Chapter 7 Further Reading: AI Decision-Making

Recommended Books

Cathy O'Neil, Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy (Crown, 2016) A landmark book on how algorithmic decision systems — in lending, hiring, policing, and education — encode bias and perpetuate inequality. O'Neil, a former Wall Street quant, writes with precision and passion about the human costs of opaque scoring systems. Essential reading for anyone interested in the real-world consequences of AI classification and prediction. Chapters on credit scoring, predictive policing, and hiring algorithms connect directly to this chapter's themes.

Eli Pariser, The Filter Bubble: How the New Personalized Web Is Changing What We Read and How We Think (Penguin Press, 2011) Pariser coined the term "filter bubble" to describe how personalization algorithms create information silos. Though published before the current AI boom, the core analysis remains relevant: recommendation systems optimized for engagement inevitably narrow the information landscape. A useful framework for thinking about the trade-offs discussed in Section 7.2.

Virginia Eubanks, Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor (St. Martin's Press, 2018) Eubanks examines how automated decision systems in public services — child protective services, homeless services, Medicaid — disproportionately impact low-income Americans. Her case studies are deeply reported and focus on the human experience of being classified, scored, and sorted by machines. Pairs well with Case Study 7.2 on credit scoring.

Ruha Benjamin, Race After Technology: Abolitionist Tools for the New Jim Code (Polity, 2019) Benjamin explores how AI systems can reinforce racial inequality under the guise of objectivity. Her concept of the "New Jim Code" — technologies that encode discrimination while appearing neutral — is directly relevant to the discussion of proxy variables and feedback loops in Sections 7.4 and 7.6.

Recommended Articles and Reports

Julia Angwin, Jeff Larson, Surya Mattu, and Lauren Kirchner, "Machine Bias," ProPublica (May 23, 2016) The investigation that brought the COMPAS recidivism prediction tool into public debate. ProPublica analyzed COMPAS scores for over 7,000 defendants in Broward County, Florida, and found that the system was significantly more likely to falsely classify Black defendants as high-risk and white defendants as low-risk. Connects directly to the discussion of prediction and proxy variables in Section 7.4. Available at: propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing

Kristian Lum and William Isaac, "To Predict and Serve?" Significance, Vol. 13, No. 5 (October 2016) A rigorous analysis of the predictive policing feedback loop discussed in Section 7.6. The researchers used drug use data from the National Survey on Drug Use and Health to show that PredPol, if trained on historical arrest data, would disproportionately target Black and Hispanic neighborhoods — even though drug use rates are roughly equal across racial groups. A clear demonstration of how biased data produces biased predictions.

Jeff Horwitz and Deepa Seetharaman, "Facebook Knew Its Algorithms Could Divide. Here's Why It Didn't Fix Them," Wall Street Journal (September 15, 2021) Part of the Wall Street Journal's "Facebook Files" series based on the Frances Haugen disclosures. Documents the internal tension between engagement optimization and content quality that is central to Case Study 7.1. Read alongside the original Facebook Oversight Board reports for a fuller picture.

Consumer Financial Protection Bureau, "Supervisory Highlights: Fair Lending" (2022) The CFPB's analysis of algorithmic lending models and disparate impact. Technical but accessible, and directly relevant to Case Study 7.2 on credit scoring. Available at: consumerfinance.gov

Recommended Videos and Podcasts

"The Social Dilemma" (Netflix, 2020) A documentary featuring former Silicon Valley employees discussing the design of social media recommendation systems. Watch critically — the film has been criticized for oversimplifying complex systems and for centering the perspectives of the people who built these systems rather than those most affected by them. Best used as a conversation starter rather than a definitive account.

Safiya Umoja Noble, "Algorithms of Oppression" (TEDxUSC, 2018) A talk based on Noble's research into how search engine recommendation and ranking systems reproduce racial and gender stereotypes. Approximately 15 minutes. A concise introduction to the intersection of recommendation systems and social inequality.

For Deep Dive Readers

Solon Barocas and Andrew Selbst, "Big Data's Disparate Impact," California Law Review, Vol. 104 (2016) A legal and technical analysis of how data mining and machine learning can produce discriminatory outcomes even without discriminatory intent. More technical than the other recommendations but foundational for understanding the legal frameworks around AI classification.

Finale Doshi-Velez and Been Kim, "Towards a Rigorous Science of Interpretable Machine Learning," arXiv:1702.08608 (2017) A foundational paper defining interpretability in machine learning and proposing frameworks for evaluating it. Written for a technical audience but approachable for motivated general readers. Useful for understanding the accuracy-interpretability trade-off discussed in Section 7.5 at a deeper level.

Arvind Narayanan, "How to Recognize AI Snake Oil" (Princeton lecture, 2019) A presentation by Princeton computer scientist Arvind Narayanan distinguishing between AI applications where prediction is well-supported by evidence (perception tasks, certain natural science predictions) and those where it is not (predicting individual human behavior). Directly relevant to Section 7.4's discussion of prediction vs. explanation. Available at: cs.princeton.edu/~arvindn/talks/MIT-STS-AI-snakeoil.pdf