AI Literacy

Can you use AI effectively without surrendering understanding, verification, or authority?

AI literacy is the capability to understand what an AI system is doing, use it for a bounded purpose, verify consequential output, and retain human authority. Purpose determines the beneficiary and valuable change.

It is broader than writing prompts or coordinating agents. Prompting expresses context and constraints. Orchestration delegates, evaluates, and integrates work. AI literacy governs when and how either practice is appropriate.

Capability Map

Question	Learner responsibility
What enables this?	Know the relevant capability envelope, inputs, outputs, limits, and failure modes.
Who authorises it?	Name the beneficiary, consent, permissions, values, and protected constraints.
Where can AI assist?	Choose bounded explanation, practice, analysis, creation, simulation, or coordination work.
What must be understood?	Retain enough domain knowledge to frame, question, explain, and notice failure.
What must be verified?	Check consequential claims, sources, reasoning, calculations, decisions, and effects.
What proves learning?	Repeat independently, transfer to a second context, and create beneficiary value.

Functional Analogy

Describe an AI system without projecting human moral or experiential properties:

• Native capability envelope: what valid evaluations show it can and cannot reliably do.
• Authorised beneficiary constraints: whose interests, permissions, and protected rules govern use.
• Compute and context budget: which resources and information are available.
• Sustained execution: how it performs within those bounds across a task.

This analogy does not imply consciousness, intrinsic morality, goodwill, consent, or flourishing.

Use Protocol

1. Name the beneficiary, valuable change, and human authority.
2. Identify the capability required and what you can do unaided.
3. Decide where AI can assist and where it must not decide.
4. Supply relevant context, constraints, evidence requirements, and a stop condition.
5. Inspect the process and output for known failure modes.
6. Verify consequential elements with primary evidence or an authorised person.
7. Remove assistance and repeat the important capability independently.
8. Transfer the capability to a materially different context.
9. Observe beneficiary value and unintended effects.
10. Adapt, redirect, stop, or reinvest.

Dependency Test

Compare:

unaided → assisted → verified → independent → second context

If output rises but independent explanation, error detection, judgment, or transfer falls, the system created leverage without learning. That may still be useful when authorised, but name it accurately and manage the dependency.

Human Authority

Humans retain authority over values, beneficiaries, consent, consequential trade-offs, acceptable risk, claims about human worth or limits, wellbeing guardrails, escalation, stopping, and accountability for effects.

Failure Modes

• Mistaking fluent output for understanding or truth.
• Delegating values, consent, consequential risk, or stop decisions.
• Using assistance without an unaided baseline or independent test.
• Verifying with the same unexamined source or system that produced the claim.

Changes my mind: robust evidence that a named AI system can own human values, consent, and moral accountability would require revising this authority boundary.

Practices

• Prompting — express purpose, context, constraints, and evidence needs.
• Orchestration — frame, delegate, evaluate, integrate, and stop multi-step AI work.
• Truth Recognition — calibrate confidence and verify claims.
• Learning — fade support and prove independent transfer.

Agent Use

• Page job: canonical owner for general AI literacy.
• Human authority: beneficiary, values, consent, risk, verification threshold, and stop decision.
• AI assistance: bounded by capability, context, permissions, and evidence.
• Proof: verified use plus independent and second-context performance where learning is claimed.

Context

• applies-to Education Platform — choose AI as a bounded scaffold rather than the purpose of learning.
• proved-by Education Performance — test verified, independent, and second-context performance.
• Working Memory — control context and external memory
• Taste — judge quality without surrendering standards

Questions

• What must remain human-authorised in this use?
• Which consequential elements require independent verification?
• How will you test what remains after assistance fades?

Next question: Does this use create independent capability, authorised leverage, or unacknowledged dependency?