Section 1

Responsible AI Foundations

Section 1 decides whom an AI system may affect, where its systematic errors come from and which kind of equality its decisions promise. It works through the nH Predict lawsuit and Volume 11's claims assistant, Bloomberg's test of GPT-3.5 and GPT-4 as résumé rankers and ProPublica's COMPAS data, recomputing the audit and fairness numbers from published counts. Each chapter ends in a priced decision: meaningful review of adverse drafts at about 1% of adjuster time, a bias audit of 5,000 rankings per job family, and one risk threshold for everyone, set from a harm weight with a named owner.