Using AI

AI tools are welcome in this course. You do not need to hide them, apologize for them, or pretend you worked without them. What you do need is to stay the author of your own understanding.

The rule, in one line: use AI if it helps you learn. Disclose how you used it, and be able to reproduce, modify, and explain your own work without it.

AI is a legitimate tool here, like Excel or Solver. Disclosure and reproducibility are what make it safe: honest disclosure is never penalized, and work you can’t rebuild, change, and explain unaided isn’t yours yet.

What this course certifies

V348 does not certify that you can compute a number. Software has done the computing for decades, and AI now does it instantly. What this course certifies is the modeling cycle: turning a messy public-sector problem into a model, testing it, knowing when its assumptions break, and advising a decision-maker who has to act. AI can produce a number. It cannot own that judgment on your behalf, and the moment you are in a real job, nobody will ask you for the number alone.

So this policy is not about catching you. It exists to make sure two things stay true: your grade reflects what you can actually do, and a classmate who leans on AI cannot out-earn you for less work.

How each task is labeled

Every graded task tells you which tier it is. Regular take-homes show it on the workbook’s Instructions sheet; the course-synthesis brief states it directly:

Acceptable ways to use AI

All of these leave you able to reproduce your own work, which is the test that matters:

• Ask it to explain a concept you didn’t catch in class. What a shadow price actually means, why EVPI is a ceiling, why a moving average lags a rising series.
• Ask it to debug a formula or an Excel/Solver error that is not behaving. Getting Solver to run is not the skill being assessed; interpreting what it returns is.
• Ask it to generate extra practice problems so you can drill a method before the take-home.
• Check your arithmetic after you have done the work yourself: a second pair of eyes, not a substitute.
• Ask it to critique your model: “what assumption am I making here that could be wrong?” This is a genuinely good use, and it is close to what the course is teaching you to do.
• Tighten writing you drafted: clarity edits on a recommendation whose substance is yours.

Unacceptable ways to use AI

• Pasting the assignment or the Instructions sheet into a chatbot and submitting what comes back.
• Submitting a model you cannot rebuild, modify, or explain. This is the core violation. The written prompts and the in-class checkpoints are both built to find it.
• Using AI during a Tier 0 in-class checkpoint.
• Leaving the disclosure blank, or writing a false one. In this course, non-disclosure is the violation, not use. You will not lose points for saying you used AI. You will for hiding it.
• Sharing a solved workbook, or a chat transcript containing one, with classmates.
• Passing off AI-written interpretation as your judgment without engaging with the numbers in your own model.

The disclosure line

Every take-home workbook has a disclosure cell on its Free responses sheet. Fill it in honestly:

I used [tool] to [what for]. One thing it got wrong, or that I had to fix: [what].

“I didn’t use AI on this one” is a perfectly good answer. And if you did use it, the “what I had to fix” part is often where the real learning shows up. Noticing that the tool sampled the wrong distribution, or quietly rounded a Solver result in a way that broke a constraint, is exactly the skill this course is after.

How this course is designed so AI doesn’t decide your grade

You should know how the assessment works, because it explains why honest AI use costs you nothing here and dishonest use buys you very little.

1. In-class checkpoints. Unaided. Six times in the term you answer a short set of questions drawn at random from the self-checks and take-homes you have already worked. Nothing on one is new, and nothing about it is a trap; what it asks is whether you can do it without help. A graded component in its own right, and the page has the detail.
2. Questions that depend on your own numbers. Most written prompts hinge on a value that exists only in your solved model: your shadow price, your cumulative error, your probability of a loss. A confident, generic answer is not merely weaker here; it is usually wrong.
3. Debug-the-model tasks. Sometimes you’ll be handed a plausible but subtly broken model and challenged to find and fix the flaw. AI is allowed. It is still hard, and it is the most realistic task in the course, because in practice you will far more often inherit someone else’s model than build one from scratch.

Work you cannot reproduce or explain unaided may not receive credit.

If you’re stuck

Reaching for a chatbot because you are lost is understandable, but it is often the slower path. It gets you a submitted file and leaves you no better prepared for the in-class check. Come to office hours, ask your TA, work the ungraded self-checks (retry them as often as you like), or bring the problem to the in-class group practice. Asking a person is not a fallback here; it is the faster route.

Academic integrity

Submitting work you cannot reproduce or explain as your own, and failing to disclose AI use, are academic-integrity violations handled under IU policy. If you are ever unsure whether a particular use is acceptable, ask us before you submit. That conversation has never once gone badly for a student who asked first.

What you’re allowed to paste into a chatbot. Everything published on this website: these pages, the week pages, the self-checks, and the starter .xlsx workbooks and .docx practice files you download from them. O’Neill requires every syllabus to say that putting instructor-produced course materials into AI needs the instructor’s permission, so here is that permission, in writing. It does not stretch to the Tier 0 in-class checkpoints, to another student’s work, or to anything handed out in class that isn’t published here.