Lab 4: Auditing an AI-Assisted Workflow

Author

Rony Rodriguez-Ramirez

Published

04 September 2026

Introduction

In this lab, you will complete a small end-to-end analysis where AI is allowed to help but every meaningful result must still be checked, justified, and documented by you.

Part 1: Define the task first

Before using any AI tool, write down:

  • your research question,
  • the unit of analysis,
  • the main outcome,
  • at least two predictors or grouping variables,
  • one likely data-quality problem.

Part 2: Use AI for one scoped coding task

Ask for help with one bounded task such as:

  • cleaning the data,
  • building a missingness summary,
  • generating grouped statistics,
  • drafting a plotting function,
  • proposing a simple model specification.

Part 3: Audit the generated output

Answer these questions:

  1. Did the code run without modification?
  2. What did you have to fix?
  3. Did the tool make assumptions you did not ask for?
  4. Would you trust this output without reading it carefully?

Part 4: Produce one verified result

Create one output:

  • a summary table,
  • a visualization,
  • a simple model,
  • or a coded text table.

Then verify it manually by recomputing one number, checking raw rows, or comparing the result to a simpler baseline.

Part 5: Reflection

Write a short reflection on:

  • where AI saved time,
  • where it created extra checking work,
  • which prompt style worked best,
  • which fall-semester task you would use AI for again,
  • which task you would approach more cautiously.