Exploring AI Tools, Evaluating AI Claims, and Discovery Interviews. Your first hands-on session using AI to research a real social problem and practice the kind of questions that get honest answers.
"Every big change starts as one small, deliberate action."
First Ripple
Student Access
Token provided in class or on Canvas.
Professor Meirelles van Vliet
Fall 2026 · T/R 3:30-4:45 PM · KWR 310
SE 395-2: AI Tools for Social Impact
01
Week 1 · Thursday · AI Lab
Exploring AI Tools, Evaluating AI Claims, and Discovery Interviews
Your first hands-on session with AI. You will research a real social problem, learn to separate hype from evidence, and practice the kind of questions that get honest answers.
Time
60 minutes in class
Tools
Claude, ChatGPT, or Perplexity (free)
Due
Sunday, Aug 31, 11:59 PM
Overview
What This Lab Is For
This is your first hands-on session with AI tools. By the end of this lab you will have used an AI tool to research a real social problem, evaluated whether certain AI claims hold up under scrutiny, and practiced writing the kind of interview questions that produce honest, useful answers instead of polite agreement.
The deeper goal is not about learning any particular tool. What you are really building is a way of working with AI where you treat it as a collaborator you can argue with, rather than a search bar that hands you finished answers. The difference shows up in the quality of your thinking: when you use AI well, you come out of the conversation with sharper questions about the problem you are studying, not just a pile of text you could have found on Google.
No experience required. If you have never used an AI tool before, you are in good company. This lab walks you through everything step by step, and you can raise your hand any time you get stuck.
Setup
Before You Start
Open one of the tools below in your browser and create a free account. You will need to be logged in so you can save and export your conversation for the submission.
Tool
URL
Best for
Claude
claude.ai
Nuanced analysis and longer conversations
ChatGPT
chat.openai.com
General purpose, the most widely used
Perplexity
perplexity.ai
Research with sources attached to answers
You can use any one of these or try more than one. Part of this lab is noticing how different tools respond to the same question. If you already have a preferred AI tool, feel free to use that instead.
AI tools fabricate things. They produce text that sounds confident and specific even when the underlying facts are wrong. This happens with statistics, organization names, dates, and citations. Throughout this lab you will practice catching these errors. Do not accept any factual claim from an AI tool without checking it.
The Exercise
Five Steps
Work through these in order, since each one builds on what came before.
1 Pick a social issue
Choose a social, environmental, or economic problem that you genuinely care about. Pick something that feels real to you rather than something that sounds impressive on paper. It does not need to connect to any of the community partners you will meet later in the course.
Some examples to get you thinking: food insecurity on college campuses, plastic waste in Central Florida waterways, mental health access for young adults, affordable housing in Orlando, immigrant language barriers, funding cuts for the arts, or isolation among elderly residents. But pick your own issue if something else matters more to you.
If you are stuck on choosing, think about the last time something in the news genuinely frustrated you, or the last time you wished someone would step in and fix a problem in your own community. That is a good place to start.
2 Map the problem landscape
Use your AI tool to explore the problem in depth. At this stage you are not looking for solutions. You are trying to build a real understanding of the issue: who it affects, what drives it, what has already been tried, and where the gaps are in what currently exists.
Suggested Prompt
I want to understand the problem landscape around [your issue]. Help me map:
Who is most affected and how
The root causes (not just symptoms)
What organizations are currently working on this
What approaches have been tried and what the results were
Where the biggest gaps in current solutions are
Be specific. Use real organizations and real data where you can. If you are not sure about a fact, say so.
Read the response carefully. Some of what it tells you will be accurate, and some of it will be wrong, outdated, or entirely fabricated. That is normal, and learning to tell the difference is part of the exercise.
Follow-Up: Verify What You Got
You mentioned [organization name] and [specific claim or statistic]. I want to verify these. For each one, tell me:
Where this information comes from
When it was last updated
How confident you are that it is accurate
Whether I should verify this independently
Flag anything you are not sure about.
Why verification matters: Later this semester you will present findings to real community partners who work on these problems every day. If your presentation includes a fabricated statistic or references an organization that does not exist, you lose credibility with the people you are trying to help. Building the verification habit now saves you from that later.
3 Evaluate AI claims about your issue
Now ask the AI what role it thinks artificial intelligence could play in addressing your issue. Once you have those answers, your job is to evaluate them honestly and figure out which claims are grounded in evidence and which are more aspiration than reality.
