You are currently viewing Grading the Human, Not the Homework: An Assessment Framework for the AI Era

Grading the Human, Not the Homework: An Assessment Framework for the AI Era

As an instructor of bachelor’s-level courses in software development, where most assignments are code-based, I teach courses that are among the most vulnerable to AI misuse and outsourced learning. This reality left me with a foundational question: How can I structure my assessments so that AI use becomes inconsequential to the grade? Whether students use AI or not, is there a way to remove the pressure on me to serve as an AI detective so that grades reflect what students actually know rather than what they can produce?

That question led me to develop the Grade the Human, Not the Homework framework. Under this approach, grades are not determined by what students submit. Instead, scores are based on how well students can articulate their understanding of what they submitted, demonstrating their ability to explain, modify, troubleshoot, or justify their work.

Testing the framework

I began testing these ideas in my own classroom, starting with a senior-level circuit design course. Previously, students completed a weekly programming assignment and lab report outside of class. Their grade was based solely on the quality of the report and whether their code executed successfully when I tested it. Under these conditions, the likelihood of receiving AI-generated code and reports was high.

With the revised assessment strategy, students no longer received automatic credit for these assignments. Instead, they had to take a validation quiz following each report’s due date. I required Students to bring their report and coding assignment to class, where I gave them a 20-minute modification task (for example, “modify your code to display the output in reverse order”). Students who had genuinely authored their own code knew exactly where to make this modification within minutes. Students who did not understand the code they submitted were unable to complete the task in the time provided.

Within weeks, a pattern emerged. On an average week, approximately one-third of the class could not complete a simple modification of their submitted code during class, despite the fact that their submitted code was entirely correct. Had I graded the submission alone, those students would have earned an A. With the validation quiz in place, students received credit only for what they could demonstrate live in class. Those who submitted correct code but could not modify it received significantly lower grades, proportional to their quiz performance.

At the end of the semester, I was more confident than ever that grades reflected student understanding rather than student production. I subsequently implemented these practices across additional courses in my software development and cloud computing curriculum. The results were consistent: Asking students to modify, troubleshoot, defend, document, or discuss their work created an environment in which AI use without genuine understanding was insufficient to earn a passing grade.

Encouraged by these outcomes, I formalized the approach into a framework of strategies designed to be applicable across disciplines, class sizes, and instructional modalities.

The framework: Four strategies for grading the human

The following strategies share a common principle: The submitted artifact is not the sole basis for the grade. Instead, each strategy creates an additional assessment moment in which students must demonstrate understanding of their own work. While the examples below highlight specific disciplines, each strategy is cross-disciplinary in nature and can be scaled across modalities and class sizes.

Strategy 1: Interactive check-ins

Engage students in brief one-on-one sessions to discuss their submitted work.

Example (literature/humanities): Rather than grade a paper in isolation, review it during the meeting and pose reflective questions, such as “Why did you prioritize this theme over that one?” or “Why did you choose this particular word to describe the thesis?”

Strategy 2: Validation checkpoints

Following assignment submission, administer a proctored exercise (e.g., a quiz, concept map, or troubleshooting task) in which students modify or troubleshoot their own work or analyze a sample artifact that mirrors their assignment.

Example (science): After a laboratory experiment, set aside 15 minutes during class (or in a proctored environment) to hand-draw a concept map connecting their lab results to course theories or to troubleshoot a sample data set. “Create a concept map from your enzyme activity results” or “This sample graph matches your experiment but mislabels an axis. Troubleshoot it and redraw the link to cellular metabolism.”

Strategy 3: Process demonstrations

After they submit an assignment, require students to record and narrate their work process through video walkthroughs, screen recordings, or annotated revision timelines.

Example (writing): Review a student-submitted screen recording in which the student navigates their own draft, explaining the evolution of their thinking and identifying key turning points in their writing process. “Highlight three key sentences in your draft and explain why they are essential to your argument.”

Strategy 4: Reflective defenses

Following submission, require students to justify their choices, articulate their reasoning, and defend their methodological decisions in writing or discussion.

Example (calculus): Following a major assessment submission, schedule a proctored time window where students can present a defense; this is similar to scheduling a major assessment or final exam. Present a complete alternative solution using a different method. Students must compare both approaches, identify which sections of their submission they would revise, and justify whether their original method was superior. “This solution uses integration by parts; you used substitution. Compare both methods for this problem and determine which sections of your solution you would modify after seeing the alternative.”

Addressing the time concern

A common concern among faculty considering this approach is the additional time it takes to implement. In practice, the time investment is often comparable to existing grading workflows. The time an instructor spends reading, annotating, and writing feedback on an assignment is frequently equivalent to the time required for a five- to 10-minute conversation about that same work or for reviewing a video in which the student explains their process. The modality of assessment shifts, but the total time commitment need not increase substantially.

By asking students to modify, troubleshoot, justify, and defend their work, we can begin to relieve ourselves of the pressure of functioning as AI detectives. We can be more confident that the grades we assign reflect genuine learning rather than a student’s ability to prompt Claude or ChatGPT for a polished answer. We can, in short, grade the human rather than the homework.


Ashley Evans is a tenured professor of software development and cloud computing at Valencia College in Orlando, Florida. Learn more about this framework and other AI-ready teaching strategies at www.teachAIready.com.

The post Grading the Human, Not the Homework: An Assessment Framework for the AI Era appeared first on The Teaching Professor.