Crossed Wires: The lost art of contemplation - where AI is failing education
AI makes student writing cleaner, clearer, and more consistent. It also risks cutting the cord between writing and thinking - the very link that turns information into insight.
That tension showed up in a graduate course where students were allowed to use AI freely. Every paper was polished. Language barriers vanished. The grading bias that comes from clunky English disappeared. But it raised a tougher question: if AI carries the prose, are students still doing the hard work of thinking?
The upside is obvious
Open AI policies level the field for multilingual classrooms and reduce surface-level penalties. With strong rubrics, students can produce well-structured arguments quicker and spend time exploring sources. Even grading can benefit: AI can draft comments, flag gaps, and align to a rubric with surprising accuracy.
Used well, this is productive. Used uncritically, it creates a polished shell over shallow ideas.
The cost we avoid talking about
Writing is not a wrapper for thought - it is the lab where thought forms. When students outsource drafting too early, they shortcut the struggle that builds judgment. The result is a thin mix of quotes, summaries, and borrowed analysis that's easy to read and easy to forget.
Educators who practice "slow teaching" argue for protected time, fewer digital prods, and deliberate reflection. The goal is depth, not speed. If that resonates, this book is worth a look: The Slow Professor.
Design principle: separate stages, not outcomes
AI can stay in the process - just not at the start. Move it from "ghostwriter" to "coach" and "checker." Require artifacts that prove thought actually happened.
A practical, AI-aware writing workflow for students
- Stage 1 - Think first (no AI): 30 minutes offline. Draft a working thesis, a rough outline, and three questions they still can't answer.
- Stage 2 - Write from memory (no AI): 45-60 minutes to produce a rough first section (300-500 words). Imperfect is fine.
- Stage 3 - Research and map: Gather sources, build a concept map, and identify counterarguments.
- Stage 4 - Use AI as a challenger: Ask for critiques, missing perspectives, and clarity checks. No full-paragraph generation. Capture prompts and outputs.
- Stage 5 - Revise with receipts: Submit the final draft plus a short reflection: what changed in your thinking and why.
Rubrics that reward contemplation (not just polish)
- Synthesis over summary: Are ideas combined into something new?
- Counterargument quality: Are the strongest opposing views engaged fairly?
- Process evidence: Outline, first-draft excerpt, prompt log, revision notes.
- Transfer: Application to a novel case, local data, or lived experience.
Use AI to speed feedback, not learning
- Draft feedback with AI, decide as a human: Let the model propose comments tied to the rubric; you accept, edit, or reject.
- Calibrate first: Co-grade a small set manually, then compare AI-assisted output to your standard.
- Spot-check with oral defenses: Five-minute viva to verify authorship and depth on the paper's core claims.
Clarify your AI policy (students need guardrails)
- Allowed: Idea prompts, outlines, critique prompts, grammar cleanup, citation formatting.
- Restricted: Full-paragraph generation, undisclosed paraphrasing, fake citations.
- Required disclosures: Tools used, prompts, and where AI influenced structure or claims.
- Consequences: Clear, proportional, and consistent with existing academic integrity rules.
Reduce LMS noise to make room for thinking
- Batch announcements into a weekly digest. Turn off push notifications that don't affect deadlines.
- Block out "deep work" windows with no submissions or reminders.
- Prefer fewer, larger milestones over constant micro-deadlines.
This isn't just about vibes. Cognitive overload degrades learning; see cognitive load for the basics.
Assessment ideas that survive AI
- Localize: Use campus data, community issues, or course-specific datasets models won't have seen.
- Constrain: In-class concept maps, whiteboard derivations, or annotated problem-solution traces.
- Iterate: Multi-part assignments where Part B depends on feedback from Part A.
- Defend: Short oral or screencast walkthroughs of key claims or methods.
A quick policy snippet you can adapt
"AI may be used to plan, critique, and proofread. Your first 400 words must be written without AI. You must submit a prompt log and mark AI-influenced sections in the margin. Full-paragraph generation without disclosure is a violation. Grading prioritizes synthesis, counterargument quality, and evidence of revision."
What about grading with AI?
Testing a rubric with an AI grader can produce helpful, consistent comments. It will likely mirror your scoring on structure and surface issues. Keep the final call human, and use the saved time for richer feedback and short orals.
Efficiency is useful. Satisfaction comes from seeing a student think sharper over drafts. Keep that part human.
Faculty development and next steps
- Pilot the workflow above in one assignment this term and collect student reflections.
- Run a 45-minute workshop on AI disclosures and "coach-not-ghostwriter" usage.
- Co-create rubrics with your cohort so criteria are visible and owned.
If you're building faculty capacity for prompt design, feedback workflows, and policy, these resources can help: AI courses by job.
Your membership also unlocks:
