Systems Thinking for Human and Machine Judgment

People and AI systems now share many of the decisions that matter. Each can think fast or slow, and each fails in ways you can predict. This book shows you how to get the combination right.

Systems Thinking for Human and Machine Judgment is an operating manual, not a theory book. Across 45 chapters in nine parts, it turns research on human judgment, machine learning and systems thinking into 81 named protocols: practical routines with clear steps, a frank grade for the evidence behind them, the ways they tend to fail, and a quick card you can keep at your desk.

What you will learn

• When intuition and fast models have earned trust, and how to set the confidence thresholds at which a model acts, asks or hands over
• How to slow down well, with decomposition, base rates, red teams and a clear sense of when to switch speeds
• How to see the system around a decision: stocks, flows, delays, feedback loops, leverage points and Goodhart’s law
• How to work with reasoning models and agents: briefing them, reading their reasoning traces honestly, setting autonomy levels and designing human checkpoints
• How to measure judgment with evals and calibration, for models, for agents over time and for yourself
• How to govern decisions through clear decision rights, oversight that works, practical risk frameworks and auditable records

Who it is for

Leaders, operators, analysts, engineers and policy makers who make or oversee decisions with AI in the room. You do not need to write code to use it.

What makes it different

The book treats the person, the model and the checks around them as one system. It is open about the strength of its evidence, grading claims from replicated experiments to practice-based judgment, and it learns from documented cases, including Robodebt, the Dutch childcare benefits scandal, Mata v. Avianca, Air France 447 and the CrowdStrike outage. Notes at the end of every chapter give the sources.