Insight into the learning process

• Tip 1: start small, think big

  You don’t need to overhaul your entire course to adapt to AI use. The Norton authors deliberately kept their interventions small: one transparency statement, one ten-minute conversation or one fifteen-minute class activity. 

• Tip 2: examine the evidence question before you set an assignment

  For every form of assessment you use, it is worth asking yourself: does this form of assessment fit my learning objective? With the rise of AI, it is increasingly difficult to gauge a student’s level of knowledge or skill from an essay or report. So ask instead, what does this task prove? And what conclusion does that justify? The question is not what the outcome looks like, but what the process leading to it proves.

  Say, for example, that you want to assess critical thinking. In the past, you may have done this with an essay, but now that AI is here, an essay no longer proves it. The essay itself is the outcome of a process, and you need to ask what exactly that process demonstrates. For example: working independently, working with feedback, academic writing, developing arguments or organizing and prioritizing information. You can then look for which tasks can actually demonstrate those skills. For example, argumentation exercises that students carry out in groups, with rounds of critical revision.

• Tip 3: make the process visible during learning, not afterwards

  The Norton authors recommend process visibility: interim drafts, learning logs or, for example, the ‘red light, then green light’ template. The latter means that AI is prohibited during the brainstorming phase and the first draft and only permitted for specific purposes after that. You run the problem exploration or brainstorming phase during a class activity in which students cannot use their digital devices. The point is: their ideas first, and only then building on them with AI. Other options include pitching research questions or defending conclusions without a laptop.

  Fawns and colleagues add a warning about reflection assignments during or after the process: a reflection submitted afterwards is in fact just another product, and just as easy to generate with AI. So, opt for process evidence that emerges as the work progresses, such as: 

  • drafts and worked-out material from class; 
  • short progress discussions;
  • screen captures in which the students show themselves at work; 
  • reflection tasks such as one-minute papers; 
  • think-pair-share activities, which you can run during a class session without digital devices. 

• Tip 4: add a performance moment to product assignments

  Performance, where you see the student in action live, is the best AI-aware form of evidence. Both sources converge here: a short conversation of five to ten minutes about the submitted work, embedded in the course and not framed as an interrogation, fundamentally changes the basis of evidence. Students who know such a conversation is coming up, also study differently; it shapes how they prepare. Do watch out for what Fawns and colleagues call ‘slippage’: a presentation that you are in fact assessing on its pre-written script and slides is, in disguise, a product assessment after all. When assessing a performance, assess thinking and application in the moment: the follow-up question, the unexpected example or the response to a counterargument.

• Tip 5: don’t use surveillance

  Strikingly, both sources independently draw the same line here. Process visibility used as a policing tool, such as version history as burden of proof or monitoring of tool use, disrupts the process itself and makes students performative. You then measure how well they are monitored, not how they learn. Opt for transparency instead: ask students to state briefly with each assignment what they did with AI and what it produced. You could use this statement of work generator or AI disclosure statement generator for this. This makes the chain of evidence easier to interpret. That way, you know what you are assessing and you open up a conversation instead of policing students. 

• Tip 6: break the product saturation and don’t average away evidence

  If almost every assessment in your course or programme is a submitted artefact, students learn that the final product is the evidence of success. Precisely the evidence that AI can fake most easily. So take stock, in the curriculum or your course, of which of the four proxies dominate (product, process, performance and practice) and diversify them.  

  Do pay attention to synthesis: a strong product and a weak oral exam do not average out to ‘sufficiently competent’. That gap is an interesting signal to probe further: why is there such a difference? For example, allow a maximum difference of two points between the product and the defense (or one point, depending on what you consider acceptable). Or require at least a pass for each component. Compare pieces of evidence explicitly with each other instead of adding them up into a single grade.