Human Agency Under AI — Organizational Diagnostic
Purpose
This tool helps leadership teams examine whether AI adoption is enlarging human capability while preserving meaningful human agency, judgment, accountability, authority, relationship, and learning.
It is a developmental reflection tool, not a psychometrically validated assessment, clinical instrument, legal audit, compliance certification, technical AI audit, or safety guarantee.
How to use it
Choose one important AI-assisted workflow or decision domain.
Score each statement:
- 0 — Not true / absent
- 1 — Rarely true / seriously underdeveloped
- 2 — Partly true / inconsistent
- 3 — Usually true / functioning well
- 4 — Clearly true / consistently demonstrated
Discuss differences in scoring before treating a number as meaningful.
1. Human Decision Ownership
- We can name the human being who owns each high-consequence AI-assisted decision.
- Decision owners understand that AI output is input to judgment, not a transfer of authority.
- Ownership remains clear across people, vendors, systems, and automated steps.
- We know which human authority can affirm, reverse, or escalate a challenged decision.
2. Capacity-Fitted Respondability
- People held accountable have enough knowledge to evaluate the relevant output.
- They have enough time, access, context, and authority for meaningful oversight.
- Responsibility is not assigned merely because a person is the final human in the process.
- Training and support increase when responsibility exceeds present capacity.
3. Discernment & Challenge
- People can explain why an important AI-assisted conclusion should be trusted.
- Teams are expected to challenge confident outputs when evidence is weak or misapplied.
- People know the failure modes relevant to the tools they use.
- Raising a legitimate concern is treated as responsible participation rather than obstruction.
4. Authority Boundaries & Delegation
- We have defined what AI may recommend, automate, draft, prioritize, or decide.
- We have identified decisions that must remain human.
- People know when they may override, stop, or bypass the system.
- Delegation does not obscure who remains accountable for consequences.
5. Correction Rights & Escalation
- There is a clear path for pausing or escalating a questionable process.
- People who surface errors are not punished merely for slowing the workflow.
- The organization can correct an AI-enabled decision after execution.
- Significant failures produce learning, not only blame.
6. Human Skill Preservation
- We have identified skills that must remain practiced even when AI can perform the task faster.
- People periodically demonstrate critical reasoning without relying entirely on AI.
- AI use is designed to enlarge human capacity rather than systematically replace participation.
- Training includes judgment, interpretation, and correction—not only tool operation.
7. Relational Consequences
- We examine effects on trust, dignity, participation, and belonging.
- We distinguish interactions that are efficient to automate from those requiring human presence.
- Important feedback, correction, care, or conflict is not routinely delegated merely for convenience.
- People affected by AI-assisted decisions have an intelligible path to human explanation or review when appropriate.
8. Accountability Without Shame
- We distinguish capacity gaps, system weakness, poor judgment, and misconduct.
- Correction gives people an actionable path.
- Accountability is proportionate to actual role, capacity, and authority.
- Leaders can preserve standards without humiliating people who expose uncertainty or error.
9. Learning & Corrigibility
- Policies and workflows can be revised when evidence contradicts earlier assumptions.
- Leaders can change course without treating revision as weakness.
- Significant errors lead to changes in process, training, policy, or design.
- Teams compare expected outcomes with actual outcomes and learn from the difference.
10. Purpose, Human Value & Governance
- We can explain why an AI use serves the people and mission affected—not only why it is efficient.
- Productivity gains are weighed alongside effects on agency, trust, participation, judgment, and development.
- Leaders can identify cases where an available automation should not be used.
- AI adoption is governed by explicit human purposes rather than capability alone.
Reflection bands
Each domain has four statements and a maximum score of 16.
- 13–16 — Functioning strength
- 9–12 — Developing but inconsistent
- 5–8 — Material vulnerability
- 0–4 — Immediate attention
These are working reflection bands, not validated norms.
Do not rely on the total score alone
Pay particular attention to:
- the lowest domain,
- the widest leadership disagreement,
- the weakest domain attached to the highest-consequence decision.
Restorative developmental response
Where capacity is weak:
Truthful assessment → capacity-fitted expectation → nourishment/support → meaningful participation → observable development → progressively enlarged respondability.
Privacy
Do not enter employee names, health information, diagnostic information, disciplinary details, or other sensitive personal information into a public version of this tool.
Next step
Use the two weakest high-consequence domains to create one 30-day action plan.