AI Automation ·
How can large Korean classes assess AI-assisted homework fairly through oral checks?
On November 15, The Cornell Daily Sun described an engineering course that uses a 20-minute oral defense with the professor or a teaching assistant for each assignment [1].
Problem definition and research question
A large required course in Korea needs evidence that individual students can explain a circuit analysis while retaining problem sets. Can instructors reduce inflated grades from answer-only marking without imposing unsustainable TA hours?
Evidence and causal analysis
The reporting describes a real course [1]. Cornell’s teaching center helps instructors set course-specific AI rules [2], and Korea University’s notice describes teaching materials organized by stage of instruction [3]. Neither establishes a university-wide oral defense requirement or a causal gain in learning. Oral fluency and anxiety could affect marks independently of conceptual understanding.
| Dimension | International evidence | Korean evidence and application |
|---|---|---|
| Evidence | One Cornell course uses 20-minute sessions [1] | Korea University provides instructional resources [3]; a universal oral assessment policy is unconfirmed |
| Assessment implication | Oral defenses count toward course grades [1] | Test staffing and accessibility in large Korean sections |
Comparison of alternatives
Option A assesses every student orally using a common rubric. It avoids selective sampling and gives individual evidence but takes more TA hours. Option B gives all students a short in-class transfer problem and samples low-stakes oral checks using a published random procedure. It saves staff time while observing fewer individual explanations. Departments should own the rubric and accommodations and provide comparable access to any graded check.
Implementation and evaluation
Rotate the approach across sections and units. Blind-grade a circuit transfer question students have not seen on homework. Compare performance, paid TA hours, appeals and inter-rater agreement. If explanations improve but gaps related to language or disability widen, revise written equivalents and train raters before expanding. These are proposed pilot criteria.
Conclusion and open questions
The goal is to establish whether homework grades represent understanding. A full rollout depends on staff time per student and gains on transfer items. Can an accessible alternative assess the same reasoning?