Exercises: The Failure Modes of the Future

Part A: Comprehension and Application

A.1. Distinguish between AI failure modes that are accelerations of existing failure modes and those that are genuinely new. Give two examples of each.

A.2. Define "algorithmic consensus" and explain why it is structurally different from human consensus. Why is the distinction dangerous?

A.3. Explain the "epistemic pollution" feedback loop. Why does it have no historical analogue? What partial analogues exist, and why are they insufficient?

A.4. Define "confidence laundering" and explain how AI systems perform it. How does this differ from the precision-without-accuracy problem (Chapter 12)?

Part B: Analysis

B.1. Apply the Red Flag Scorecard (Chapter 31) to an AI-generated summary or recommendation in your field. Which questions work well for AI-generated content? Which need modification? Design two supplementary questions specifically for evaluating AI outputs.

B.2. Apply the Epistemic Health Checklist (Chapter 32) to an AI-dependent decision-making process in your field or organization. Which dimensions are most affected by AI dependency? Where are the greatest risks?

B.3. The chapter identifies four new tools needed for AI-era failure modes: model diversity, training data audits, output labeling, and feedback loop monitoring. For each, design a specific implementation for your field. What institutional changes would be required?

Part C: Synthesis and Evaluation

C.1. The overcorrection framework (Chapter 21) predicts a pendulum dynamic: AI winter → overcorrection (uncritical enthusiasm) → crisis → meta-correction. Where are we in this cycle? What would the "crisis" look like? What would "calibrated integration" look like?

C.2. Is it possible to build AI systems that are less susceptible to failure modes than human institutions? Could an AI system implement the seven design principles (Chapter 37) more reliably than a human institution? Or does AI introduce more failure modes than it addresses?

Part D: Mixed Practice (Interleaved)

D.1. A hospital is considering using an AI diagnostic system to assist physicians. Using the complete toolkit from this book — Red Flag Scorecard for the AI's claims, Epistemic Health Checklist for the hospital's decision-making process, Seven Principles for the institutional design, and the AI-specific failure modes from this chapter — produce a comprehensive risk assessment and recommendation.