Back to blog
AI•August 19, 2023•4 min read

Designing AI that supports the teacher, not replaces them

CoThink.AI grew out of my own teaching. The idea is to help faculty redesign assignments for a world where students use AI, and to support grading against a rubric. From the start I had one rule for the design: the AI does the heavy lifting, but the teacher always makes the final call.

That sounds obvious, but it shapes almost every decision once you take it seriously.

Anchor the AI to the rubric, not its own opinion

If you ask a model to just grade an essay, you get a confident answer with no accountability. So instead of letting it free-form a judgment, we anchor it to the teacher's own rubric. The AI's job is to read the work against the criteria the teacher set and surface where it sees evidence for each one. That keeps the standard the teacher's standard, not the model's vibe.

The teacher owns the rubric. The AI just helps apply it consistently and fast. That line is the whole product.

Make the human step the point, not a checkbox

Human-in-the-loop fails when the human becomes a rubber stamp. So we designed the grading flow so the teacher is reviewing something useful - the evidence and a draft read - and can disagree easily. The goal is to save them the tedious first pass, not to talk them into the model's answer. On a high-stakes call like a grade, the friction of a real review is a feature.

How I think about building AI for educators:

  • Anchor the model to the teacher's rubric, not a generic notion of quality.
  • Surface evidence, not just a verdict, so the human can actually check it.
  • Make disagreeing with the AI easy and fast, otherwise review becomes rubber-stamping.
  • Save the teacher the tedious pass, but leave the judgment firmly with them.

The whole point of CoThink is to give teachers their time back without taking away their authority. The AI is the assistant, and the teacher stays the one who decides.