AI Literacy - Cycle 3 - CIRAS AI Summit

Home – Cycle 1 – Cycle 2 – Cycle 3 – Cycle 4 – Cycle 5 – Cycle 6

Cycle 3 – Direct It – Experiment: Make More of Your Intent Explicit

How to Use This Page

Run the guided experiment with the instructor.
Notice what the AI has to decide when your intent is not fully defined.
Use RTIO to decide what should be made more explicit.
Build and run an aligned prompt using the same information.
Then choose one independent experiment and apply RTIO in a different way.

Guided Experiment

Scenario: Understanding a Rise in Customer Complaints

Goal: See how a reasonable request can still leave important decisions to the AI, and how RTIO helps you make more of your intent explicit.

Step 1 – Start With a Basic Request

Situation: A manufacturer has received a noticeable increase in customer complaints during the last six weeks. Leadership wants to understand what is happening before deciding what to do next.

What you know:

The company sells configurable industrial workstations directly to business customers.
Orders are built to customer specifications and usually ship within 10 business days.
Six weeks ago, the company changed its online ordering form.
Six weeks ago, the company also began using a new packaging supplier.
Sales volume has increased about 12% during the same period.

Run this prompt:

Review these customer comments and tell me what's going on.

Customer Comments:

1. "The workstation itself is great, but two of the accessory brackets we ordered were not in the shipment."

2. "We were told 10 business days. It arrived on day 14 and we did not hear anything until we called."

3. "The new ordering page was confusing. I selected the wrong power option and did not realize it until the confirmation email."

4. "Packaging looked rough when it arrived. One corner of the work surface was chipped."

5. "Customer service fixed the issue quickly, but we should not have had to call to find out the order was delayed."

6. "The shipment arrived on time, but the hardware bag contained the wrong bolts for one of the accessories."

7. "We have ordered from you three times before. This was the first time the configuration process felt difficult."

8. "One carton was crushed. Nothing inside was unusable, but several pieces were scratched."

9. "Our order confirmation showed the right model, but the mounting kit that arrived did not match it."

10. "Product quality is still very good. My biggest complaint is that nobody told us the order had slipped by three days."

Discuss: The answer may sound reasonable. What did the AI have to decide for you?

Step 2 – Build an Aligned Request

Use RTIO to decide what you want the AI to understand. Choose the statements that fit your intent, then copy and paste them into ChatGPT with the customer information above. You do not need to use every category or every statement.

Role – What perspective would be useful?

You are an operations manager focused on process reliability, delivery performance, and customer impact.

You are a customer experience manager focused on customer frustration, communication, and loss of confidence.

You are a quality manager focused on defects, recurring failure patterns, and process breakdowns.

You are a business leader evaluating which customer issues deserve management attention first.

Task – What should the AI actually accomplish?

Identify the recurring problems appearing across the customer comments.

Group related complaints into meaningful themes rather than treating every comment as a separate issue.

Prioritize the issues based on frequency, severity, and likely business impact.

Identify likely contributing causes when the available information supports a reasonable inference.

Recommend practical actions the company should investigate or take next.

Input – What information should it consider?

Base the analysis on the customer comments and company background provided.

Pay attention to repeated issues rather than giving equal weight to one-off comments.

Consider the recent change to the online ordering form and the new packaging supplier as possible contributing factors, but do not assume they are causes without evidence.

Consider the 12 percent increase in sales volume as a possible source of operational strain.

Do not invent facts that are not present in the comments or background information.

Output – What would make the result useful?

Identify the three most important issues.

For each issue, briefly explain the customer evidence that supports the conclusion.

Clearly separate what the comments directly support from what you are inferring.

For each priority issue, recommend one or two practical next steps.

Keep the response concise enough for a leadership discussion and avoid unnecessary background explanation.

Run it: Copy the statements you want, combine them into a prompt, and run that prompt with the same background information and customer comments from Step 1.

Compare / Reflect

What changed when you made more of your intent explicit?
Did the AI focus on different issues?
Did it make fewer decisions on your behalf?
Did the response become more useful for the purpose you actually had in mind?
Did someone else choose different RTIO options and still get a useful result?

Independent Experiments

Choose one scenario below. Each experiment uses RTIO in a different way.

Scenario 1 – Same Information, Different Intent

Task: Use the same meeting transcript for a different purpose.

Choose one purpose:

Prepare leadership to make a decision.
Create an action plan for the project team.
Identify process problems worth investigating further.

