Epistemic AI Literacy: 78.8% non-mastery in student-GenAI

Mengqian Wu's paper, submitted 30 June 2026, introduces Epistemic AI Literacy (EAIL) and analyzes a large dialogue dataset of human-AI co-programming to measure how students enact epistemic aims and processes when using generative AI. The study finds 78.8% of student-GenAI interactions relied on non-mastery-oriented aims, while 11.1% showed high epistemic engagement.

What is Epistemic AI Literacy?

Epistemic AI Literacy, or EAIL, is defined in the paper as an epistemic, process-oriented view of AI literacy that emerges through human-AI interaction across domains. The study adapts the AIR framework (epistemic aims, ideals and reliable epistemic processes) to frame AI literacy around observable aims and processes rather than static skills.

The paper situates EAIL within computer science and human-computer interaction research, arguing learners must construct queries, evaluate and validate AI outputs, and regulate problem-solving strategies during co-programming. The author names specific epistemic constructs the analysis targets: epistemic aims (including mastery-oriented aims) and epistemic processes.

What did the dataset reveal about student-GenAI co-programming?

A large dialogue dataset of human-AI co-programming was coded for observable dimensions of epistemic aims and processes, revealing a prevalent lack of EAIL: 78.8% of student-GenAI interactions relied on non-mastery-oriented aims and less reliable strategies. Conversely, only 11.1% of interactions showed high epistemic engagement.

The coding identified five epistemic processes that students enacted: outsourcing, explanation seeking, verification seeking, prompt monitoring, and epistemic justification. The paper treats mastery-oriented aims as the indicator of stronger epistemic engagement and pairs those aims with advanced strategies such as epistemic justification to mark "high epistemic engagement" in the dataset.

Why it matters

If three quarters of student-GenAI interactions prioritize non-mastery aims, educators and tool designers face a gap between AI access and epistemic development. The paper implies that access to generative AI alone does not produce reliable inquiry practices; students often default to outsourcing or simple verification rather than constructing mastery-oriented queries and offering epistemic justification. That pattern risks encouraging surface-level reliance on AI outputs instead of deeper understanding.

This matters for curriculum designers, platform developers, and researchers in computer science and human-computer interaction because the study provides concrete, measurable constructs to evaluate students' epistemic behavior with GenAI and a baseline percentage of current engagement patterns.

What to watch

Follow efforts that operationalize EAIL in classroom interventions or tooling: the next signals will be datasets or studies that report increased rates of mastery-oriented aims and higher-than-11.1% proportions of interactions classified as high epistemic engagement. Also watch implementations that measure reductions in outsourcing and increases in epistemic justification during co-programming.

Methodological notes: the paper explicitly draws on the AIR framework and focuses on dialogue data from human-AI co-programming; the author is Mengqian Wu and the submission date is 30 June 2026. The study's primary subject classifications are Artificial Intelligence (cs.AI) and Human-Computer Interaction (cs.HC).