How a computational framework can account for the successes and failures of human cognition

At the heart of human intelligence rests a fundamental puzzle: How are we incredibly smart and stupid at the same time? No existing machine can match the power and flexibility of human perception, language, and reasoning. Yet, we routinely commit errors that reveal the failures of our thought processes. What Makes Us Smart makes sense of this paradox by arguing that our cognitive errors are not haphazard. Rather, they are the inevitable consequences of a brain optimised for efficient inference and decision making within the constraints of time, energy, and memory — in other words, data and resource limitations. Framing human intelligence in terms of these constraints, Samuel Gershman shows how a deeper computational logic underpins the 'stupid' errors of human cognition.

Embarking on a journey across psychology, neuroscience, computer science, linguistics, and economics, Gershman presents unifying principles that govern human intelligence. First, inductive bias: any system that makes inferences based on limited data must constrain its hypotheses in some way before observing data. Second, approximation bias: any system that makes inferences and decisions with limited resources must make approximations. Applying these principles to a range of computational errors made by humans, Gershman demonstrates that intelligent systems designed to meet these constraints yield characteristically human errors.

Examining how humans make intelligent and maladaptive decisions, What Makes Us Smart delves into the successes and failures of cognition.