Read the meeting transcript.
Choose one purpose.
Use RTIO to construct a request that fits that purpose.
Run it.
Compare with someone who chose a different purpose if possible.

Question: The information stayed the same. What changed because your intent changed?

Example Input

Meeting Transcript: Process Improvement Discussion

Participants: Sarah (Operations Manager), Mike (Process Engineer), Jenna (Customer Support Lead), Alex (Data Analyst)

Sarah: We’ve been seeing delays in order processing, and it’s starting to affect customer satisfaction. I want to pinpoint where things are breaking down.

Jenna: Customers are frustrated because they don’t know where their order is once it’s submitted. Sometimes the status just sits in “processing.”

Alex: About 40% of orders remain in “processing” longer than 24 hours, even when work has already started.

Mike: Statuses are updated manually, and people do not always remember to change them.

Alex: We could automate some status changes when workflow steps are completed.

Sarah: What about actual bottlenecks?

Alex: The approval step is the biggest delay. It averages 18 hours, while every other stage averages under six.

Mike: Approvals depend on one person per shift. If that person is busy, work stacks up.

Mike: We could also use threshold-based approvals so smaller routine orders do not require manual sign-off.

Jenna: Proactive customer messages during delays would reduce complaints.

Sarah: Let’s focus on three things: automate status updates, analyze approval thresholds, and improve customer communication.

Scenario 2 – Diagnose a Reasonable Request

Task: Find what is still undefined in a prompt that already sounds pretty good.

Situation: Your company is considering buying an automated pallet wrapper.

You know:

Purchase price: $48,000.
Estimated installation and training: $6,500.
Current wrapping process uses two employees for about 3 total labor-hours per day.
Loaded labor cost averages $31 per hour.
The vendor says the system can reduce wrapping labor by 70%.
Current annual stretch-film cost is about $29,000.
The vendor estimates a 15% to 25% reduction in film use.
The machine requires preventive maintenance every six months.
No maintenance cost estimate has been provided.
The plant expects volume to increase next year, but no forecast has been approved.

Starting Prompt:

Review this information and recommend whether we should purchase the pallet wrapper. Explain the major advantages and disadvantages.

Before you run it, use RTIO to identify what is still undefined.
Decide which missing pieces actually matter.
Improve only those parts.
Run the original prompt and your revised prompt.
Compare the recommendations.

Watch for: Does the original prompt sound competent while still making assumptions you never intended?

Scenario 3 – Fix the Result, Not Just the Prompt

Task: Use a disappointing result to diagnose what the request left undefined.

Starting Prompt:

Turn these field notes into a clear daily summary for the project file.

Field Notes:

Tuesday. Six people on site this morning. Mike was about twenty minutes late. Ground was frozen until around 10:00.

Worked mostly on the east side. Started trenching near station 12 and worked toward station 18. Hit unexpected rock around noon and it slowed us down.

Excavator ran normally. Skid steer is still leaking hydraulic fluid. The leak is worse than yesterday and needs attention.

Concrete truck arrived around 1:30 for the footing pour. Inspector did not arrive until after 2:15, so only about half the planned pour was completed.

Safety meeting covered trench safety and cold weather. No incidents. Trench walls are getting steep in a couple spots.

Rebar delivery arrived and looked good. Two pipe sections that were supposed to arrive yesterday are still missing.

Plan for tomorrow is to continue trenching if weather allows and finish the footing pour if inspection happens on time.

Office needs to follow up with the pipe supplier first thing tomorrow.

Your supervisor says:

This is a good summary, but I still can't tell what needs my attention tomorrow.

Look back at your original request.
Use RTIO to diagnose why the result is not useful enough.
Revise only what needs clarification.
Run the revised request.
Compare the result to the first summary.

Watch for: Which RTIO element actually needed to change?

Scenario 4 – Use Your Own Situation

Task: Apply RTIO naturally to something you might actually ask an LLM to do at work.

Choose a real work task.
Write the request naturally, the way you would normally type or say it.
Before sending it, ask: What am I leaving for the model to decide?
Use RTIO as a checklist and clarify only the parts that matter.
Run the request and evaluate the result.
If the result misses, use RTIO again to diagnose what remained undefined.

Goal: You are not trying to fill out RTIO. You are using it to communicate more deliberately.

Developed in collaboration by:
CIRAS (Center for Industrial Research And Service, ISU) • IIAI (Iowa Initiative for Artificial Intelligence, UofI) • TrAC (Translational Artificial Intelligence Center, ISU)