Suggested Prompt
What role could AI play in addressing [your issue]? Give me 5 specific applications. For each one:
Describe what the AI would do
Name a real organization doing this (if one exists)
Rate the maturity: proven, experimental, or theoretical
Identify the biggest risk or limitation
Do not oversell. If an application is mostly hype right now, say that.
Once you have the response, run it through the hype filter below. You are not doing more research here. You are practicing critical thinking about the claims you just received.
Hype Filter Prompt
Look at the 5 AI applications you just described. I want to evaluate each one honestly. For each application, answer:
Is this actually working somewhere today, or is it a concept?
What specific evidence exists that it works? (Published results, pilot programs, peer-reviewed studies, not press releases or blog posts)
Who benefits and who could be harmed?
What would a skeptic say about this claim?
On a scale of 1-10, how much of this is real vs. marketing?
Be direct. I would rather know the truth than feel good about AI.
"The question is never 'Can AI do this?' The question is 'Should AI do this, and does it actually work?'"
Keep this frame for the rest of the course.
4 Write discovery questions using the Mom Test
Imagine you are about to interview someone at one of the organizations you identified in Step 2. Your goal in that conversation is to understand their actual day-to-day work, not to pitch them on an idea you have.
The Mom Test, a short book by Rob Fitzpatrick, makes a simple observation: if you ask your mom whether your business idea is good, she will say yes because she loves you and does not want to hurt your feelings. The people you interview will do the same thing. They will be polite, they will nod along, and you will walk away thinking you learned something when you actually did not. The only way to get real information is to ask the right kinds of questions.
Bad questions
"Do you think AI could help with this?"
"Would you use a tool that does X?"
"How much would you pay for this?"
These ask for opinions about hypotheticals. People give encouraging answers that tell you nothing useful.
Good questions
"How do you currently handle X?"
"Tell me about the last time you dealt with X."
"What have you tried so far? What happened?"
These ask about past behavior and facts. People have a harder time fabricating what already happened.
Suggested Prompt
I am preparing to interview someone who works at [organization you identified in Step 2] on [your issue]. Help me draft 8 discovery interview questions using Mom Test principles. The rules:
No hypotheticals. Do not ask "would you" or "do you think."
Ask about past behavior, not future intentions.
Ask about their current workflow, not my idea.
Ask about pain points, workarounds, and things that take too long.
Ask about what they have already tried and why it did or did not work.
For each question, explain:
What I am trying to learn
What a useful answer would sound like
What a polite-but-useless answer would sound like (so I know when to dig deeper)
Read through the questions and ask yourself whether someone could answer any of them with a quick "yes" or "no." If they can, the question is too closed and needs to be rewritten as something more open-ended.
Follow-Up: Stress-Test the Questions
Pretend you are the person I am interviewing. You work at [organization] and you deal with [issue] every day.
Answer my 8 questions the way a real person would, sometimes with useful detail, sometimes with vague politeness. For the vague answers, tell me what follow-up question I should ask to get something real.
Then tell me: which 5 of these 8 questions are the strongest? Cut the weakest 3 and explain why they are weak.
5 Challenge your own assumptions
At this point you have a problem map, an evaluation of AI claims, and a set of discovery questions. Before you wrap up, take a few minutes to let the AI push back on your thinking. This step can feel uncomfortable, and that is the point.
Suggested Prompt
I have been researching [your issue] for the last 30 minutes. Here is what I think I know:
[Write 3-4 sentences summarizing your understanding of the problem]
Now challenge me:
What am I getting wrong?
What am I missing entirely?
What would a skeptic say about my framing of this problem?
What perspective am I not considering? (Think about people who are affected but whose voices are often left out of these conversations.)
If I walked into an organization working on this and shared my analysis, what would make them roll their eyes?
Be honest. I would rather be corrected now than embarrassed later.
This is the most important step in the lab. Being willing to have your assumptions challenged, and actually engaging with the pushback rather than brushing it off, is what makes the difference between using AI in a way that deepens your understanding and using it in a way that just confirms what you already believed.
