The AI Student Test
The problem this fixes
Higher education has always rested on one distinction: students who learn versus students who copy. Learning means transformation, wrestling with material until you can restate it in your own understanding. Copying means substitution, handing in someone else's words.
Generative AI breaks the visible signals instructors used to tell those apart. AI-assisted writing can look fluent, structured, and coherent without any of the struggle that produces real learning. Existing academic integrity rules, built around intent, human authorship, and originality of phrasing, translate poorly. You can't always tell whether AI was used. Detection tools are unreliable, especially for multilingual students. And the deeper question isn't even detection. It's whether learning happened.
What the test does
The AI Student Test reframes the question. It doesn't try to identify AI in the writing. It evaluates the writing itself, against four criteria that capture what assessment is supposed to measure in the first place.
The test rests on a simple analogy: imagine a "student who learns" and a "student who copies" as two reference points. AI-assisted work can resemble either. The four criteria help instructors place a given submission closer to one pole or the other.
The four criteria
Sources are accurately represented. Required readings are clearly used. Citations follow the conventions of the field. If AI helped summarize or paraphrase, the student verified the output against the original source rather than trusting the machine's interpretation.
The work goes beyond restating what the source said. It interprets, synthesizes, critiques, or applies the material. AI outputs, where used, were reorganized and integrated with course concepts, not pasted in as-is.
The student gives a clear, specific account of how AI was used and can explain key choices when asked. Vague disclaimers don't satisfy the criterion. The standard is: an instructor following up with "why this section?" gets a coherent answer rooted in the student's reasoning.
The submission demonstrates the knowledge and skills the assignment was designed to teach. AI assistance, where present, supported access or expression, it didn't bypass the cognitive work the course is meant to develop.
What changes when institutions adopt this
Three things shift:
- Assignment design becomes the front line. If existing assignments can't distinguish AI-substituted work from learned work, the assignment is the problem. The test pushes instructors toward tasks that require source comparison, contextualization, critique, or integration, tasks unedited AI output can't satisfy.
- Detection stops being the focus. Investing in surveillance tools that flag "AI use" stops being the policy lever. Investing in pedagogy and rubric design takes its place.
- Policy becomes durable. Tool-based bans break the moment a new tool launches. Learning-centered standards apply across whatever AI ships next year, because they're tied to learning, not to a specific technology.
What it does not do
The test isn't a detection tool. It won't tell you whether AI was used. It won't catch a student who used AI invisibly. What it does is make the question of detection less central, because if the work demonstrates learning under the four criteria, the AI-or-not question is largely moot. And if it doesn't, the work fails the test regardless of what tools were involved.