Iterative perturbation process
Definition
Iterative assignment perturbation is an instructor-led process of revising a question and testing a target chatbot’s responses against a rubric.
Explanation
Begin with the learning goal and the question students will solve. Choose the target model and establish how its responses will be graded. Test the original question, then revise it when the response meets the chosen threshold. Keep the wording and results of each version together.
The AIED work describes spatial injection, axiomatic replacement, and contextual embedding as perturbation strategies. The instructor needs domain knowledge to choose a change, judge its effect on the task, and grade the response. Making a question harder for the chatbot is useful only if the question still serves its instructional purpose.
The study used a 60% cutoff and three attempts to classify a tested question. Those are operational choices in the procedure. They do not establish that the question will resist every chatbot or remain resistant after a model update.
Open the Perturbation Workbench to record questions, rubrics, and attempts.
Scope and evidence
The process is described in the AIED paper. This page explains its purpose and decision logic; use the paper’s Table 1 for the formal eight-step protocol.
Related concepts
One-Shot GenAI Resistance · Three-Attempt Consistency Protocol · Contextual embedding