Appendix E — On Writing in the Age of AI

Writing as thinking and learning

Good writing clarifies thought. Writing is difficult because it forces us to confront our own thoughts, rationale, and reasoning. This is just one reason why writing is learning (Zinsser, 1993,Yan, 2020, Ahrens, 2022). Good writing is a feedback loop of multiple processes - reading, drafting, reflecting, reading, drafting, revising, revising, revising, revising, reflecting, writing, etc. In this course, writing is not a last-minute wrapper; it will be an integrated part of what we do, how we integrate, and how we think. Some of the writing may be AI assisted - and some of that will be your own (in which I will ask you to turn in an audit trail of reading notes and comments on our class textbook wiki).

What AI changes—and what it doesn’t

  • AI can help you brainstorm, outline, critique, and rephrase.
  • AI cannot understand your specific field data, lab context, or instructor’s expectations unless you do.
  • You are responsible for factual accuracy, citations, and integrity.

Our separate AI Policy covers boundaries (originality, attribution, and prohibited uses). This chapter focuses on practice.

AI: tools vs. foundations

  • Foundation models (e.g., general LLMs) learn broad patterns from massive data; apps (chatbots, copilots) wrap those models with UX and guardrails.
  • You should know the difference well enough to evaluate reliability, bias, and when to escalate to a human expert.
  • We will use AI tools to accelerate thinking and learning—but not to replace reading, reasoning, or evidence.

Required AI use log (attach with any assignment submissions where AI was used)

Include a brief section at the end of your assignment document as follows:

AI Use Log
- Models: (e.g., GPT-5, Claude, Gemini)
- Purpose(s): brainstorming, outline critique, code debugging, copyediting
- Prompts (or gist): 
- What I kept vs. discarded:
- Verification steps (how I checked claims / code):
- Date(s):

A pragmatic workflow

  1. Define the deliverable (audience, purpose, constraints).
  2. Outline yourself: produce a simple structure (thesis → sections → evidence).
  3. Draft 0 with AI (optional): ask for a skeletal outline or a list of counterarguments.
  4. Write Draft 1 yourself: fill sections using your results, figures, and citations.
  5. Targeted AI passes:
    • “Point out unclear claims and missing evidence.”
    • “Suggest stronger topic sentences for each paragraph.”
    • “Propose 3 alternative figure captions grounded in the text.”
  6. Fact-check & code-check: run your analysis; verify numbers and references.
  7. Polish: style, transitions, title, abstract.

On “prompt engineering” for writing

Much of modern writing with AI is asking better questions. You’ll get better results by being explicit about tone, audience, structure, constraints, and evaluation criteria (see the Prompt Engineering appendix).

Citing AI influences

When AI meaningfully shapes a paragraph or figure, treat it like any other tool or collaborator: acknowledge it in the AI Use Log and ensure all sources are properly cited.

Two podcast perspectives to consider

  • On study craft and deliberate practice in the LLM era (and why “just reading” isn’t enough).
  • On transforming prompts into better prompts and using AI to build a reusable memory system.

(See references for specific episodes.)

References