5 Research Responsibility and Intellectual Ownership
This chapter is part of a book in active development and has not yet been through the author’s review. Content may change as the review advances.
The research decision. For every sentence an AI helped produce, you decide two things: whether you will personally stand behind it as your own, and what your disclosure will say about how it came to be. Your name on the work means you answer for it, whatever tool helped.
5.1 Why this decision matters
The decision on the table: which sentences in your work you are prepared to answer for as your own, and what you tell your reader about how each one was made.
“When something in this paper turns out to be wrong, I do not ask which software wrote it. I ask whose name is on it. That person answers for every line.” — a journal editor, to a first-time author
You will finish this book with work that carries your name: a poster or a one-figure summary, a research note, and a defense of both in front of people who can ask you anything. An AI tool may have helped you find sources, draft a paragraph, or compute a number. None of that moves the responsibility off you. The reviewer, the reader, and your future self all hold you, the named author, accountable. This chapter draws that line on purpose, before someone else draws it for you.
5.2 The concept
Intellectual ownership is being the author of, and answerable for, every claim that appears under your name, whatever tool helped produce it (International Committee of Medical Journal Editors 2026). Example: if an AI drafts your methods paragraph and it states that creek chloride exceeded a safety threshold, you own that sentence, and if it is wrong, the mistake is yours.
Ownership rests on accountability, being the person who answers for the work’s correctness and integrity when it is questioned (National Academies of Sciences, Engineering, and Medicine 2017). Example: at your defense, when someone asks how you know a number, “the AI computed it” is not an answer. “I recomputed it from my own measurements, and here is the check” is.
The habit that makes accountability visible is AI-use disclosure, a short, honest record of which AI tool did which task and how you verified its output (Allen et al. 2014). Example: “an AI assistant located candidate sources; I retrieved and confirmed each one in the library catalog.” Disclosure is not a confession. It protects you, because it shows exactly where your own judgment entered the work.
Disclosure has gotten harder in one specific way, and it is worth naming. You rarely send one prompt anymore. You run a loop: you ask, you read, you push back, you ask again, and sometimes an agentic tool runs dozens of those cycles on its own and hands you a finished result. “I used AI for analysis” describes none of that. An honest disclosure says what you specified going in, roughly what the loop did, and what you opened and checked when it stopped. Your ledger is what makes writing that sentence possible, because nobody remembers cycle eleven a month later.
Some calls never move to the tool at all. A never-delegate decision is a judgment that stays yours no matter how capable the tool becomes: which question you pursue, what your evidence licenses, the ethics of the study, and the decision to put your name on the work. Example: an AI can list candidate measures of water quality, but choosing the one you will defend is yours.
Put together, the rule is simple. Use AI freely for reach and speed. Keep ownership, accountability, and honest disclosure for yourself.
5.3 A worked example
Picture a small freshwater-biology project. You are testing whether winter road salt raises chloride, a dissolved salt ion that harms freshwater life above certain levels, in a creek behind campus. You collect ten water samples and ask an AI tool to help write up the result.
The tool returns a fluent paragraph: “Mean chloride was 310 mg/L, well above the U.S. EPA chronic aquatic-life criterion of 230 mg/L, so the creek is unsafe for aquatic life.” It reads like a finished result. Here is the ownership decision: which parts of that sentence will you personally stand behind?
You split it into pieces you can own. The mean of your ten samples is a number you can recompute in a minute, so you do, and you get 304, not 310. The EPA criterion is a published value, so you open the EPA’s National Recommended Water Quality Criteria and confirm the real figure yourself rather than trust the tool’s number. The word “unsafe” claims something about the whole creek all winter, from ten samples on one day, so you narrow it to what your evidence licenses: elevated chloride in these samples, at this site, on this date.
Now your disclosure writes itself: the tool drafted the paragraph, and you did the checking, retrieving, and rewriting that made the claim yours. You own the corrected sentence. You would not have owned the original.
That authorship cannot be delegated to a tool, and that the named author answers for every claim, is the position taken by editorial guidance on AI use (International Committee of Medical Journal Editors 2026).
