Positions for human judgment in complex systems.

The position

AI literacy is a practice that develops through reflection, not a checklist to complete. It means understanding enough about AI systems to use them responsibly, evaluate outputs critically, and retain human authority over how they are employed.

The position joins two modes of practice: explicit verification rules when consequences are high, and reflective habits that improve judgment during ordinary use.

Why this matters

AI is increasingly present in work, education, health, law, and daily decisions. The practical problem is not simply that systems can be wrong; it is that people can over-trust plausible outputs and offload judgment without noticing.

The aim is calibrated trust: use a system where it helps, identify when its output needs independent checking, and remain responsible for the decision that follows.

Substantive claims

The position does not depend on treating every interaction alike. It distinguishes between what an AI system can assist with and what a person remains accountable for.

1. AI outputs can appear reasoned without carrying equivalent human understanding or responsibility.
2. The required level of verification should increase with the consequence of a decision.
3. Trust calibration can be practiced; it is not a fixed personal trait.
4. Reviewing what was trusted, changed, or rejected can improve later judgment.
5. The user remains the custodian of how an AI system is directed and used.

Rules for high-stakes decisions

For medical, legal, financial, safety-critical, or other decisions that materially affect another person’s wellbeing, an AI output is not enough to justify action.

1. Verify consequential claims with independent, non-AI sources.
2. Consult appropriate domain expertise before a final decision.
3. Cross-reference the claim and record the reason for accepting or rejecting it.

Reflection practice for everyday use

Not every interaction merits a formal review. For significant everyday use, the practice is to make a small return to the interaction and learn something concrete from it.

1. What did I take from this interaction that I trust, and did I verify it?
2. What did I change or reject, and why?
3. What did this case show about the system’s strengths or limits?
4. What is one thing I would do differently next time?

The strongest counterargument

A serious objection is that reflection is too soft to change behavior: people may not verify outputs, and a concrete measurable skill set may be more effective than a reflective practice.

The position accepts that objection as a limit. Reflection is not presented as a complete solution; it is one necessary discipline alongside rules, expertise, and institutional safeguards.

What would change this position

A public position should name the evidence that could require revision rather than treating its current formulation as final.

1. Evidence that prescriptive frameworks reliably outperform reflection-plus-rules approaches.
2. Evidence that reflective practices fail to reduce AI-related harms in measurable settings.
3. Evidence that a user-custodian model becomes technically impossible for increasingly agentic systems.
4. Evidence that critical-thinking interventions have no effect on later reliance behavior.

Evidence and revision record

The published position brings together policy, research, and learning-practice sources. Its source set is part of the claim boundary, not an afterthought.

Initial public record: 2026-09-23. The complete source list and revision history remain available in the full position record.