False Patterns and the Overactive Brain: Why We Can't Stop Finding Meaning
Show someone a random sequence of coin flips and they will find a pattern. Show someone a random scatter of dots and they will see shapes. Show someone a sequence of unrelated stock price moves and they will construct a narrative of cause and effect.
The mind is a pattern-detection machine that cannot decline to detect. This is not stupidity and not ignorance — it is the design of a cognitive system that evolved in an environment where pattern detection was survival-critical and the cost of a false positive (seeing a tiger that wasn't there) was much lower than the cost of a false negative (not seeing a tiger that was there).
Nassim Taleb summarizes this: "Our overactive brains are more likely to impose the wrong, simplistic narrative than no narrative at all."
The consequences of this design, in a complex modern world full of noise, are exactly what you'd expect.
The Pattern-Detection Bias
The mismatch is precise. The ancestral information environment was: - Data-sparse: limited inputs from a small geographic area and social network - High-stakes: most patterns had survival implications - Low-noise: most variation was either signal or white noise, not the structured near-noise of financial markets
The modern information environment is: - Data-rich: staggering input volumes from global sources in real time - Variable-stakes: most inputs are irrelevant to anything consequential - High-noise: financial prices, political news, and social commentary are dominated by structured noise that resembles signal
The brain was calibrated for the first environment. It operates in the second. The result is pattern-detection in overdrive in a domain where most detected patterns are false.
"There is this mismatch between the messy randomness of the information-rich current world, with its complex interactions, and our intuitions of events, derived in a simpler ancestral habitat."
Randomness Looks Like Order; Order Looks Like Randomness
"Randomness is indistinguishable from complicated, undetected, and undetectable order; but order itself is indistinguishable from artful randomness."
This is the deep problem. It's not just that we see false patterns where there are none. It's that genuine patterns and genuine randomness look, to our pattern-detection system, similar. Complex structured processes generate output that looks random to casual inspection. Random processes generate output that looks patterned to the same inspection.
The financial market example: stock price movements are partly driven by real information (earnings, macro conditions, competitive dynamics) and partly by noise (market microstructure, liquidity effects, behavioral patterns). The noise and the signal look similar. Our pattern detection cannot reliably distinguish them. We confidently detect patterns in the noise and dismiss the signal because it "looks random."
The remedy is not better pattern detection — the brain has already maxed out its pattern detection. The remedy is calibrated skepticism about detected patterns: the discipline of asking, before acting on a pattern, whether this detected pattern is more likely genuine or manufactured.
Narrative Restraint as a Cognitive Skill
Taleb's recommendation is the discipline of withholding the explanation — not rushing to close the open question with a story.
Most cognitive environments penalize narrative restraint. In meetings, the person without a confident explanation appears unprepared. In investment pitches, the manager without a thesis sounds aimless. In media, the commentator who says "this is noise and I can't distinguish it from signal" gets replaced by the commentator who generates a confident narrative every fifteen minutes.
The incentive structure rewards narrative generation, not narrative accuracy. The confident explainer wins the immediate social exchange. The person who withholds explanation until evidence is strong looks hesitant.
But in high-noise domains, the confident explainer is systematically generating expensive false patterns and acting on them. The person who withholds is correctly identifying the signal-to-noise ratio and behaving accordingly.
Practical Application: The Test Before the Story
Before accepting a detected pattern as genuine:
What was the sample size? A pattern detected in five events is more likely false than a pattern detected in 500 events. The brain generates high-confidence patterns from tiny samples; the pattern's significance should scale with sample size.
What was the alternative hypothesis? What patterns would you have found if the data had been random? If random data would have generated similar-looking patterns (it almost always does), the detected pattern is not strong evidence.
Has the pattern survived prediction? A pattern detected in historical data and then tested against new data is more likely genuine than a pattern only found in the historical fitting. Most patterns that survive historical fitting do not survive forward prediction.
Does acting on this pattern require being right more often than a coin flip? If your edge is small, the transaction costs of acting on it may eliminate the edge. The conservative estimate of how genuine your pattern is should account for the noise floor.
"The average of expectations is different from the expectation of averages." Most of what we call "pattern" is the former dressed up as the latter.
For the full framework, read The Bed of Procrustes Explained.