Learning Cannot Be Proven by the Final Answer

This week brought weak research evidence, unreliable AI detection, and more calls for schools to pause. Together, the stories showed why final submissions are no longer enough to prove that learning happened.

What Happened This Week

Stanford's SCALE Initiative combed through more than 800 academic studies on AI in K-12, expanded to a repository of 1,100 papers, and found just twenty rigorous causal studies.¹²

Seven doctoral nursing students rated three blind responses to their statistics questions.³⁴

Eleanor Canina got a 0 and a note that said 'evidence of AI, Please redo.'⁵⁶

On April 29, parents, students, and teachers filed into a Panel for Educational Policy meeting in lower Manhattan and stayed for seven hours.⁷⁸

A panel at last month's ASU+GSV Summit did something unusual for an industry conference.⁹¹⁰

What These Stories Have in Common

A polished answer can come from careful thought, heavy AI assistance, or no understanding at all. Detection tools cannot reliably separate those cases after the work is finished. The useful evidence exists earlier, in the process that produced the answer.

Koan starts from a simple premise. The final answer is no longer a reliable measure of learning. Schools need to understand how students reached that answer and whether the work made them more capable.

What Schools Can Do

Schools should preserve drafts, revision history, student questions, and short reflections. This gives teachers a fairer basis for judgment and gives students a way to show that the thinking is their own.

When the final answer becomes easy to produce, the path to that answer becomes the evidence that matters.