Simple Heuristics That Make Us Smart

1. Bounded rationality replaces the unrealistic "demons" of unbounded optimization with simple, fast, and frugal heuristics.

We propose replacing the image of an omniscient mind computing intricate probabilities and utilities with that of a bounded mind reaching into an adaptive toolbox filled with fast and frugal heuristics.

The demise of demons. Traditional models of rationality assume the human mind operates like an omniscient "demon" with unlimited time, knowledge, and computational power. In reality, humans must make decisions under severe constraints. Bounded rationality offers a realistic alternative, viewing the mind as an adaptive toolbox of specialized, domain-specific tools.

The three building blocks. Fast and frugal heuristics are built from simple, easily computable rules. These rules govern how the mind searches for information, when it stops searching, and how it makes a decision.

• Search rules: Guide the search for alternatives or cues (e.g., random or ordered search).
• Stopping rules: Terminate the search without complex cost-benefit calculations (e.g., stop at the first discriminating cue).
• Decision rules: Make the final choice based on the gathered information (e.g., one-reason decision making).

A realistic perspective. Rather than striving for unattainable optimization, these tools allow us to make "good enough" decisions. This perspective shifts the focus from logical consistency to real-world survival and success.

2. Ignorance can be an asset, enabling the "less-is-more" effect where knowing less leads to more accurate inferences.

If one of two objects is recognized and the other is not, then infer that the recognized object has the higher value.

Exploiting partial ignorance. The recognition heuristic is the simplest tool in the adaptive toolbox. It relies on the ancient, highly developed capacity for recognition memory to make rapid inferences. When faced with a choice between a recognized and an unrecognized object, the mind bets that the recognized one has a higher value on the target criterion.

The less-is-more effect. This heuristic leads to a highly counterintuitive phenomenon: those who know less can make more accurate inferences than those who know more. When we recognize only some objects, we can use our partial ignorance to guide our choices.

• An ignorant person who recognizes no objects must guess (50% accuracy).
• A partially ignorant person can use the recognition heuristic to achieve high accuracy.
• An expert who recognizes all objects cannot use the heuristic and must rely on other, potentially noisier cues.

Ecological fit. The recognition heuristic is ecologically rational because recognition is systematically correlated with real-world criteria. Media coverage, social discussion, and environmental exposure act as mediators that reflect inaccessible criteria like city size or company power.

3. Simple name recognition can construct stock portfolios that outperform professional mutual funds and market indices.

The predictive power of the recognition heuristic corroborates the notion that a lack of recognition can contain implicit knowledge as powerful as explicit knowledge.

Ignorance-based investing. To test the real-world power of the recognition heuristic, researchers threw it into the highly volatile stock market. They asked laypeople and experts in the US and Germany which major companies they recognized, then built portfolios based on these recognition rates. The results challenged the core tenets of traditional finance theory.

Outperforming the experts. Portfolios consisting of highly recognized stocks consistently outperformed unrecognized stocks, major market indices, and professionally managed mutual funds.

• German laypeople's recognized portfolio gained 47% in six months, easily beating the Dax 30 index.
• International portfolios based on the recognition of completely ignorant foreign laypeople performed the best.
• Randomly selected "dartboard" portfolios and expert-selected portfolios fell far behind.

The value of a name. This success occurs because public name recognition is strongly correlated with economic indicators like market share and profitability. Companies with dominant market presence are highly recognized by the public, making collective ignorance a highly profitable investment guide.

4. Complex calculations are unnecessary when you can "take the best and ignore the rest" to make highly accurate decisions.

Take The Best first tries the cue with the highest validity, and if it does not discriminate, the next best cue, and so on.

The power of one reason. When recognition fails because both objects are recognized, the mind must search for further cues. The Take The Best (TTB) heuristic does this by searching through cues in order of their validity. It stops searching the moment it finds a single cue that discriminates between the two options, basing its entire decision on that single reason.

Noncompensatory choices. TTB is a noncompensatory strategy, meaning that no amount of contrary evidence from later, unexamined cues can overturn the decision made by the first discriminating cue. This stands in stark contrast to classical "moral algebra" which requires weighing and adding all available reasons.

• It avoids the need to convert different cues into a common currency.
• It bypasses the complex calculations required to resolve conflicting information.
• It drastically reduces the cognitive effort and time required to make a choice.

Astonishing accuracy. Despite its extreme frugality, TTB matches or exceeds the accuracy of complex statistical models like multiple regression. It proves that the mind can achieve high accuracy by focusing on the most important cue and ignoring the rest.

5. Simple heuristics generalize better to new situations because they do not overfit noisy data like complex statistical models do.

In generalization, in contrast, more is not necessarily better.

The danger of overfitting. When statistical models like multiple regression are fitted to existing data, they use many parameters to capture every detail, including random noise. While this produces a perfect fit for past data, it often fails miserably when predicting new data. This failure of generalization is known as overfitting.

Frugality breeds robustness. Simple heuristics like Take The Best are highly robust because they use very few parameters. By ignoring the noise inherent in minor cues, they focus only on the most significant, systematic forces in the environment.

• In a 20-environment test, multiple regression's accuracy dropped by 9% when generalizing to new data.
• Take The Best's accuracy dropped by only 5%, making it the overall winner in predictive accuracy.
• Even the ultra-simple Minimalist heuristic, which searches cues randomly, performed on par with regression in generalization.

Less is more in prediction. This finding turns the "more-is-better" information ideology on its head. In the real world, where we must constantly make predictions about the future, limiting information search is not a regrettable compromise but an adaptive necessity.

