Workshop 2: Verifying STEM AI Outputs
The Check the Machine protocol and error-hunting skills for AI-generated code and STEM content.
Learning Objectives
- Identify common LLM failure modes in code and content
- Apply the Check the Machine (CtM) 4-step protocol
- Conduct structured error hunts
- Design a CtM classroom activity for their subject/grade
- Distinguish productive AI use (learning) from unproductive (copying)
Key Activities
- Code error hunt (breakout: find bugs in AI-generated code snippets)
- Content error hunt (audit AI-generated science/math explanations)
- Check the Machine protocol deep dive (Task → Before → After → Takeaway)
- CtM template customization for participant's classroom
- AI policy committee exercise (draft verification-based classroom norms)
Talk:Do Ratio
~25 min facilitator-led / ~155 min participant activities (1:6.2)
Participant Takeaways
- CtM template (editable)
- Content Audit Checklist
- AI Error Gallery (annotated examples)
- Prompt library for verify-able outputs
Date: April 18, 2026 · 8:30 AM – 12:00 PM PST · Virtual (Zoom)
Zoom: Join Workshop 2 (Password is in your calendar invitation.)
Focus: Academic integrity, error analysis, critical thinking in STEM
Talk:Do Ratio: ~25 min facilitator / ~155 min participant activities (1:6.2)
Your students are using AI — but do they know when it’s wrong? This session equips you with the “Check the Machine” protocol and hands-on error-hunting skills for both AI-generated code and STEM content.
Surveys
- 📝 Pre-Survey — complete at the start of the session
- 📊 Post-Survey — complete at the end of the session
Shared Workspace
- 📋 Group Workspace (Google Doc) — collaborative document for all breakout and group activities
- 🗓 Live Agenda (Google Doc) — facilitator-maintained timing board, updated in real time during the session
Learning Objectives
- Identify common LLM failure modes in code generation and STEM content
- Apply the Check the Machine (CtM) 4-step protocol: Task → Before → After → Takeaway
- Conduct structured error hunts on AI-generated code and scientific explanations
- Design a CtM classroom activity customized for their subject and grade band
- Distinguish productive AI use (learning) from unproductive AI use (copying)
Session Resources
- 📋 Planned Agenda — participant-facing timeline with activity links and print view
- 🖥 Web Slides — keyboard-navigable presentation (← → arrows, F for fullscreen)
- ⬇ Download Slides (PPTX) — import to Google Slides
- 🐛 AI Error Gallery — annotated code snippets with deliberate bugs for the error hunt
- 🔍 Content Audit Checklist — structured checklist for evaluating AI-generated explanations
- ✅ CtM Template — editable Check the Machine template for classroom use
- 🔬 AI Output Audit Workbench — fillable CtM-structured audit tool with AI-audits-AI round-trip and multi-format export
- 📚 Prompt Library — includes verification-focused prompts
- ⚖️ Platform Comparison — ChatGPT vs Claude vs Gemini cheat sheet
- 📄 CRAFT Cycle One-Pager — printable overview of the CRAFT framework
Code Snippets for Error Hunt
Downloadable Python scripts with deliberate bugs for the breakout error hunt activity:
Key Activities
| CRAFT Phase | Activity | Duration | Type |
|---|---|---|---|
| Contextualize | CRAFT Orientation + Self-Assessment | 15 min | You Do |
| Reframe | Poll + “The Real Problem” | 15 min | Listen |
| Reframe | Breakout: Rewriting the AI Policy | 15 min | You Do |
| Assemble | I Do + Breakout: Code Error Hunt (3 snippets) | 30 min | You Do |
| Assemble | Solo: Break Your Own AI | 15 min | You Do |
| Assemble | I Do + Breakout: Content Error Hunt (audit checklist) | 30 min | You Do |
| Fortify | CtM: Live Cycle + Partner Practice + Build Your Own | 30 min | You Do |
| Transfer | Debrief + Pair-Share Commitment + Post-Survey | 15 min | You Do |
The Check the Machine (CtM) Protocol
A reusable 4-step classroom verification framework from the CRAFT pedagogy:
- Task — What you asked the AI to do
- Before — Your expectation or prior belief about the answer
- After — What the AI actually produced
- Takeaway — What the comparison reveals about the tool AND about your own understanding
Participant Takeaways
- Check the Machine template (editable, customized for your grade level)
- Content Audit Checklist for evaluating AI-generated explanations
- AI Error Gallery with annotated examples across STEM subjects
- Prompt library designed to generate verify-able outputs
- Digital Toolkit PDF with all resources
Reframe Theme
“The problem isn’t that students USE AI — it’s that they don’t VERIFY AI.” — Banning teaches avoidance; verification teaches engineering thinking.
Acknowledgement
This material is based upon work supported by the National Science Foundation award #2230997. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.
Other Workshops: Workshop 1: AI for STEM · Workshop 3: Edge/IoT with AI
Your students don't need to be AI experts. They need to be AI skeptics. That's a skill you just learned to teach.