Scaffolded critique-of-AI-output assignment sequence with paired verification of individual understanding¶

Strategy · All strategies
Evidence · no claims cited

Description¶

A strategy from the Evaluating & Fixing AI-Produced Work room in which the object of study is output the student did not write: students read, test, critique, debug, and repair AI-produced code. The room's approaches include giving students AI-generated solutions to critique, improve, or test—"validating behavior, finding defects or unnecessary complexity, identifying missing edge cases, designing unit tests"—and pairing AI-supported coursework with in-class quizzes or examinations that verify individual understanding. Responsibility for formulating prompts and evaluating results shifts gradually onto the student.

Design Implications¶

Context¶

Requirements¶

• Assignments must be designed so the student, not the AI, performs the critique, since any text handed to a student can become a prompt.
• Critique activities outside controlled environments remain open to students using AI to complete the evaluations.

Constraints¶

• Oral examinations provide strong evidence of individual understanding but are impractical for large cohorts.
• AI often produces good-quality solutions for introductory programming tasks, leaving little to critique.

Target Learners¶

• undergraduate computing students
• students in large programming cohorts

Target Learning Goals¶

• code reading, evaluation, and critique
• software engineering practices such as testing and code review

Related Strategies¶

• Use weekly oral code review interviews to incentivize students to understand AI-assisted code
• Assess process over product using staged, connected tasks with iterative evidence
• Replace detector-centred assessment with a four-stage process-oriented model: declared starting point, versioned drafting, process artefacts, and targeted oral verification
• Pair AI-permitted take-home assignments with supervised modifications, oral defense, prediction, and transfer tasks to verify understanding
• Deliberately scaffold AI-output verification habits rather than assume they develop alongside rising AI use
• Frame GenAI labs explicitly as verification and repair practice alongside prompting

Examples¶

-

Key Sources¶

• Muhammad Sajjad Akbar, Geoffrey Challen, Fraida Fund, Casey Hopkins, Oscar Karnalim, Kevin Lin, James McGuffee, Shubbhi Taneja, Ranysha Ware. (2026). AI Can Do Your Homework. Now What? Report from an online workshop on computing assessment in the age of generative AI. https://arxiv.org/abs/2609.27842