6. The success of a heuristic lies not in its logical consistency, but in how well its structure matches the information structure of the environment.

A heuristic is ecologically rational to the degree that it is adapted to the structure of an environment (see below).

The match of mind and world. Rationality is not a purely internal matter of logical consistency or probabilistic coherence. Instead, ecological rationality is defined by how well a cognitive tool fits the structure of the physical and social world. A simple heuristic can outperform a complex calculator simply because its structure exploits the natural distribution of information.

Exploitable environmental structures. Heuristics are designed to tap into specific environmental patterns to bypass heavy computations.

• Noncompensatory environments: When cue validities decay exponentially, no linear model can beat Take The Best.
• Scarce environments: When very few cues are available, simple heuristics easily outperform equal-weight linear models.
• Skewed distributions: J-shaped environments allow heuristics to make rapid, coarse-grained estimates.

Domain-specific tools. Because different environments have different structures, the adaptive toolbox contains highly specialized tools rather than a single, general-purpose power tool. The key to understanding human intelligence is analyzing this mutual adaptation between the mind and its environment.

7. Misremembering our past predictions is a small price to pay for a memory system that constantly updates itself with new information.

We suggest that hindsight bias is a by-product of an adaptive process, namely knowledge updating.

The reconstruction of memory. Human memory is not a permanent, static archive of past events. Because maintaining an infinite number of past traces is computationally and biologically expensive, our memory system constantly updates itself with new, highly relevant information. When we are asked to recall a past judgment and direct retrieval fails, we reconstruct it using our currently updated knowledge.

The RAFT model. The Reconstruction After Feedback with Take The Best (RAFT) model explains how this process inevitably leads to the "knew-it-all-along" effect.

• Step 1: Direct retrieval of the past judgment is attempted; if it fails, reconstruction begins.
• Step 2: The mind uses its current, updated cue knowledge to run the Take The Best heuristic again.
• Step 3: Because feedback has altered the cue values to align with the correct outcome, the reconstructed judgment shifts toward the feedback.

A price worth paying. Hindsight bias is therefore not a design flaw or a sign of cognitive decay. It is the inevitable, minor cost of a highly efficient, self-updating memory system that prioritizes currently useful information over outdated past states.

8. The QuickEst heuristic exploits skewed, J-shaped environments to make rapid quantitative estimates using minimal cues.

QuickEst's policy is to use environmental structure to make estimates for the most common objects (e.g., in the cities environment, the smallest cities) as quickly as possible.

The law of the higher, the fewer. Many real-world quantities—such as city populations, company sizes, and the distribution of wealth—follow highly skewed, J-shaped distributions. In these environments, there are a vast number of very small items and only a tiny handful of extremely large ones. The Quick Estimation (QuickEst) heuristic is specifically designed to exploit this skewed structure.

The sorting mechanism. QuickEst works like a conveyor belt that sorts coal, sifting out the most common, small items first using a negative-biased stopping rule.

• It ranks cues by the average size of the objects that lack the cue, starting with the smallest.
• It stops searching the moment it encounters a "no" (a negative cue value).
• It estimates the size based on this first negative cue, rounding it to the nearest "spontaneous number" (e.g., 100, 150, 200, 300).

Frugal and accurate. By stopping at the first negative cue, QuickEst estimates the size of the many small objects almost instantly. In environments where knowledge is scarce, this fast and frugal heuristic easily beats multiple regression and complex estimation trees in absolute accuracy.

9. Complex categorization can be achieved quickly by sequentially eliminating possibilities using one cue at a time.

Categorization by Elimination is a fast and frugal, noncompensatory, cue-based model of categorization.

Sequential narrowing. Traditional psychological models of categorization assume that we must integrate all available features to place an object in a category. However, in urgent real-world situations, we must categorize objects quickly using minimal information. The Categorization by Elimination (CBE) heuristic achieves this by using cues sequentially to eliminate impossible categories.

The elimination process. CBE processes cues in order of their success, using each cue to narrow down the set of possible categories.

• It maps the observed cue value to a pre-constructed "bin" of possible categories.
• It intersects this new set of categories with the previous set, eliminating any categories that do not match.
• It stops immediately when only a single category remains, ignoring all other unexamined cues.

Graceful and robust. In tests on complex, multi-cue datasets (such as identifying wines or classifying mushrooms), CBE matched the accuracy of complex neural networks and exemplar models. Crucially, when some cues were missing, CBE proved far more robust, while the competing models suffered catastrophic drops in accuracy.

10. In sequential search with mutual choice, successful pairing requires learning one's own mate value through feedback rather than seeking absolute perfection.

Choosing a mate should not be a scientific affair.

The challenge of mutual choice. In simple sequential search, like shopping for a television, the objects of our search cannot reject us. But in mate search, choice is mutual: You must choose someone who also chooses you. Standard search rules like the 37% rule or the "Try a Dozen" heuristic fail miserably in mutual search because they set aspirations too high, leaving most of the population unmated.

Learning your own value. To search effectively in a mutual market, you must know your own mate value so you can target prospects of similar quality. Because we cannot know our own value a priori, we must learn it through the feedback of offers and rejections we receive from others during an initial "adolescence" phase.

• If someone of higher value proposes, we raise our self-image and aspiration level.
• If someone of lower value rejects us, we lower our self-image and aspiration level.
• This feedback loop allows us to quickly calibrate our expectations to our actual market value.

Stable and egalitarian sorting. This simple, feedback-based satisficing heuristic successfully pairs up a large proportion of the population. It naturally sorts individuals into well-matched, stable pairs without requiring any complex calculations of population means, standard deviations, or game-theoretic equilibria.