Rubric
What Good AI Usage Looks Like
Strong (full credit)
Weak (minimal credit)
You verify claims and flag what the AI got wrong
You copy the AI output without checking any facts
Your hype filter catches at least one exaggerated claim
You accept every AI application as proven and ready to deploy
Your discovery questions focus on past behavior and real workflows
Your questions are hypotheticals: "Would you use..." or "Do you think..."
You stress-test your questions and cut the weak ones
You submit all 8 without evaluating which ones actually work
You challenge your own assumptions and engage with the pushback
You skip Step 5 or dismiss the criticism without thinking about it
Your reflection shows what changed in your thinking
Your reflection is a summary of the AI output
Your conversation shows 10+ exchanges with follow-ups
Your conversation shows 5 prompts and 5 outputs, no iteration
Deliverables
What to Submit
Upload a professionally formatted document to Canvas as a PDF or Word file (.docx). Include all four sections below.
Section 1: Problem Landscape (from Steps 1-2)
What social issue you chose and why it matters to you
Who is affected and what the root causes are
At least 3 organizations working on this issue, verified
Where the biggest gaps are in what currently exists
What you verified and what turned out to be wrong or could not be confirmed
Section 2: AI Claims Evaluation (from Step 3)
The 5 AI applications you explored
Your hype filter results, including which claims held up and which did not
The most interesting finding and the most overblown claim
Section 3: Discovery Questions (from Step 4)
Your final 5 discovery questions after cutting the weakest three
For each question, what you are trying to learn and what a useful answer would look like
A brief note on which question you think is strongest and why
Section 4: Reflection (from Step 5)
What assumptions did the AI challenge, and did you agree or push back?
What surprised you most during this lab?
What is one thing you thought you understood about your issue that turned out to be more complicated than you expected?
How did working with AI as a collaborator feel different from doing a regular internet search?
Conversation Log
You must also submit your full AI conversation. This is not optional. Choose one of the three options below.
Option A: Ask the AI to generate a conversation document (recommended)
At the end of your session, use this prompt to generate a clean, structured document of everything you discussed:
Conversation Export Prompt
Create a complete document of our entire conversation that I can submit to my instructor. Structure it as follows:
LAB 1: AI-ASSISTED PROBLEM DISCOVERY
Student: [my name]
Date: [today's date]
Tool used: [Claude / ChatGPT / Perplexity]
For each stage of our conversation, include:
The question or prompt I asked (paraphrased is fine)
The key points from your response
Any follow-up exchanges and how my thinking evolved
End with:
Key insights that changed my thinking during this conversation
Open questions I still need to explore
Format this as a clean, readable document. Do not leave anything out. I need my instructor to see the full analytical process, not just the conclusions.
In Claude, this will generate an artifact you can copy or download directly. In ChatGPT, copy the output into a Google Doc or Word file and export as PDF.
Option B: Share a link
Both Claude and ChatGPT let you generate a shareable link from within the chat. Before you submit, open the link in an incognito or private browser window and make sure the full conversation loads without requiring a login. If I cannot open it, I cannot grade it.
Option C: Copy and paste
Copy and paste the entire conversation into a Word document (.docx). Include both your prompts and the AI responses, in order. Do not edit or rearrange anything.
The instructor needs to see the full back-and-forth, not just the AI's final output. Multi-turn conversations with follow-up questions score higher than single-prompt interactions.
Grading
How This Gets Graded
Problem landscape is specific, verified, and shows critical thinking
25%
AI claims evaluation distinguishes real evidence from marketing
25%
Discovery questions follow Mom Test principles
25%
Reflection shows genuine learning and changed thinking
15%
Conversation log shows iterative prompting with follow-ups
10%
Due: Sunday, August 31 by 11:59 PM. Upload your report and your conversation log to Canvas.
Length: The written report should be 3-5 pages, professionally formatted (PDF or .docx). The conversation log can be any length.
On honesty: Do not fabricate data, do not present AI-generated quotes as things real people said, and do not claim you verified something you did not. These rules apply to every assignment in the course. The goal is to use AI in a way that makes your thinking sharper, not to produce something that just looks polished.
Exemplary Submission
Sample Report
This is what a strong Lab 1 submission looks like. Use it as a reference for structure, depth, and tone. Your issue and analysis will be different, but the level of detail, verification, and critical thinking should be similar.
This shows what a strong conversation log looks like: iterative prompting, follow-ups, verification, and genuine pushback. Notice how each exchange builds on the last.
