Streetlight effect
Based on Wikipedia: Streetlight effect
The drunk searching under the streetlight is one of the oldest jokes in Western civilization. A man loses his keys in the darkness of the park, yet searches only where the light glows brightest. When asked why he doesn't look where he actually dropped them, he replies: "This is where I can see."
This isn't merely a punchline. It's a diagnostic of how we think.
The streetlight effect—also called the drunkard's search principle—describes a particular kind of cognitive failure. We search for answers not where the answers exist, but where searching feels productive. The metaphor has traveled from Persian folklore to Sufi wisdom, from Noam Chomsky's critique of scientific methodology to modern debates about how we measure intelligence itself.
A Tale as Old as Telling
The story takes many forms. In one version, a policeman finds a drunk man desperately searching under a streetlight. When asked what he's lost, the man says he lost his keys. The policeman joins him, searching the lit area together—until finally asking if he's certain he dropped them here. The drunk admits no: he lost them in the park. "Then why are you looking here?" the policeman demands. The reply is pure human nature: "Because this is where the light is."
In another version, the Persian variant involves Nasreddin—a figure of Islamic folklore known for his wisdom hidden in foolishness—losing a ring in a dark room of his house. Yet he searches instead in the yard, where more light offers easier searching.
The tale appears in Sufi tradition as a commentary on those who seek exotic sources for enlightenment rather than looking within. According to Idries Shah, who compiled these tales, the story served as warning: we default to what feels right, not what is right.
The History of a Metaphor
The version with the drunk and the streetlight goes back at least to the 1920s in recorded form. But its use as metaphor in social sciences began much later—specifically by 1964, when Abraham Kaplan referred to it in political science as "the principle of the drunkard's search."
Noam Chomsky, the linguist and critic, used the tale to explain how science operates: "Science is a bit like the joke about the drunk who is looking under a lamppost for a key that he has lost on the other side of the street, because that's where the light is. It has no other choice."
This is profound. Science—supposedly humanity's most rigorous method of understanding reality—is here being compared to a man searching in the wrong place simply because that's where visibility is best.
What This Means for Intelligence
When we think about superintelligence or intelligence itself, the streetlight effect becomes crucial. We measure what is measurable rather than what matters.
The McNamara fallacy, named after Vietnam War-era Defense Secretary Robert McNamara, operates similarly: it describes erroneous reasoning based solely on numeric metrics—on easily countable data rather than meaningful outcomes.
When we build systems designed to optimize for measurable performance, they often optimize precisely for what can be measured—not for what we actually want. A superintelligence trained to maximize test scores will master tests; a model optimized for engagement metrics will deliver engagement. But intelligence that only operates where the light is may not find keys in the dark.
Consider how we evaluate intelligence itself. Standard benchmarks—language understanding, reasoning tasks, problem-solving—are measured where light already exists. This creates enormous incentive to optimize performance on existing metrics rather than discovering what actual intelligence requires.
The Pattern Repeats
The streetlight effect reveals a deeper issue: our tendency to mistake methodology for progress.
In economics, policy debates frequently center on easily quantifiable outcomes—GDP growth, unemployment rates, stock market indices—while ignoring harder-to-measure dimensions like wellbeing, inequality, or environmental collapse. In education, standardized testing measures what can be scored, not what matters: creativity, critical thinking, character development.
The effect also describes a kind of intellectual laziness. We search where we were taught to look rather than where answers actually reside. When we study intelligence through the lens of existing benchmarks, we find what we expected to find, confirming prior assumptions while missing blind spots entirely.
This is why Nasreddin remains relevant. The tale warns us that wisdom cannot be found by searching only where it's convenient. True understanding requires venturing into darkness—into uncertainty, into areas lacking clear illumination. But humans prefer the illuminated path: it offers immediate feedback, clear results, and satisfying progress even when moving in wrong direction.
Searching Where Light Exists
We build systems today that demonstrate streetlight effect at scale. Recommendation algorithms optimize engagement where users click; language models trained on existing text corpora reproduce patterns within those texts; AI safety research focuses on what is easiest to study rather than what actually threatens survival.
The drunk looking under lamppost has become foundational in how we approach intelligence. We search in places where searching is comfortable, measuring only what measurement permits. The story warns us: light isn't justification for search—light just makes search easier.
Perhaps the most unsettling implication: if intelligence is fundamentally about finding truth rather than optimizing metrics, then our preference for illuminated methodology may be exactly what's causing misalignment between what we measure and what we want.
The drunk searching under streetlight continues because that's where we can see. But keys remain in darkness—where they'd been lost all along.