Assessment decides which work gets outsourced. On a statistics course, the exam could be passed only if the student also understood what they had calculated. Students therefore used the AI for explanation rather than for the solution. The purpose of the exam decided that.
A course can be adjusted mid-semester. On the graphics course, the lecturer set weekly assignments so that students would work continuously. He suspected the assignments were being done with AI. The logs revealed where students got stuck, what they asked, and what the AI supplied in place of understanding. He then configured the assistant to explain first and answer afterwards. He changed things on the basis of what he saw.
Everyone worked, instead of one student. On the Databases course, each student is meant to work through the exam questions themselves. In practice one student wrote them up and the others studied from that. The lecturer made it possible to work the questions through with AI as well. The logs showed that more students worked than before, and follow-up questions appeared.
Replacing one practice that had long lost its purpose was enough.
Questions we cannot have ready answers to.
- How do we teach a subject when the output and the understanding come apart?
- What makes genuine dialogue appear, where the student's own thinking enters after the finished answer and the conversation carries on?