Original title: A Human Capability Test for Learning Outcomes in the AI Era
Navid B Saleh · September 2026
Generative AI can do schoolwork so well that submitted assignments no longer prove what a student actually knows. Universities recognize this problem, but simply giving permission to use or not use AI still fails to show real student learning. To fix this, researchers created a "human capability test." It works by asking three questions about an assignment: What must the student do completely on their own? What parts can AI help strengthen? And what must the student verify, defend, and take responsibility for? The authors tested this using an engineering problem. Instead of asking whether a student used AI, schools can now find out what skills belong to the human.
Imagine a cooking class where students submit a dish. The teacher tastes it and assumes the student made it. But now a student could have ordered takeout and claimed it as their own. The teacher can no longer trust the dish alone. In this paper, the test asks three questions: what ingredients must the student choose themselves, what steps can they practice with a recipe book, and what must they explain about why the dish tastes that way. This separates the student's actual cooking skill from the takeout.
For universities, the old way of grading no longer works because AI can produce perfect assignments. This test gives educators a way to prove what a student actually knows, restoring trust in degrees. Students benefit because their genuine skills become visible, not hidden behind AI-generated work.
The framework is intentionally not prescriptive, so it does not give step-by-step instructions. It is illustrated with only one example, so its effectiveness across all subjects is not shown. The abstract does not discuss how to prevent AI use during the test itself or how to grade the responses. No experimental results are provided to validate the approach.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
A Human Capability Test for Learning Outcomes in the AI Era — Navid B Saleh — EdArXiv — doi:10.35542/osf.io/gjrb9_v1
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.