Why we still can't stop plagiarism in undergraduate computer science (2018)
Universities are struggling to curb plagiarism and AI‑assisted cheating in undergraduate computer science, prompting questions about what grades and degrees actually certify. Commenters contrast resource‑intensive but effective approaches—like in‑person coding exams, oral defenses, and heavily weighted finals—with homework-heavy models that are easy to game and often misaligned with learning. Many argue the root problem is structural: institutions are incentivized to sell credentials rather than fail weak students, so cheating persists until employers or accreditation pressures force a shift toward more robust, harder-to-cheat assessment.
Project‑based and Exam‑heavy Approaches
- Several comments endorse ungraded or low‑weight homework plus:
- A substantial project built over the term.
- A timed, in‑person practical where each student must modify their own project; tasks are chosen to both prove authorship and stress-test design/complexity.
- Variants: oral/whiteboard exams, viva voce defenses of projects, pen‑and‑paper finals, and “no take‑home” weekly in‑class assignments.
- Benefits: plagiarism becomes pointless, understanding and architecture are directly tested.
- Costs: extremely time‑ and staff‑intensive, hard to scale, often “brutal” with lower pass rates; fairness and accessibility issues (e.g., large cohorts, weaker language skills).
Role and Weight of Homework
- One camp: homework should be primarily for practice; grades should come mostly or entirely from proctored exams.
- Optional or low‑weight homework often leads to more exam failures, but that’s seen by some as the student’s responsibility.
- Another camp: the deepest learning and “real‑world” skills come from large, graded projects and sustained homework; exams can’t fully measure that.
- Suggested compromises: homework to qualify for the exam (or provide bonus points), or multi‑part assignments where suspected plagiarists get extra work.
AI/LLMs and Changing Cheating Patterns
- Many note that traditional plagiarism signals (identical code, whitespace quirks) are largely obsolete; LLMs can generate and “rewrite” solutions.
- Instructors report:
- More students getting perfect homework scores and then failing exams.
- Students turning in AI‑generated work they cannot explain in oral exams.
- Proposed responses: heavily exam‑weighted grading, in‑lab coding with logging/keystroke replay, and using LLMs to generate many variant problems.
Incentives, Institutions, and Culture
- Strong view that degree value as a hiring filter drives cheating: when the diploma matters more than the learning, cheating is rational.
- Some argue universities, especially revenue‑driven ones with many international students, have weak incentives to crack down hard; enforcement and sanctions are often mild.
- Others insist institutions must protect their signal: unchecked cheating will erode program reputation and harm honest students.
Honor Codes, Ethics, and Empathy
- Honor codes are seen as:
- Weak direct deterrents but useful as legal/administrative evidence that students knew the rules.
- Culturally dependent; cheating remains common in many “honor code” environments.
- Debate over how much to factor desperation, mental health, and unequal preparation into responses to cheating:
- One side emphasizes strict, consistent consequences to protect trust.
- The other stresses understanding underlying causes and avoiding life‑ruining penalties for a single bad decision.