Noise

The verdict

Noise is the intellectual sequel to Thinking, Fast and Slow, and it addresses a blind spot in the behavioral science conversation: while everyone talks about bias (systematic errors in judgment), almost no one talks about noise (random variability in judgment). Kahneman, Sibony, and Sunstein argue that noise is at least as costly as bias and far easier to reduce — if you know it exists.

The book’s core demonstration is devastating. In study after study, professionals who should reach similar conclusions about identical cases produce wildly divergent judgments. Judges give sentences that vary by years for the same crime. Doctors recommend different treatments for identical symptoms. Insurance underwriters assign premiums that differ by hundreds of percent for the same risk. This variability is not explained by different information or different values. It is noise.

The book is too long, repetitive in its middle sections, and could have delivered its argument in half the pages. But the argument itself is important enough to justify the slog.

The noise audit

The book opens with a concept it calls the noise audit — a simple test that most organizations have never conducted. Take a set of cases that have already been decided. Present the same cases (disguised) to the same professionals. Measure the variation. In every organization that has conducted this audit, the results have been shocking: far more variability than anyone expected.

The reason organizations do not conduct noise audits is not that they are difficult but that they are threatening. If two doctors in the same practice would treat the same patient differently, what does that say about the quality of medical judgment? If two loan officers would make opposite decisions about the same applicant, what does that say about the fairness of the lending process? Noise audits reveal that the expert judgment people rely on is less reliable than they assume, and this revelation is uncomfortable enough that most organizations avoid it.

System noise, level noise, and pattern noise

Kahneman and colleagues decompose noise into useful subcategories. System noise is the total variability in judgments across an organization. Level noise is the variation in average severity — some judges are consistently harsher than others, some doctors are consistently more conservative. Pattern noise is the variation in how individuals respond to specific cases — one judge is harsh on white-collar crime but lenient on drug offenses, while another shows the reverse pattern.

Occasion noise is perhaps the most disturbing subcategory: the same person, evaluating the same case on different days, reaches different conclusions. Your judgment varies with your mood, your fatigue, the weather, and whether you ate recently. This is not a minor effect. Studies of judicial sentencing show measurable differences in rulings based on the time of day and whether the local sports team won.

Algorithms versus judgment

The book makes a careful case for simple algorithms and decision rules as noise-reduction tools. The argument is not that algorithms are brilliant but that they are consistent. A simple model that weighs a few factors will produce the same output every time it encounters the same inputs. Human experts will not.

Kahneman is not arguing for replacing human judgment with algorithms in all cases. He is arguing that in cases where consistency matters — sentencing, hiring, medical diagnosis, risk assessment — structured decision processes dramatically outperform unstructured expert judgment, primarily because they eliminate noise.

The resistance to algorithms is psychological, not evidential. People trust expert judgment because it feels more thoughtful, more nuanced, more human. But the evidence consistently shows that this feeling is wrong — unstructured expert judgment is noisier, not more accurate, than simple rules.

Decision hygiene

The book’s prescriptive section introduces “decision hygiene” — a set of practices designed to reduce noise in organizational decisions. These include sequencing information exposure (evaluating dimensions independently before forming an overall judgment), using structured decision protocols, aggregating independent judgments from multiple evaluators, and conducting noise audits.

The analogy to physical hygiene is deliberate: you practice hand-washing not because you can see the germs on a specific occasion but because the general practice reduces a general risk. Similarly, you practice decision hygiene not because you can identify the specific noise in a specific decision but because the general practice reduces the random variability that degrades all decisions.

Read this if…

You make consequential judgments as part of your work — hiring, evaluating performance, assessing risk, diagnosing problems, or allocating resources. The book is also valuable for anyone who has read Thinking, Fast and Slow and wants to understand the complementary problem of noise alongside bias. If you manage a team that makes repeated judgments about similar cases, the noise audit concept alone is worth the read.

Skip this if…

You have limited patience for academic writing at book length. The core argument of Noise can be stated in a few pages, and the book takes several hundred to develop it. The middle sections are repetitive, revisiting the same point from different angles without adding proportional insight. A detailed summary captures the essential framework.

Start here

Read the introduction and Chapters 1-3 for the core concepts. Then jump to Chapter 7 (occasion noise), Chapter 21 (the case for algorithms), and Chapter 24 (decision hygiene). These sections contain the ideas with the highest practical return.