← All articles
AI and assessment

The Rubric That Reads Itself

· 2 min

Building a rubric with students, then testing it against AI-written work. Step by step, including the part that failed.

This was a ninth-grade English class in early May, four periods, and it was supposed to take one lesson. It took three, and the third one is the only reason I am writing it up.

Day one: build it

We wrote the rubric together, which is not new and not mine. Four criteria, in their words, on the board, with an argument about each one. The argument is the lesson. "Evidence" took eleven minutes and ended with a definition I would not have written: evidence is a quote that would annoy someone who disagreed with you.

Day two: test it

I brought four essays on the set text. Two were generated, two were from last year's class with permission and names removed. The students did not know the mix. They scored all four against their own rubric, in pairs, and then we compared.

  • The two human essays scored lower on organisation. Every pair agreed.
  • The two generated essays scored top marks on organisation and structure, and the room noticed within about four minutes that they had nothing to annoy anybody with.
  • One pair gave a generated essay full marks on evidence. They were not wrong by their own definition, which is how we found the hole.

Day three: the part that failed

Here is what I had planned and what actually happened. I expected the rubric to hold up and the exercise to end in a satisfying conclusion about voice. Instead the students found that their own rubric, applied honestly, rated the hollow essays above the real ones on three of four criteria.

Our rubric measured whether an essay was assembled correctly. We had never once asked it to measure whether anyone had thought anything.

So we spent a third lesson fixing it, and the fix was a fifth criterion they named themselves: does this essay take a risk that could be wrong? It is harder to score, it produces more disagreement, and their teacher tells me it is the only criterion the students now argue about before they hand work in.

I have run this with three classes since, and the rubric fails every time, in roughly the same place. I no longer treat that as the exercise going wrong. It is the exercise.