Overview
This assignment emerged from my experimental use of AI tools in my own research. I developed it to teach students to use AI as a thinking partner while maintaining human judgment in research design. For this assignment, students work in pairs to collaborate with AI tools to design research workflows, then test their methodologies on real sources. The activity has three main phases: AI-enhanced question generation and selection, refinement based on AI feedback, and implementation testing with document analysis tools.
I use the “Earth” chapter of Kate Crawford’s Atlas of AI as the background reading and frame the session around the environmental impacts of AI. The assignment can accommodate any complex topic requiring systematic analysis.
Assignment
Pre-Class Preparation
Read “Earth” in Kate Crawford, Atlas of AI.
Phase 1: AI-Enhanced Design
Step 1: Generate Options with Content Knowledge Working in pairs, choose any LLM and record which model you’re using. Begin with this context-rich prompt:
“Based on Kate Crawford’s analysis in Atlas of AI where she traces AI’s material infrastructure from lithium mining in Nevada to rare earth extraction in China, generate 10 different research question ideas for systematically analyzing the environmental impact of AI. Each question should cover a different dimension and help avoid common research mistakes.”
Step 2: Expert Assessment Ask the same LLM: “Now act as an expert researcher on the environmental impact of AI. Assess and rank these 10 questions according to these criteria: 1) Specificity: avoids vague language, 2) Buildability: provides foundation for follow-up questions, 3) Evidence-seeking: likely to yield concrete data from sources. Rank them 1-10 and explain your reasoning.”
Step 3: Human-in-the-Loop Selection Review the AI’s ranked list and choose 3 questions for your workflow. You don’t have to accept the AI’s top choices. Select questions that make sense based on your reading and judgment, or write your own, then explain your reasoning.
Step 4: Class Sharing Post to shared document: your AI’s top 3 recommendations, your actual selections, and whether you followed AI suggestions or made independent choices.
Phase 2: Refinement
Ask your LLM two critical questions: “As an expert on AI’s environmental impact, what are the biggest methodological problems with this workflow? What biases might we encounter? What types of evidence might be misleading or missing?”
“What assumptions might you be making about AI’s environmental impact that could limit these questions? What perspectives might you be missing?”
Revise your questions based on AI feedback and class discussion.
Phase 3: Testing and Analysis
Prediction Phase: Before testing, predict what each question will reveal based on your Crawford reading. Post predictions to shared document.
Implementation: Create Google LM Notebook, upload provided sources and test your questions. Also try one question from another pair’s workflow for comparison.
Challenge Round: Identify something in sources that contradicts or complicates your background reading.
Class Discussion and Reflection
Review discoveries and methodological insights across all pairs.
Students address these questions in a brief written reflection:
- What was different from what you expected when testing your questions?
- Which of your 3 questions worked best and why?
- Did you choose different questions than the AI recommended? What informed your choices?
- What contradicted or complicated your understanding of the topic based on the Crawford reading?
- What surprised you most about working with AI for research design rather than getting answers?
- How could you apply this systematic questioning approach to other academic topics?
This assignment helps students experiment with AI for process improvement rather than content generation. It reminds them to remain in control of the research process even when collaborating with AI.
