Paul Bloom on the Psychology of Children, and the Morality of Empathy and Disgust

Episode Summary

Executive Summary: Tyler Cowen and Paul Bloom explore what psychology can and cannot explain, emphasizing infants’ early moral and social intuitions, the mixed role of emotions like empathy and disgust in morality and politics, and how large language models are reshaping theories of language, therapy, and theory of mind. Bloom is skeptical of grand claims but optimistic about AI’s practical usefulness, especially in communication and counseling.

Main Topics: What psychologists actually understand (Priority: 5/5): Bloom argues psychologists know meaningful subdomains—memory, child development, mental disorders, group behavior—but are not especially better than others at real-time social perception or everyday judgment. Infant minds, morality, and development (Priority: 5/5): A major theme is that babies are not blank slates: they show early preferences for helpers over hinderers, early in-group/out-group cues, and early capacities that later develop into moral and religious understanding. Empathy, disgust, and moral reasoning (Priority: 5/5): Bloom defends his skepticism toward empathy as a moral guide, criticizes disgust in the moral sphere, and discusses how emotions can distort policy and everyday judgment while still serving social functions. Large language models and language theory (Priority: 5/5): Bloom says LLMs are the biggest surprise of his intellectual life because they challenge Chomsky-style claims about innate language modules and show that statistical learning can produce impressive language competence. AI therapists and theory of mind (Priority: 4/5): He thinks AI will soon handle a large share of therapy-like interactions and that successful use of LLMs requires a special kind of theory of mind and conversational flexibility. Moral boundaries, family, and sex (Priority: 3/5): The conversation includes Bloom’s view that adult sibling sex is morally wrong, largely because family-sex possibilities can corrode childhood relationships and sex is not always appropriate within intimate bonds. Psychology of prestige, interviewing, and work habits (Priority: 3/5): They discuss overrating intelligence, the value of flattery in job interviews, the bias toward smarts in hiring, and Bloom’s own unusual six-minute work bursts.

Key Arguments: Psychologists are not generally better than non-psychologists at reading people in everyday situations, though they do understand some specific domains such as memory, child development, and mental disorders. Babies are not blank slates; they show early preferences for helpers, sensitivity to social categories, and an early us-versus-them orientation. Some moral intuitions are likely innate and evolutionarily shaped, including early responses to helping vs. harming and social affiliation markers like language and accent. Empathy is a poor guide to good policy because it is biased, parochial, and often manipulated; overt empathic signaling can obscure substantive reasoning. Disgust is especially dangerous when used morally because it has historically fueled prejudice against gay people, interracial marriage, immigrants, and other groups. LLMs strongly challenge the idea that language requires an innate language module, since large-scale statistical learning appears capable of producing language competence. AI therapists may soon outperform humans in many short-term counseling interactions, and human-to-AI therapy will likely become common. Theory of mind is best tested in sustained, novel conversation or by predicting what happens next in a novel or TV episode, not by standardized tasks that can be gamed. Education increases intelligence somewhat, not just test-taking skill, by training abstraction and hypothetical reasoning. Bloom believes many people overvalue intelligence in hiring and relationship judgments, but honesty, productivity, and work ethic often matter more. He thinks flattery is underused in professional contexts because it is cheap and effective, while honest criticism is often strategically stupid when seeking a job.

Data Points: Baby preference for helpers: 9 months - Bloom cites experiments showing babies prefer characters that help a climber up a hill over those that hinder it. Child belief in God: about 4 years old - Bloom gives a rough age at which cultural support can make belief in God coherent for children. Belief in Santa Claus: 2–3 years old - He says Santa is an easier cognitive leap than God because Santa is concrete and sometimes visible in a department store. Infant happiness: around 1.5 to 2 years old - Bloom says life starts looking up as toddlers gain autonomy and better physical conditions, especially in wealthy countries. LLM therapy share: human interaction likely a minority within not too many years - Bloom predicts AI-driven therapy-like conversations will dominate many interactions soon. Education effect on IQ: each year in school raises IQ a little bit - He says education modestly boosts intelligence beyond mere test familiarity. Work interval: 6 minutes - Bloom says he often works in six-minute bursts using a timer. Chosen-item preference experiment: three-year-olds and capuchins - He cites research showing both show choice-induced liking changes consistent with cognitive dissonance. Monkey cognition study: capuchins - Bloom refers to studies with Laurie Santos showing capuchins display dissonance-like effects. LLM benchmark: GPT-4o was discussed on the day it came out - Bloom noted the timing of the interview coincided with GPT-4o’s release.

Pivotal Quotes: "We have no special ability to manipulate people, to understand people." — Paul Bloom: On whether psychologists are better at reading people than smart non-psychologists. "Probably the most misunderstood thing, period, is that they're blank slates." — Paul Bloom: On infant cognition and the common misconception that babies know nothing. "This is the biggest surprise in my intellectual life." — Paul Bloom: On the success of large language models and what they imply about language learning.

Implications: The episode suggests psychology is most useful when humble and domain-specific, while AI may soon transform language, therapy, and communication. It also pushes listeners to distrust empathy and disgust as default moral guides and to value practical judgment over status signals.