Week 1: Generative AI for Analysts

Module: Understand | Deliverable: Use case reflection (due before Week 2 begins) | See assignment →

Learning objectives

By the end of this week, you should be able to:

Explain what generative AI and LLMs are at a high level
Distinguish using AI as a tool from delegating judgment to AI
Identify analyst tasks where GenAI can be useful and where it can go wrong
Score GenAI use cases on five risk dimensions quantitatively and compute a composite risk score
Compare risk scores across groups and identify sources of disagreement
Describe the course architecture: understand, use, evaluate, govern
Make API calls to an LLM in Python, including batch processing with structured JSON output
Parse AI-generated responses and compute summary statistics from the results

Sessions

Session 1 · Class 1

What does GenAI change about analytic work?

We'll start with introductions and a discussion of where you've already used AI tools. Then a short lecture on how GenAI is reshaping analyst workflows - and why fluency is not the same as accuracy. The core activity is a quantitative risk-scoring exercise: you'll rate use cases on five dimensions (1-5 each), compute composite scores, and compare your ratings with another group's.

Opening discussion Mini-lecture: GenAI for analysts Quantitative risk scoring Cross-group comparison

Session 2 · Class 2

First hands-on: seeing what GenAI does

We'll introduce the course structure (understand, use, evaluate, govern), then get hands-on with Python. You'll make your first API calls, experiment with temperature, and process a batch of 10 texts programmatically by parsing structured JSON output and analyzing the results as a dataset. We'll also do a prompt diagnosis exercise.

Course structure Python: API calls and temperature Python: batch processing with structured output Prompt diagnosis activity

Prepare before class

Before Session 1:

No readings this week - come ready to discuss where you’ve already used AI tools (ChatGPT, Claude, Copilot, Gemini, or others)
Think about one time AI was useful and one time it failed or surprised you

Before Session 2:

Check that you can access Google Colab
The Python notebook link will be posted on Canvas

Key concepts

Concept	What it means
Fluency ≠ accuracy	GenAI output can sound confident, polished, and professional while being factually wrong, missing context, or fabricated
Analyst responsibility	AI can produce work products; analysts remain responsible for quality, verification, documentation, interpretation, and accountability
Risk dimensions	Stakes, data sensitivity, affected population, reversibility, and degree of automation - scored 1–5 each for a composite risk score
Structured output	Requesting JSON-formatted responses from LLMs so results can be parsed and analyzed as data
Understand / Use / Evaluate / Govern	The four-verb framework that organizes every topic, reading, and assignment in this course

Readings and resources

This week’s readings are optional with the main learning happening in class activities. That said, I highly recommend reading them.

Békés, Gábor. Data Analysis with AI: Course overview - skim the course philosophy to understand the applied framework we’ll build on
Microsoft. Generative AI for Beginners, Lesson 1 - accessible introduction to GenAI concepts

Deliverable

GenAI Use Case Reflection

~1 page + scoring table | 30 points | Due before Week 2's first class | Submit via Canvas

Describe one realistic GenAI use case in a future analyst role. Explain what AI would do, what the human analyst would still be responsible for, what would need to be verified, and the risk level. This will include a quantitative score on the five risk dimensions.

Full prompt and rubric →

Looking ahead

Next week we’ll move into LLMs, harnesses, and setup - you’ll learn the conceptual model behind how these systems work (tokens, context windows, model landscape), set up your development environment, and run a systematic model benchmarking exercise. See the setup guide to get a head start.