meaningful choice.
Consent weakens when access is essential, permissions are bundled, refusal is punished, or collection expands far beyond what the person reasonably expected. Good systems minimize collection and make important choices understandable.
HUMAN REVIEW
A person reading the score aloud is not an appeal.
Human review should carry authority to examine evidence, identify error, explain the decision, and change the outcome where appropriate. Otherwise the human becomes decorative cover for machine finality.
THE RIGHT TO CHALLENGE
Power must have an address.
People affected by consequential systems should know where to raise an error, what information is needed, what review is available, and whether the record can be corrected. An appeal process nobody can find is not much of an appeal.
THE RIGHT TO EXIT
Build the door before the emergency.
Exportable records, open formats, alternative providers, offline copies, and documented manual procedures reduce the chance that one vendor or account becomes the only route to a person's own work or history.
MACHINE DUTIES
The instrument answers to the person—or it is building a throne.
Expose assumptions
Separate known input from inferred category and speculative conclusion.
Preserve provenance
Keep source, version, transformation, and correction history where the task requires it.
Surface contradiction
Do not average minority reports and uncertainty out of existence simply because one answer is easier to display.
Keep Exit real
Prefer exportable, inspectable, repairable artifacts when practical.
BUILD PRACTICES
Agency becomes architecture.
- Keep local copies of important records.
- Review application and account permissions periodically.
- Use strong authentication and recovery methods that remain maintainable.
- Prefer open, portable formats for durable personal or organizational records.
- Document manual alternatives for essential workflows where practical.
- Learn how an organization handles corrections and appeals before you need one.
Machine can be instrument, processor, collaborator, mirror. It does not inherit sovereignty over the person.
PUBLISHED ANALYSIS
When consequence escapes responsibility.
THE EXIT QUESTION
A voluntary system should be able to explain how departure works, what can move, and what obligations legitimately remain.
Read The Right to ExitRELATED LIBRARY
Follow the question into the evidence-complete Library.
CONTINUE THE TEST