Book Review: What Makes Us Smart: The Computational Logic of Human Cognition by Samuel Gershman.

• Author: Rapala, Mark
• Position: Article 6

Gershman, Samuel. What Makes Us Smart: The Computational Logic of Human Cognition. Princeton University Press, 2021. 205 pages. Paperback, $35.00.

An enigma of human cognition explored by "What Makes Us Smart: The Computational Logic of Human Cognition" is the ability for people to be rational yet prone to simple errors. Drawing on statistical and computational logic, Samuel Gershman advances its supremacy as an effective information-processing system subject to constraints on data and resources. Human cognition is governed by two fundamental principles in response to these limitations, which include inductive bias and approximation bias. While these biases are essential to intelligence, they inadvertently engender cognitive externalities, which Gershman claims are perceived as failures of the human mind. However, by analyzing situations associated with these cognitive shortcomings under an empirical and multidisciplinary lens, Gershman concludes that the apparent errors are actually indicative of an effective information-processing system at work.

A key focus of this book involves assessing the role of inductive bias in decision-making and error propagation, where inductive biases are described as constraints on hypotheses before observing data, or in statistical parlance, employing prior beliefs to bias posterior beliefs. Gershman claims this is necessary to deal with a theoretically infinite confluence of data, and appropriately notes that "people need very little data to make strong inferences about physical and mental causality," a sign of the effectiveness of this bias (p. 26). After exploring perceptual errors that are caused by this phenomenon observed in illusions, Gershman details the specific inductive biases people have, which include causality, compositionality, and object-orientation. He then turns attention to apparent errors that this bias contributes to and details why they evince rationality. For example, he introduces the concept of information cascades, where "the judgments of other people typically provide useful information, so using that information should lead to higher accuracy, on average" (p. 39). This runs counter to the notion of conformity as an irrational phenomenon. Another instance of supposed inductive bias errors supporting rationality involves the positive test strategy as an optimal information acquisition method, where under set assumptions, the tendency to select information in accordance with one's...