The block below builds the ten samples and recomputes the mean, which is the whole of the one-minute check. Run it before you read the next line.
import numpy as np, pandas as pd
SEED = 464
rng = np.random.default_rng(SEED)
# Ten chloride samples (mg/L) from one site, on one day.
chloride = np.round(rng.normal(297, 18, size=10), 1)
print(pd.DataFrame({"sample": range(1, 11),
"chloride mg/L": chloride}).to_string(index=False))
print(f"\nmean of your ten samples : {chloride.mean():.0f} mg/L")
print("the drafted paragraph said: 310 mg/L")
print("the criterion it quoted : retrieve it yourself, then compare")5.4 An AI failure case
You ask the tool to write your AI-use disclosure for you. It produces a tidy, confident sentence: “AI was used only for minor language editing.” It sounds professional, and it is false. The tool actually drafted your entire results paragraph, including the 310 mg/L figure and the “unsafe” claim. Left unchecked, that disclosure understates the tool’s role and overstates your ownership, which is exactly the integrity problem disclosure exists to prevent.
You catch it with a record the tool cannot see: your own AI Research Ledger. Lay the one-line disclosure next to your ledger rows and the mismatch is plain. The ledger says the tool drafted the analysis; the disclosure says it only edited language. You rewrite the disclosure to match what happened. The lesson is blunt: never let a tool describe its own role. You describe it, from your record.
5.5 It is your turn
You are working inside Studio 2: Set your rules, shape your question. Keep what you write here; the studio’s milestone chapter is where it joins the other lessons’ pieces into one artifact you can defend.
You have a ledger, a director’s task list, and one delegation you ran end to end. This step puts your name on all of it, on purpose, in writing.
The hands-on half of this section lives in the chapter’s companion notebook: open it in Colab with the badge at the top, and work the steps there.
Commit your own answer first, then delegate. Each prompt below hands the tool reach, not judgment.
The tool may fetch, draft, and compute. It may never decide which claims you put your name on, whether your evidence supports a word like “unsafe,” the ethics of your sampling, how you state your uncertainty, or what your AI-use disclosure says. Authorship is a responsibility, not a task, and it does not transfer.
Write your ownership statement: three or four sentences naming the work you intend to put your name on, what you will personally answer for, and what you will not claim to have done. Start with the question you are about to declare, because its words are the first thing your name goes on.
Write your never-delegate list for this specific project. Five lines is plenty. Make each one a call, not a topic: which question you pursue, what counts as a fair comparison, where the claim stops. Keep the list somewhere you will actually see it.
Draft the AI-use disclosure you would attach to your work as it stands today, built from your ledger rows rather than from memory. Name each tool, what it did, and how you checked it. If an agent ran a long loop on your behalf, say so, and say what you opened and inspected when it finished.
Read the disclosure and the ledger side by side. Wherever they disagree, the ledger is right. Fix the disclosure.
Find one sentence you have already written that you could not defend if someone asked how you know it is true. Produce the check, or cut the sentence.
List every claim to verify.
Here is my draft results paragraph: [paste]. List every empirical claim in it as a table: the claim, the number or source it rests on, and whether that number came from my data or from you. Do not add any new claims.After running, verify: check the table against your own list of what you actually measured. Counters illusion of completeness, a claim the draft slipped in that never shows up in the tool’s own accounting.
A second angle, optional:
Optional depth. The milestone needs only this lesson’s core cycle; run this second angle when you want more practice.
Red-team your ownership.
Act as a hostile reviewer. Read my draft and name every sentence I could NOT defend myself if someone asked how I know it is true. Do not rewrite anything. List the weak spots.After running, verify: for each flagged sentence, either produce your own check or cut the claim. Counters sycophantic agreement, praise that reviews your ego instead of your evidence.
Log all of it in your AI Research Ledger, and verify at least one surviving claim with a named method from the Verification Guide: primary-source reading if it rests on a source, direct calculation if it rests on a number. An AI reviewer may run the check with you; the decision to accept or reject stays yours.
Optional depth. The milestone needs only this lesson’s core cycle; run this second angle when you want more practice.
Locate the real source.
Act as a research assistant in freshwater ecology. I need the current U.S. EPA aquatic-life criterion for chloride. Give the exact value, the document it comes from, and a link. Only state a figure you are confident is published; if you are unsure, say so and tell me where to look.After running, verify: open the EPA document and read the figure yourself (primary-source reading). Counters confident fabrication, a number quoted with a made-up source.
You have rules, a ledger, a verified delegation, and an ownership statement. You are ready to declare the formal question those protections will serve.