# Cognitive Biases (General)

> **L0:** Cognitive Error Pattern
> **L1:** When your confident judgment turns out to have been predictably wrong — in the same direction as everyone else's.
> **Also known as:** systematic errors; cognitive illusions; thinking traps; judgment fallacies

Source: /concepts/cognitive-biases
You're a radiologist at a major hospital, and you've just examined a chest CT scan. You found a small nodule on the lung and flagged it for follow-up. What you did not notice is the image of a gorilla, forty-eight times the size of the average nodule, superimposed directly onto the scan. This is not a thought experiment. In 2013, researchers Trafton Drew, Melissa Vo, and Jeremy Wolfe ran exactly this study at Brigham and Women's Hospital. Eighty-three percent of radiologists missed the gorilla. These were experts with years of training, examining images in their core area of competence. They missed it because they were looking for nodules, and their cognitive systems were configured to find what they expected to find and filter out everything else. This is what cognitive biases do. They don't make you stupid. They make you systematically wrong in ways that feel indistinguishable from being right.

 What Cognitive Biases Are

A cognitive bias is a systematic pattern of deviation from rationality in judgment. The key word is systematic. Everyone makes mistakes, but cognitive biases are not random errors. They are predictable, directional, and consistent across populations. If you flip a coin and guess wrong, that's random error. If you consistently overestimate the probability of events you can easily recall, that's a cognitive bias, specifically the availability bias, and it will push your judgment in the same direction every time. This also means cognitive biases are NOT a sign of stupidity or low intelligence. The radiologists in the gorilla study were highly trained experts. Kahneman himself has described how he continues to notice cognitive biases in his own thinking decades after characterizing them. Biases are not failures of IQ — they are features of the cognitive architecture that every human brain runs on, which is precisely what makes them so difficult to correct.

This is NOT the same as heuristics, though the two are frequently confused. Heuristics are mental shortcuts, cognitive rules of thumb that allow the brain to make fast decisions with limited information. Many heuristics are adaptive and efficient. Cognitive biases are what happen when those shortcuts produce systematic errors. The availability heuristic, for example, is a useful shortcut: estimating probability based on how easily examples come to mind works well in environments where memorable events are also common events. The availability bias is what occurs when that shortcut misfires, when vivid but rare events (plane crashes, shark attacks) dominate your probability estimates because they are easy to recall, not because they are likely.

The modern study of cognitive biases began in the early 1970s with psychologists Daniel Kahneman and Amos Tversky, whose collaboration produced one of the most influential research programs in the history of social science. Their work demonstrated that human irrationality is not random but patterned, and that these patterns could be catalogued, predicted, and studied with the same rigor applied to any natural phenomenon. Heuristics are the underlying mental shortcuts from which many biases emerge, and understanding their adaptive function helps explain why biases are so persistent: they are side effects of cognitive strategies that are usually beneficial, which means the same mental process that generates a bias under one set of conditions generates a correct answer under another.

 The Machinery Underneath: Why the Brain Generates Biases

Understanding why cognitive biases exist requires understanding the architecture of the brain that produces them. Kahneman's framework, popularized in Thinking, Fast and Slow (2011), divides cognition into two systems. System 1 operates automatically, rapidly, and with little sense of voluntary control. It recognizes faces, completes the phrase "bread and ," and flinches at a loud noise. System 2 allocates attention to effortful mental activities, including complex computation, logical reasoning, and self-monitoring. The critical insight is that System 1 runs constantly and generates impressions, feelings, and inclinations that System 2 may endorse, correct, or override, but System 2 is lazy. It requires metabolic resources, tires easily, and defaults to accepting System 1's outputs unless something triggers active scrutiny. Cognitive biases are, in large part, the artifacts of System 1 doing its job efficiently in situations where efficiency produces systematic error.

This is not a design flaw. The evolutionary psychologist Gerd Gigerenzer has argued persuasively that many cognitive biases represent "ecological rationality": they produce correct answers more often than not in the environments where they evolved. The human brain developed in small-group, information-poor settings where quick decisions based on limited cues were more adaptive than slow, comprehensive analysis. Overestimating the probability of threats (the negativity bias) kept our ancestors alive in environments where the cost of missing a predator was death and the cost of a false alarm was a wasted sprint. The problem is that we now live in environments radically different from those that shaped our cognitive architecture. Statistical reasoning, probabilistic inference, and the evaluation of large datasets were not selection pressures during the Pleistocene. The same machinery that kept your ancestors alive on the savannah now causes you to misjudge risk, overreact to headlines, and make investment decisions based on feelings rather than fundamentals.

 The Biases That Matter Most

Hundreds of cognitive biases have been catalogued, but a handful account for a disproportionate share of real-world damage.

