The 1%

Multiple Choice

If a fraud detection AI is 99% accurate and checks 56 billion US transactions per year. Roughly how many errors does it make?

• 560,000

• 560,000,000

• 56,000

• 5,600

Open Ended

The fraud AI makes 560 million errors on 56 billion transactions.

Write that as a percentage.

Does the percentage or the raw number feel more significant to you, and why does that matter?"

Multiple Choice

Scenario: A hospital deploys an AI readmission tool that flags high-risk patients for follow-up calls. It's wrong 8% of the time. With this system, the hospital reduces costs by $2M annually. A patient whose risk is missed ends up having a second hospitalization and pays $30,000 out of pocket. 

This is an example of: A) A false positive B) A tradeoff where benefit and risk fall on different parties C) A malfunction that should be reported to the developer D) A problem that better accuracy would eliminate"

• A false positive

• A tradeoff where benefit and risk fall on different parties

• A malfunction that should be reported to the developer

• A problem that better accuracy would eliminate

Multiple Select

Scenario: Maya is writing a research paper on climate change for her AP Environmental Science class. She uses AI to generate a list of five statistics about rising sea levels, then copies them directly into her paper. She submits it without checking any of the sources. Her teacher finds that two of the statistics were fabricated and one had the wrong year.

Select all responsible use steps Maya should have taken.

• Verified each statistic against a reliable source before including it

• Treated the AI output as a first draft, not a final answer

• Avoided using AI entirely for research tasks

• Noted in her paper that she used AI and was uncertain about some figures

Open Ended

Think about the last time you used AI for schoolwork: a search, a summary, a draft, a calculation. Knowing what you know now about error rates and scale, would you do anything differently? What would you verify, and how?

OR

In what kinds of tasks would you trust AI most? Least? What's the difference? Is it about the stakes, the type of task, or something else? Explain your thinking.