Quiz — Chapter 24: Grant Writing with AI
Answer from memory, then check. These test what AI is, where it helps and fails, the human-in-the-loop discipline, the confidentiality red lines, and disclosure.
1. Which best captures the threshold concept of this chapter? a) AI can write a fundable proposal if the model is good enough. b) AI can sharpen your argument but cannot own it — accountability stays with you. c) AI should never be used in grant writing. d) Disclosing AI use makes any use acceptable.
Answer
(b). AI helps you think, structure, and edit, but it cannot supply original ideas, real data, funder judgment, or accountability — you remain the author of every claim.
2. What is a large language model, and why does it hallucinate?
Answer
An LLM predicts plausible language, not verified truth. Because it generates what sounds right with no built-in mechanism for accuracy, plausible-but-false output (invented citations, statistics, findings) is intrinsic to how it works — and it sounds just as confident when wrong as when right.
3. What property makes an AI use safe?
Answer
The output is easy for you to verify and you supply the substance — e.g., editing your own prose, brainstorming options you judge, summarizing documents you provide, devil's-advocate critique. You remain the judge of truth and quality.
4. What is the single most important and most violated rule in the chapter?
Answer
Never include an AI-provided citation or statistic without independently verifying it against the real source. AI fabricates flawlessly-formatted references and confident statistics; treat every one as a lead to verify, never a citation to use.
5. Why is including AI-generated "preliminary data" a serious problem?
Answer
It is fabrication — research misconduct. AI has no access to your lab, pilot, or program outcomes; any data it "produces" is fiction. Your preliminary data must be real (Chapter 9).
6. Name the four commitments of human-in-the-loop.
Answer
(1) Verify every fact, citation, number, and claim against a real source; (2) supply the substance and judgment yourself; (3) own the voice (rewrite AI drafts into your specific voice); (4) remain accountable — "the AI did it" is never a defense.
7. State the peer-review red line.
Answer
A reviewer must never upload a confidential proposal they are reviewing into a public AI tool. It breaches the applicant's confidentiality, may expose their unpublished ideas, and violates peer-review integrity — and major funders prohibit it.
8. Why is AI's generic, voiceless prose a liability even when it's accurate?
Answer
Your reader is a specific, often tired human reviewer (Chapter 2) moved by concrete, specific, passionate writing. Generic AI boilerplate reads as "anyone could have written this about any project" — the opposite of what engages a reviewer and wins funding. Rewrite it into your real voice.
9. Why can't you build a budget from AI estimates?
Answer
AI doesn't know your actual costs, your institution's indirect rate, your funder's budget rules, or your real timeline. A budget is precise, verifiable, and accountable (Chapters 11–12); AI guesses, and its guesses will be wrong in ways that matter. Build from real numbers.
10. How should you handle confidential or sensitive material with AI tools?
Answer
Don't paste it into public AI tools, which may retain or train on it — protecting unpublished ideas, others' confidential information, personal data, and proprietary content. Treat a public tool as a non-private channel unless your institution has vetted contractual privacy guarantees.
11. How do you decide about AI disclosure?
Answer
Follow your funder's current policy (which varies and evolves); where it requires disclosure, disclose; when unsure, lean toward transparency. Never use AI in a funder-prohibited way or hide a use you suspect wouldn't be permitted. Accountability is stable regardless of policy.
12. A perfectly-formatted AI citation looks completely real. Why is that exactly the danger?
Answer
Fluency disarms skepticism — a fabricated citation indistinguishable from a real one slides into your reference list, and the reviewer who can't find it concludes you fabricated it, collapsing the whole proposal's credibility. Fluency is no evidence of accuracy; the better it looks, the more (not less) you must verify.
13. (Synthesis) In one sentence, what defines whether an AI-assisted proposal is authentically "yours"?
Answer
It's yours when you originated the substance (ideas, real data, funder strategy) and can stand behind every word as your own accountable claim — even if AI helped express it; it stops being yours when AI supplied substance, data, or citations you didn't verify or can't stand behind. Accountability, not the tool, defines authorship.