Who Gives Feedback When No One Is There? Designing Module 2 for Coursera
Removing the instructor does not remove the need for teaching. It moves teaching into the design.
For my capstone, I am converting one module of Dartmouth’s Foundations of Implementation Science from Canvas to Coursera. On Canvas, the course has live Zoom sessions, paid textbooks, and an instructor who grades the work. On Coursera, none of that exists. Learners are working health professionals and graduate students around the world, often studying on a phone, in short sessions, with no one to ask. The question I kept coming back to was simple: who gives feedback when there is no instructor?
I chose Module 2, “Choose your EBI,” because picking an evidence-based intervention is the first high-stakes decision in the course project. Everything learners build later rests on it. Bates (ch. 5) notes that MOOCs struggle most with feedback on exactly this kind of complex, higher-order work, so that is where I put the most design effort.
Collins and Kapur’s cognitive apprenticeship gave me a structure. The “Propose your EBI” template breaks the decision into eight boxes, each with a worked example, which is modeling and scaffolding. An AI check gives feedback, which is coaching. A short reflection log asks learners to explain their thinking and look back on it, which is articulation and reflection. Designing it step by step showed me that scaffolding does not mean giving the answer. It means making the hard part visible.
The part I thought about most was the AI. I did not want learners to trust it blindly (“if the AI says it’s fine, it’s fine”) or to avoid it because it feels like cheating. So the reviewer is told to challenge first, then name strengths, and to separate what came from the learner’s own text, what is general knowledge, and what still needs checking. I hope this pushes learners to question why they chose their EBI, the double-loop learning Rigolizzo (2012) describes, instead of only fixing surface errors. AI appears only in ungraded steps. Peers grade the learner’s own reasoning, not the AI’s output.
This connects to what I wrote about accessibility. Not every learner has a good AI account, a fast connection, or the same digital literacy, and a design that depends on AI would quietly leave them out. That is why the plan includes a no-AI self-check with the same six criteria, captions and transcripts, and videos of about six minutes, where Guo, Kim and Rubin (2014) found engagement starts to fall.
What I am taking from this project is that removing the instructor does not remove the need for teaching. It moves teaching into the design: the worked examples, the order of the steps, the questions the AI asks, and the plan-and-review prompts that help learners manage their own learning. Before the gallery walk, I will test the module with two or three users in a think-aloud, to find out whether this structure actually feels like support from the learner’s side.
Three ideas I’m carrying forward
- Put the support where the decision is hardest. Scaffold the high-stakes choice, not the easy steps around it.
- Make AI a challenger, not a grader. Use it to question reasoning in ungraded steps, and let people judge the work.
- Design the no-AI path at the same time. Equal access means the course works without the tool too.