Noise

Imagine you're at a shooting arcade. Two different shooters aim at the same target. The first shooter's shots all land in the same spot—but that spot is three inches to the left of the bull's-eye. Consistent, predictable, and wrong in a systematic way. The second shooter's shots scatter everywhere: above, below, left, right, some even hitting the bull's-eye by pure chance. No clear pattern, no predictable direction, just random spray.

The first shooter has a *bias* problem. The second shooter has a *noise* problem.

Most of us are familiar with bias. We know what it means when judgments are systematically skewed in one direction—when a hiring committee consistently favors candidates from certain backgrounds, or when a doctor routinely overdiagnoses a particular condition. Bias makes headlines. It's the subject of training programs, policy reforms, and public outrage.

But noise? Noise is invisible. And that's precisely what makes it so dangerous.

The Two Kinds of Error. Every human judgment contains error. But error comes in two distinct forms. Bias is the systematic deviation—the consistent lean in one direction. Noise is the random scatter—the unwanted variability in judgments that should be identical.

Here's the critical insight: you can have a system with zero bias that's still full of noise. The average judgment might be perfectly on target, but individual judgments swing wildly around that average. Think of that second shooter—shots centered on the bull's-eye on average, but individually all over the place.

The authors make a startling claim: noise is often as damaging as bias, sometimes more so. But because it's harder to see, it goes largely unaddressed.

The Problem of Invisible Noise. Why does noise stay hidden? Three reasons.

First, we rarely see the same case judged by multiple people. When a patient sees one doctor, they get one diagnosis. They have no way of knowing that another doctor would have reached a completely different conclusion. When a defendant appears before one judge, that's the sentence they get. The alternative sentence from a different judge never materializes for comparison.

Second, we have an "illusion of agreement." Most professionals genuinely believe their colleagues see things the same way they do. They don't realize how much their judgments diverge because they rarely test the assumption. Teams nod along in meetings, assuming consensus, while each member has actually reached a different conclusion.

Third, organizations actively avoid situations that would reveal noise. Disagreement feels uncomfortable. Conflict feels unproductive. So companies design processes that minimize exposure to divergent judgments, creating a false sense of harmony that masks the variability underneath.

The Core Argument. The book's central argument is straightforward but devastating: noise is high, it's costly, and it's everywhere. Wherever human judgment is required—in medicine, law, hiring, performance reviews, forecasting, bail decisions, asylum rulings—noise is present in scandalous amounts.

Consider what this means in practice. Two patients with identical symptoms see two different doctors. They get different diagnoses, different treatments, different prognoses. Two people commit the same crime. They get wildly different sentences depending on which judge hears their case. Two equally qualified candidates interview for the same job. One gets hired, one doesn't, based largely on which interviewer they drew.

These aren't edge cases. They're the norm. The variability isn't a bug in the system—it's a feature of human judgment itself.

What Noise Is and Isn't. Let's be precise. Noise is unwanted variability in judgments that should be identical. It's the scatter around the bull's-eye, not the consistent miss.

Bias is when every shot lands in the same wrong place. You can predict the error. You can correct for it. Noise is when shots land everywhere—no pattern, no predictability, just randomness.

Here's what makes noise particularly insidious: you can measure it without knowing anything about what the correct answer should be. You don't need to know the "true" sentence for a crime to see that judges are giving wildly different sentences for the same case. You don't need to know the "correct" diagnosis to see that doctors disagree dramatically on the same patient. Noise reveals itself through disagreement alone.

The authors put it simply: wherever there is judgment, there is noise—and more of it than you think.

A Fresh Perspective on Error. Most organizations focus on eliminating bias. They train employees to recognize their prejudices, design processes to reduce discrimination, and celebrate progress toward fairness. These efforts matter. But they're incomplete.

A system can be perfectly unbiased—treating every group equally on average—while still being deeply noisy. And noise produces its own kind of injustice. It creates a lottery where outcomes depend on which judge, which doctor, which interviewer you happen to get. That lottery operates without anyone's consent, often without anyone's awareness.

The authors aren't arguing that bias doesn't matter. They're arguing that noise matters just as much, and that fixing only one half of the error equation leaves the other half untouched.

What's at Stake. The stakes are enormous. In the insurance industry, a noise audit revealed that underwriters' quotes for identical cases varied by 55 percent—costing the company hundreds of millions of dollars. In medicine, the variability between doctors' diagnoses means patients' lives literally depend on which physician they see. In criminal justice, the difference between probation and a decade in prison can come down to which judge happens to be on the bench that day.

Noise isn't an abstract statistical concept. It's a concrete source of error that produces real harm: financial losses, misdiagnoses, unjust sentences, bad hires, failed projects, missed opportunities.

The Path Forward. The first step is recognition. Noise exists. It's measurable. It's damaging. And it's been hiding in plain sight while we focused almost exclusively on bias.

The authors don't promise to eliminate noise entirely. Human judgment will always involve some variability. But they argue that noise can be dramatically reduced through systematic approaches—and that the effort is worth making.

Here's the question that sets up everything that follows: If noise is as pervasive and costly as the evidence suggests, why have we been so blind to it? And what would it take to finally see—and fix—the problem?