Chapter 7 Exercises: AI Decision-Making — Recommendations, Classifications, and Predictions

Part A: Foundational (Understanding)

Exercise 7.1 — Mode Identification For each of the following AI applications, identify whether the primary decision mode is recommendation, classification, or prediction. Some may involve more than one mode — if so, identify all that apply and explain how they interact.

a) A music streaming service creates a "Discover Weekly" playlist for each user b) An airport security scanner flags luggage for additional inspection c) A weather app forecasts tomorrow's high temperature d) A social media platform decides whether a post violates community guidelines e) An insurance company estimates the probability that a driver will file a claim in the next year f) A job board ranks which listings to show a user first g) A hospital system estimates which emergency room patients are most likely to deteriorate in the next two hours

Exercise 7.2 — False Positive vs. False Negative Consequences For each scenario below, explain the real-world consequences of (i) a false positive and (ii) a false negative. Then state which type of error you consider more harmful in that context, and explain your reasoning.

a) A fraud detection system for credit card transactions b) A content moderation system screening for child exploitation material c) A medical screening test for a rare but treatable cancer d) An automated hiring system that screens resumes for a job interview e) A spam email filter

Exercise 7.3 — Vocabulary Practice Define each of the following terms in your own words and provide an original example (not from the textbook) for each:

a) Collaborative filtering b) Content-based filtering c) Proxy variable d) Feedback loop e) Interpretability f) False positive

Part B: Intermediate (Apply / Analyze)

Exercise 7.4 — Recommendation System Audit Choose a recommendation system you use regularly (e.g., Netflix, YouTube, Spotify, TikTok, Amazon, a news aggregator). Over the course of three days, keep a brief log:

  • What does the system recommend to you?
  • Can you identify any pattern in the recommendations?
  • Deliberately interact with something outside your usual pattern (watch a different genre, listen to different music, search for something you wouldn't normally search for). How does the system respond? How quickly does it adjust?

Write a one-page reflection covering: 1. What you observed about the recommendation patterns 2. Whether you could detect evidence of collaborative filtering, content-based filtering, or both 3. Whether you noticed any filter bubble effects 4. What the system seems to be optimizing for (engagement, satisfaction, diversity, or something else)

Exercise 7.5 — The Accuracy-Interpretability Worksheet Consider the following scenario: A university wants to use an AI system to identify first-year students at risk of dropping out so it can offer early intervention (tutoring, advising, financial aid referrals).

Two systems are available:

  • System A: A logistic regression model using five variables (high school GPA, family income, first-generation status, distance from home, declared major). Accuracy: 72%. Fully interpretable — each student's risk score can be traced to specific factors.

  • System B: A deep neural network using 150 variables (including social media activity, campus WiFi login patterns, dining hall usage, library check-outs, and more). Accuracy: 88%. Not interpretable — produces a risk score but cannot explain why.

Answer the following: a) What are the advantages of System A over System B? b) What are the advantages of System B over System A? c) What ethical concerns does System B raise that System A does not? d) What data privacy concerns does System B raise? e) If you were the university's decision-maker, which system would you choose? Justify your answer. f) Is there a third option that might address the limitations of both? Describe it.

Exercise 7.6 — Tracing a Feedback Loop Choose one of the following systems and diagram its feedback loop. Include at least five steps in the cycle, clearly label the data flows, and identify the point(s) where intervention could break the loop.

a) A predictive policing system (like CityScope Predict) b) A hiring system that learns from past hiring decisions c) A social media recommendation system that optimizes for engagement d) A credit scoring system

For your diagram, identify: - Where the loop reinforces existing patterns - Whether the loop is self-correcting or self-amplifying - Two specific interventions that could break or modify the loop

Part C: Advanced (Analyze / Evaluate)

Exercise 7.7 — Perspective-Taking: The Stakeholder Analysis CityScope Predict, the predictive policing system, affects many different stakeholders. For each of the following stakeholders, write 3-4 sentences describing how they might view the system — their hopes, concerns, and what they'd want to know:

a) A police officer assigned to patrols based on CityScope's predictions b) A resident of a neighborhood consistently flagged as "high risk" c) The city council member who authorized purchasing the system d) A data scientist who built the system e) A defense attorney whose client was arrested during an algorithmically directed patrol f) A crime victim in a "low risk" neighborhood that receives fewer patrols

After completing all six perspectives, write a paragraph reflecting on what this exercise reveals about the difficulty of evaluating AI decision systems.

Exercise 7.8 — Policy Proposal You are an advisor to a government agency that is considering adopting an AI system for one of the following purposes:

a) Allocating public housing (classifying applicants by priority level) b) Setting bail amounts for arrested individuals (predicting flight risk) c) Distributing social service resources across neighborhoods (predicting need)

Write a 500-word policy memo that: 1. Identifies the decision mode(s) involved 2. Describes the accuracy-interpretability trade-off relevant to this context 3. Identifies at least two potential proxy variables and explains why they're problematic 4. Describes a potential feedback loop and how to prevent it 5. Recommends whether the agency should adopt the system, with specific conditions or safeguards

Exercise 7.9 — Comparative Analysis: Prediction vs. Explanation Read the following scenario:

A school district uses an AI system that predicts with 85% accuracy which students will fail to graduate on time. The system identifies zip code, household income, number of school transitions, and third-grade reading level as the strongest predictive variables.

a) What does this system predict? b) What doesn't it explain? c) A school board member proposes using the system to identify "at-risk" students for mandatory after-school tutoring. Evaluate this proposal: what are its strengths and weaknesses? d) Propose an alternative intervention strategy that addresses the explanatory gaps the prediction system leaves open. e) Could the prediction system itself cause harm, even if the predictions are accurate? Explain.

Part D: Research Extension (for Deep Dive readers)

Exercise 7.10 — The Recommender System Debate Research the "rabbit hole" effect in YouTube's recommendation system. Find two sources: one that argues recommendation systems are primarily harmful (amplifying extremism, polarization) and one that argues the effect is overstated.

Write a 750-word essay that: 1. Summarizes both positions fairly 2. Evaluates the evidence each side presents 3. Identifies the methodological challenges in studying recommendation effects 4. States your own assessed position with reasoning

Exercise 7.11 — Case Research: COMPAS Research the COMPAS recidivism prediction tool and the 2016 ProPublica investigation. (You may use the Further Reading suggestions for this chapter as starting points.)

Answer: a) What does COMPAS predict, and how is the prediction used? b) What did ProPublica find about the system's error rates across racial groups? c) How did Northpointe (the company that made COMPAS) respond? d) Can a system be "fair" to all groups simultaneously? Why or why not? (This question connects to Chapter 9 — note your initial thinking and revisit after reading that chapter.)