Right now · 15 minutes
Picture running a coLearn-AI activity in your own class next term.
- 11:25 · 6 min: running a class session: before, during, and after.
- 11:31 · 6 min: turning a lab you already have into an activity.
- 11:37 · 3 min: questions.
The other sections on this page are reference for later: group formation, the instructor view, hosting, and early results.
Group Formation & Role Assignment
📖 Reference: not covered live.
Group Size
Groups of 3–4 work well. Smaller groups mean roles collapse; larger groups make it harder to ensure every student contributes a submission.
Role Assignment
When the platform forms groups, it assigns roles in this order: Facilitator, Process Analyst, Quality Control, Spokesperson. The typing turn passes to the next member after every submission. Instructors can also rotate the turn by hand from the View Groups page.
Forming Groups
Heterogeneous groups (mixed ability) tend to produce more visible reasoning in the trace. Random assignment works for in-class activities; instructor assignment gives more control.
A note on slower participants
Rotating submission control can delay the group if one member is slower to type or to reach consensus. This is a known design tension. The friction is partially intentional (it keeps the group talking) but can become frustrating if too extreme. If a group is held up, rotate the turn by hand from the View Groups page.
What You See During Class
📖 Reference: you saw this live at the end of Part 2.
The View Groups page updates live as groups work. For each group you can see:
- Progress (Question Group 3 of 6), the section timer, and who is connected and holding the turn
- The number of submissions each question group took
- In Observation view: a timeline with accepted and sent-back submissions, an idle timer, and one-tap behavior tags
- Review Activity: the group's full work and epistemic trace, with every submission, AI response, and revision
Using the trace during class discussion
After an activity, projecting 2–3 traces side-by-side for the same question is a powerful discussion starter. Two groups with the same final answer but completely different reasoning paths make visible something that grades alone would hide.
Reviewing the Epistemic Trace After Class
The trace is persistent and exportable. After class, you can:
- Identify questions where most groups were sent back. These are candidates for revising the question or for a short mini-lecture.
- Spot groups where one member dominated. The trace shows who submitted each answer.
- Use the Research Statistics pages and CSV exports for participation, revision, and outcome measures
Running a coLearn-AI Class Session
- Before class, create or select the activity in the authoring tool and verify the activity guidance and question-level guidance are complete.
- At the start of class, assign or confirm groups, explain the roles, and give students the class join code. Allow 3–5 minutes for setup and role selection.
- Launch the activity. Keep the View Groups page open (Observation view on a tablet works well) so you can see group progress without walking to each table.
- During the activity, intervene at the instructor level: answer procedural questions, unblock groups that are genuinely stuck after their allowed retries, but avoid resolving content questions directly. Let the AI do that.
- After the activity completes, run a 5–10 minute whole-class debrief. Project one or two traces. Ask the Process Analysts to share.
- After class, review the traces and adjust feedback prompts for questions that were too easy, too hard, or produced unexpected AI feedback.
One account per browser
A browser can be logged in to only one coLearn-AI account at a time. Logging in to a second account in another tab signs every tab in that browser into the new account. To be an instructor and a test student at once (for example, to watch View Groups while you try your activity as a student), use a different browser for the second account (Chrome and Firefox, say) or a private / incognito window.
Hosting Options
📖 Reference: not covered live.
colearn-ai.com (Hosted)
Use the platform directly at colearn-ai.com. No installation. Contact the project to discuss access for your courses or research. Today’s workshop runs there too.
Self-Hosted
The platform can be deployed on your institution's servers. Suitable for courses with data privacy requirements. Requires a Linux server running Node.js and MySQL. Contact the project for the installation guide.
Pilot / Research Use
If you are interested in running coLearn-AI as part of a research study, the project team can provide IRB-friendly data export formats and support for study design. See Part 5.
Turning Existing Lab Materials into coLearn-AI Activities
You don't need to start from scratch. Most CS1 and data structures lab assignments can be converted into coLearn-AI activities by:
- Identifying the reasoning steps: break the lab into 3–6 sub-questions that each require a distinct cognitive act (prediction, explanation, debugging, design choice).
- Adding a prediction question first: before any code is run, ask the group what they expect. This is the cheapest cognitive engagement move and the most powerful for the trace.
- Writing a feedback prompt for each sub-question: one criterion per question. If a lab problem has three grading criteria, make it three questions.
- Providing sample responses: use the reference solution your TAs would accept.
Example conversion
A typical CS1 Python lab: "Write a function that takes a list and returns the sum of its elements. Test it with [1, 2, 3, 4, 5]."
Converted to three coLearn-AI questions:
Q1 (Prediction): Before writing any code, predict: if we pass [1, 2, 3, 4, 5] to a correct sum function, what should it return? Feedback prompt: Require the correct value (15). Accept any explanation that arrives at 15.
Q2 (Accumulator design): Write a function my_sum(lst) that returns the sum using a for loop and a running total. Show the loop and return statement. Feedback prompt: Require a for loop that accumulates and a return statement. Do not accept a solution that calls Python's built-in sum().
Q3 (Explanation): Explain in one sentence how your running total changes after processing each element of [1, 2, 3]. Feedback prompt: Require an explanation that traces the value of the total variable at each step (0→1→3→6). Push back on answers that describe the result without describing the process.
What Early Classroom Use Has Shown
📖 Reference: not covered live.
✅ Promising
- Participation appeared more evenly distributed. Turn-taking and revisions made each student's contributions visible.
- Instructors saw the reasoning process, not just final answers, for the first time.
- Groups that were sent back and revised often produced the clearest post-activity explanations.
⚠️ Open Tensions
- Guidance that is too restrictive can block progress. Guidance that is too permissive can allow cognitive bypass.
- Rotating control can delay a group if one member is significantly slower.
- Writing good feedback prompts takes iteration. Plan to revise after the first use.
These are preliminary observations, not results from a controlled study. coLearn-AI is research software under active development. If you observe something interesting, positive or negative, please document and share it. See Part 5 for how to get involved in the research.