AI literacy

AI literacy is the knowledge and skill side of AI use: what people understand about AI systems, their capabilities, their limits, and the conditions under which their outputs should be trusted. It should not be collapsed into ai-receptivity. A person can understand AI well and still reject it, or understand it poorly and still be eager to use it.

2025-tully-ai-literacy-receptivity is useful because it treats AI literacy as a predictor rather than as the same thing as adoption attitude. Across cross-country secondary data and six preregistered studies, lower AI literacy often predicted greater ai-receptivity, especially for tasks associated with distinctly human attributes. The paper’s mechanism is that lower-literacy users were more likely to perceive AI as magical and awe-inspiring, not that they believed AI was more capable.

For work and training, the construct implication is sharp: raising AI literacy may improve calibration, verification, and stewardship even if it reduces some awe-driven willingness to use AI. That makes AI literacy closer to a capability and judgment construct than a persuasion or adoption metric.

2025-kellogg-novice-risk-work adds an organizational-learning boundary: novice users can be willing to teach seniors about GenAI while still lacking enough understanding to diagnose where risk mitigation belongs. AI literacy here is not generic familiarity or use confidence; it includes knowing when a risk is caused by model capability, interface design, deployer governance, ecosystem behavior, or local human routines.

2025-passlack-great-power-responsibility extends this boundary from individual knowledge to role-appropriate responsible practice. In 28 interviews across five knowledge-worker groups, participants described AI literacy as a context-sensitive toolbox of technical, domain, communication, ethical, and learning-related capacities, with personal dispositions such as healthy skepticism and continual learning doing work that generic skills training cannot. Its role-ability-fit claim is useful: organizations should not assign AI accountability to a role unless that role has the authority, capability, and support to exercise it.

2026-idan-generative-ai-productivity-divide adds preliminary causal evidence that a narrower interactional capability may determine whether access becomes performance. In a three-day student self-study experiment, the authors’ AI Interaction Competence (AIC)—eliciting, filtering, verifying, and applying model outputs—moderated the benefit of LLM access, whereas GPA and baseline LLM knowledge did not. The construct should not yet be treated as a settled subscale of AI literacy: its measurement needs validation and the setting is educational. But it sharpens the practical boundary between knowing about AI and being able to use its output with disciplined verification and synthesis.

2026-ali-ai-adoption-multinational-workforce gives that boundary a workplace-HR form: effective use required employees to formulate and rephrase queries, interpret summaries, inspect sources, and know when to use an alternative route or human support. In its small descriptive survey, 76% said they would rephrase a poor GenAI answer, while 52% said they never or rarely clicked cited sources. This is not a validated literacy measure, but it shows why rollout training should develop situated search, verification, and escalation practices rather than treating system access as competence.

2025-sun-employee-ai-collaboration-proactive-behavior adds an employee-behavior boundary condition rather than a measurement solution. In its two-wave survey, higher reported AI literacy strengthened the association between employee–AI collaboration and lower workload, but the reported indirect path from collaboration to proactive behavior through workload was statistically significant only in the lower-literacy group. Literacy may therefore change how workers convert AI-assisted capacity into initiative; it should not be narrated as a linear assurance that more literacy always produces more proactivity.

Connections

• Relates to critical-thinking because literacy should support verification, boundary-setting, and skepticism toward AI outputs.
• Relates to human-ai-collaboration because calibrated collaboration requires knowing what the AI is doing well enough to scope, review, and contest it.
• Relates to construct-validity because AI literacy measures can mix knowledge, skill, exposure, confidence, and self-reported familiarity unless the construct is bounded carefully.
• Relates to ai-adoption because literacy can shape adoption quality even when it does not increase adoption volume.
• Relates to novice-risk-work because novice enthusiasm can spread AI use faster than calibrated risk understanding.
• Relates to ai-mediated-learning because interactional capability and structured learning designs may determine whether AI assistance supports comprehension rather than merely a short-run answer.
• Relates to ai-as-a-job-resource because literacy may determine whether a collaborative AI arrangement produces workload relief that workers can reinvest in discretionary action.

Contradictions & open questions

• 2025-tully-ai-literacy-receptivity suggests AI literacy can reduce some forms of AI receptivity, which conflicts with simple training narratives that assume education always increases adoption willingness.
• The workplace question is still open: does higher AI literacy reduce usage, or does it shift usage toward better-calibrated tasks, stronger verification, and less anthropomorphic overreach?
• 2025-kellogg-novice-risk-work suggests AI literacy may need a risk-level component: users must distinguish what can be handled by individual checking, project routines, deployer controls, or ecosystem governance.
• AI literacy may operate differently by task type. For tasks not framed as requiring distinctly human attributes, 2025-tully-ai-literacy-receptivity found that higher literacy could predict greater receptivity.
• 2026-idan-generative-ai-productivity-divide gives AIC a causal-moderation role in a bounded student learning task, but the wiki does not yet know whether AIC is distinct from AI literacy, prior domain expertise, self-efficacy, general learning strategy, or prompt familiarity—or whether a workplace measure predicts accepted quality, verification, and durable performance.
• 2025-sun-employee-ai-collaboration-proactive-behavior treats AI literacy as a self-report moderator in one Chinese organizational survey. The contrasting conditional indirect effects are useful evidence against a simple literacy-is-always-better story, but their interpretation is limited by the observational design, broad collaboration measure, and unexplained heterogeneity in the higher-literacy group.