By now, nearly every faculty member has a story. A student submission that reads a little too polished. An assignment that could — and probably did — get completed in sixty seconds with a chatbot. A policy discussion that turned into an argument no one felt equipped to resolve.
The instinct for many institutions has been to treat AI as an integrity problem: detect it, police it, ban it. In our recent Inside Higher Ed webinar, Rethinking Course and Assessment Design in the AI Era, Jocelyn Wright, PhD, and Geri Atanassova-Boft, PhD, made the case for a different starting point — one that centers course design rather than surveillance.
A quick note on framing: Jocelyn and Geri were clear about what this session was and wasn’t. There is no list of perfectly AI-proof assignments, and there is no universal template that works across every discipline. What the session offered instead was a set of practical design moves faculty and instructional designers can apply right now — without waiting for a comprehensive institutional AI policy. The core argument is this: AI has exposed a design problem that was already there. Many traditional assignments were never built to surface the thinking we actually care about. AI has simply made that gap harder to ignore.
Four Strategies from the Session
Make Thinking Visible
Polished output is easy to delegate. Thinking is not. When an assignment asks only for a final product, it measures the submission more than the learning. Process logs, annotated drafts, staged proposals, recorded explanations, and post-assessment reflections all shift the focus to reasoning rather than results. A three-stage sequence works well in practice: a short proposal before the work begins, the task itself, and a brief recorded reflection walking through key decisions and what changed from the original plan. Evaluation moves from the final answer to the process behind it.
Assess Application, Judgment, and Transfer
Generic prompts produce generic responses, and AI handles generic very well. The more a task asks students to engage with a specific context, interpret particular evidence, justify their choices, and explain what they ruled out, the harder it becomes to outsource meaningfully. The epidemiologist example from the session makes this concrete: asking a student to describe the outbreak investigation process is a very different task from placing them in a specific county, asking them to identify a likely exposure source, name a limitation in the data, and defend why they eliminated the two most plausible alternatives. Specific context and required judgment raise the bar considerably.
Distribute Evidence Across Multiple Touchpoints
Concentrating all assessment weight in a single high-stakes submission creates the conditions where AI assistance is most tempting and hardest to detect. Low-stakes checkpoints that build toward a synthesis assignment — combined with timed submission windows at each stage — spread the evidence of learning across the arc of a course. Students who are genuinely engaging with the material show up differently across those checkpoints than students who are not, and the pattern is more informative than any single submission.
Be Explicit About AI Use
In many cases, the right answer is clearer expectations rather than an outright ban. Requiring students to disclose which tools they used, how they used them, and what they changed based on the output makes AI use visible and instructive. Asking students to critique an AI-generated response — identifying errors, omissions, and weak assumptions — requires exactly the judgment skills that matter most. Rubrics built around process and decision-making, alongside prompt guidelines that set transparent expectations for AI-enabled tasks, give both students and faculty a shared framework for what counts as acceptable support versus substitution.
“You don’t need a perfectly AI-proof assessment. You need assessments that produce better evidence of student thinking, judgment, and learning.”
— Jocelyn Wright, PhD & Geri Atanassova-Boft, PhD · Six Red Marbles
Assessment redesign cannot be mandated from above. Faculty who feel surveilled, judged, or overwhelmed are unlikely to experiment meaningfully. The institutions making the most progress are those that have created conditions for low-stakes experimentation and peer exchange — with dedicated instructional design support, peer examples faculty actually trust, and spaces where uncertainty is welcome.
No comprehensive AI policy is required before getting started. Find one high-friction assignment and design your way out of it. Bring two or three colleagues together around a single question. Make one meaningful change to one assessment and pay attention to what happens. Any of those is enough.
Further Reading
-
Outsmarting AI in the Classroom
ASU News / CoRAL Lab · April 2026 -
Understanding the Evolving Needs of Online Learners
UPCEA -
More on AI + Human Insight from Six Red Marbles
SRM Insights
This recap is based on the Inside Higher Ed webinar Rethinking Course and Assessment Design in the AI Era, presented April 28, 2026, by Jocelyn Wright, PhD, and Geri Atanassova-Boft, PhD, of Six Red Marbles. External research referenced: Vivek Gupta and the Complex Data Reasoning and Analysis Lab (CoRAL) at Arizona State University.
