AI Literacy as a Developing Practice

Position

AI literacy is not a badge or a one-time checklist. It is the developing ability to understand enough about a system and its use to ask what it can do, where its limits matter, how to check its output, and who remains responsible for the decision.

Hearth & Code treats that as both a personal practice and a governance question. People need usable ways to question AI; providers and deployers also shape the systems, contexts, training, and safeguards around use. This is our working position, not a validated literacy curriculum or a claim of compliance with a standard.

What public frameworks say

The sources describe different audiences and responsibilities, so they should not be collapsed into one universal checklist:

• The OECD and European Commission AI Literacy Framework describes knowledge, skills, and attitudes for primary and secondary education. It is an education framework, not a certification for every AI user.
• The European Commission’s AI Act Article 4 Q&A describes measures providers and deployers take to support literacy among staff and other people dealing with systems on their behalf. There is a wording mismatch within the FAQ: its opening summary still says “ensure a sufficient level,” while its later amendment answer says no specific—or “sufficient”—level is mandated for any individual. The amended Regulation (EU) 2026/1744 and the consolidated Article 4 text state the updated rule: organizations take measures to support development, consider the staff and other persons’ knowledge, experience, education, training and use context, and consider the people or groups on whom the systems are used. They need not guarantee any specific literacy level for an individual. The FAQ says Article 4 does not require measuring each person’s knowledge or issuing a certificate; training or guidance records are optional examples. Separately, duties concerning staff training for human oversight remain for high-risk AI systems. Read the amended text with the FAQ rather than relying on its unrevised opening summary. This is not legal advice.
• UNESCO’s Recommendation on the Ethics of AI frames public awareness and literacy through accessible education, civic engagement, human rights, and participation. It is a policy instrument for states and institutions, not a personal conformance scheme.

The Commission also points to the AI Act’s broader rationale: affected people may need enough understanding to recognize how AI-assisted decisions affect them. That public-facing aim is not the same as an Article 4 requirement to measure or certify each affected person.

Together, these sources support a broader view: literacy includes individual judgment, organizational responsibility, and the conditions that let affected people understand and question systems.

What evidence can and cannot tell us

AI reliance is context-sensitive. NIST’s Generative AI Profile identifies over-reliance and automation bias as risk considerations in human-AI configuration. It is voluntary risk-management guidance, not proof that every user or task experiences the same effect.

A 2026 Scientific Reports experiment studied 295 participants judging whether faces were real or AI-generated with purported human or AI guidance. Its results concern that experimental task; they do not establish how people rely on AI across professional or high-stakes decisions.

A 2026 classroom study of 226 participants found that a critical-thinking intervention reduced direct adoption of AI-generated content in its study setting, while self-reported critical thinking did not significantly improve over the short intervention. That is a useful, bounded result—not proof that reflection alone improves AI literacy generally.

These findings support care and further inquiry. They do not establish that one reflection routine, verification rule, or interface will work for every person, system, or domain.

A short reflection practice

After a consequential or otherwise significant AI interaction, consider:

1. What did I rely on, and what source supports it?
2. What did I change, reject, or leave uncertain?
3. Who could be affected by the result, and who is accountable for the next step?
4. What would I check or do differently if I returned to this task?

This is an optional aid for noticing decisions and uncertainty. It is not a validated assessment, a substitute for domain expertise, or evidence that reflection by itself changes future behavior.

What could change this position

This working position should change as better evidence and governance practice emerge. We would revise it if robust evidence showed that another approach better supports understanding and appropriate reliance across different tasks—or if these practices burdened users without improving their ability to question consequential outputs.