Generative AI has unsettled familiar assumptions about how assignments provide evidence of student learning. When polished products no longer reliably reveal the thinking behind them, assignment design must focus more directly on the judgments, decisions, interpretations, and revisions students make. The central challenge is not detecting AI use or keeping pace with changing tools, but creating stronger evidence of critical thinking, learning, and authorship.

Preorder Artificial Intelligence, Authentic Voices from Bloomsbury

Making thinking visible

Finished product
The essay, report, analysis, solution, design, presentation, code, or other work a student ultimately submits.

Finished products still matter, but under AI conditions they no longer reliably show who made the important decisions, how alternatives were weighed, or whether the student can evaluate and take responsibility for the work.

Process record
Drafts, annotations, prompt histories, revision logs, reflections, or other records of how the work was produced.

Process records can make activity visible, but activity is not the same as learning. They do not necessarily show whether the student exercised judgment, applied disciplinary standards, or understood the significance of the choices being made.

Designed evidence of judgment
A response, annotation, comparison, revision, or explanation created to show how a student made a consequential decision.

Unlike a general process record, it is tied to a specific inference: how the student weighed alternatives, applied criteria, responded to evidence, or decided what to retain, reject, or change.

The goal is not to document everything a student does, but to make the consequential thinking in an assignment more visible.

Diagnose an assignment

1. What important decision must the student make?

2. Are there genuinely plausible alternatives?

3. What criteria should guide the decision?

4. Does the decision change what happens next?

5. Will the student have to revisit that decision when evidence, feedback, or conditions change?

6. What evidence of the student’s judgment will the instructor actually see?

If the assignment has already made the important choices, students may be left mainly to execute a predetermined path. Redesign begins by locating a decision that matters and deciding what evidence would make the basis of that decision interpretable.

Sample Assignment

This assignment asks students to commit to an initial explanation and then revise it when a new pattern appears in the same evidence. Because students must work directly on the original model rather than replace it, the instructor can see how explanatory weight shifts, which earlier decisions are retained, and how the student responds to evidence that complicates the first account.

Shifting Explanatory Weight (Environmental Studies)

In this unit, we have been examining patterns of coastal wetland decline across a ten-year period using regional summary data.

Stage 1: Initial Model (in class)

Using the data discussed today, construct a one-page causal map explaining the overall decline in coastal wetlands across the region.

Your map must include all four drivers we examined:

• Coastal development (housing, roads, shoreline alteration)

• Agricultural nutrient runoff

• Storm frequency

• Sea-level rise

On your map:

• Draw arrows to show causal relationships among drivers and wetland decline.

• Indicate direction of influence.

• Indicate which driver is doing the most explanatory work in your model and which is doing the least. Mark each driver with a relative weight (1 = most explanatory; 4 = least explanatory).

• Visually distinguish the driver you judge to be most explanatory in your current model.

You may not introduce additional drivers at this stage.

Submit your map at the end of class. It will be returned to you next week.

Stage 2: New Pattern in the Same Dataset (in class, one week later)

Today you are receiving additional information about the same region.

Across the dataset, wetland loss remains relatively low across a wide range of development levels and then increases sharply once development density exceeds a particular level. In several areas with moderate development below that level, wetlands remain largely stable over the ten-year period.

Using your original map (which must remain visible), revise your model in a different color. Your revision must work from your original model rather than replacing it with a new explanation.

You must:

• Reassess the relative explanatory weight of the four drivers in light of this pattern. Do not erase your original markings.

• Adjust your map as needed so that your explanation accounts for the pattern described above.

• Revise directly on your original map. You may not discard it or redraw from scratch.

Submit the revised map at the end of class.

The first map records an initial judgment about the relative importance of four causal drivers. The revised map shows how the student responds when new evidence puts pressure on that judgment. Because the original decisions remain visible, the instructor can interpret what changed, what remained stable, and how the explanation was reorganized.