6 ideas
Baseball players do not compute a fly ball's trajectory. They fix their gaze on the ball, start running, and adjust their speed so the angle of gaze stays constant. This one rule gets them to the landing point while ignoring every variable a physics model would need, such as wind, spin and initial velocity.
If you recognize one of two options and not the other, infer that the recognized one scores higher on the criterion. This works whenever recognition correlates with the target, as it does between name familiarity and city size. So people who know less can outperform experts: German students did better than American students at judging whether San Diego or San Antonio is larger, because the Americans recognized both and could not use the cue.
A heuristic is neither rational nor irrational in itself. Its quality depends on how well its structure matches the structure of the environment, such as cue validities, noise and predictability. This shifts the question from 'does the mind violate logic?' to 'in which environments does this rule succeed or fail?' Many laboratory 'biases' look sensible once you see the real-world environment the mind is adapted to.
Complex models fit past data well by absorbing noise, so they overfit and generalize poorly when the future is uncertain and samples are small. Simple rules that ignore information have less variance and are more robust out of sample. For example, spreading money equally across N assets (1/N) often matches or beats mean-variance portfolio optimization, and Markowitz himself reportedly used 1/N for his own retirement savings.
Doctors at a Michigan hospital were sending too many patients to the coronary care unit. They switched to a fast-and-frugal decision tree that asks at most three yes/no questions in sequence: is there an ST-segment change, is chest pain the chief complaint, and is any one of several other risk factors present. The tree used less information than a logistic-regression tool yet was more accurate at identifying true heart attacks, and physicians could actually remember and use it.
Conditional probabilities ('90% sensitivity, 9% false-positive rate') confuse doctors and patients alike. The failure lies in how the information is represented, not in human reasoning ability.