Confirmation bias is the tendency to seek, interpret, and remember information that confirms pre-existing beliefs while discounting information that contradicts them. Peter Wason's 1960 "2-4-6 task" first demonstrated this experimentally: subjects who were asked to discover a rule governing number sequences consistently tested only sequences that confirmed their initial hypothesis, never sequences that could disprove it. Confirmation bias does not merely cause you to notice favorable evidence. It actively distorts how you process all evidence, making confirming information feel more credible and disconfirming information feel weaker, regardless of its actual quality. The political scientist Charles Taber showed in 2006 that giving partisans balanced information on gun control and affirmative action actually increased polarization: each side processed the same data in ways that reinforced their prior position.

The sunk cost fallacy causes people to continue investing in a failing course of action because of resources already spent rather than future prospects. The Concorde supersonic jet became the textbook case. By the early 1970s, both the British and French governments knew the project would never be commercially viable, yet they continued funding it because they had already invested billions. Hal Arkes and Catherine Blumer demonstrated in 1985 that sunk cost reasoning is not limited to large institutional decisions. In their experiments, people who had paid more for a theater subscription attended more performances, even bad ones, than those who had paid less, because the higher investment created stronger psychological pressure to "get their money's worth." The rational calculation is identical: the money is gone regardless. But the bias makes past expenditure feel like a reason to continue.

Anchoring bias is the tendency for an initial piece of information to disproportionately influence subsequent judgments. In one of Kahneman and Tversky's most famous experiments, subjects watched a rigged wheel of fortune land on either 10 or 65, then were asked to estimate the percentage of African countries in the United Nations. Those who saw 65 guessed significantly higher than those who saw 10, even though the wheel spin was obviously random and irrelevant. Anchoring affects real-world decisions constantly: salary negotiations where the first number spoken disproportionately determines the outcome, real estate appraisals influenced by listing price, and criminal sentencing affected by the prosecution's requested term. The mechanism is a failure of stopping criteria: the brain uses the anchor as a starting point and adjusts toward a plausible answer, stopping as soon as the answer feels reasonable — but plausibility is itself assessed relative to the anchor. Move the anchor and you shift the entire range of what feels reasonable, not just the starting point. The same number that reads as extravagant against one anchor reads as modest against another, which is why experienced negotiators fight so hard to set the first number.

 Where You Have Already Seen These Work

At systemic scale, the 2008 financial crisis was a cascade of cognitive biases operating simultaneously. Confirmation bias led rating agencies to interpret the performance of mortgage-backed securities through models that assumed housing prices would continue rising, because they always had. Anchoring bias caused analysts to benchmark risk against recent history (a period of low defaults) rather than against the broader historical record. Herding behavior, a social form of bias, led institutions to replicate each other's strategies rather than independently assess risk, because deviating from the group felt more dangerous than following it off a cliff. None of the individuals involved were irrational in isolation. The system was irrational because individual biases aligned and reinforced each other at scale.

At personal scale, consider how you chose your current home. Research by behavioral economist Dan Ariely showed that real estate decisions are heavily influenced by "decoy" properties, listings that are clearly inferior to one alternative but similar in type. The decoy makes the superior similar option look better by comparison, not through any change in its actual value, but through the contrast effect, a form of anchoring. If you toured three apartments and one was obviously worse than another of the same style, the "better" one likely received a psychological boost it didn't earn. You probably experienced this as a genuine preference. That's what makes biases difficult: they don't feel like errors. They feel like judgments.

 Where the Concept of Cognitive Biases Breaks Down

The framework of cognitive biases is powerful, but it has real limitations that deserve scrutiny.

The replication crisis has narrowed the list. Many celebrated biases have failed to replicate under more rigorous conditions. Ego depletion, the idea that willpower draws from a limited pool, was once among the most cited findings in psychology. Large-scale replication attempts found effects that were much smaller than originally reported, or absent entirely. Priming effects, where exposure to a stimulus unconsciously influences subsequent behavior, have similarly weakened under scrutiny. The honest conclusion is that the catalogue of biases is smaller and less dramatic than the popular literature suggests, and treating every entry in a "list of 188 cognitive biases" as equally robust is itself a failure of evidence evaluation.

The bias bias. Knowing about cognitive biases can make you worse, not better, at reasoning. This happens through two mechanisms. First, people use bias labels as rhetorical weapons: "you're just anchored on that number" becomes a way to dismiss an argument without engaging with its substance. Second, the awareness of biases can produce a false sense of immunity. Research by Emily Pronin shows that people who learn about biases consistently believe they are less susceptible than average, a phenomenon she calls the "bias blind spot." Knowing about biases does not automatically correct them. It can, paradoxically, increase your confidence that you've already corrected for errors you haven't. This is