Episode Summary
Executive Summary: Celeste Kidd explains her lab’s work on belief formation, curiosity, and probabilistic learning, showing that humans and infants continuously update expectations, seek information that is neither too familiar nor too surprising, and become less curious once certain. She argues these mechanisms matter both for understanding cognition and for designing technologies—especially online platforms and AI systems—that don’t amplify misinformation, bias, or overconfident false beliefs.
Main Topics: From journalism to science and data analysis (Priority: 4/5): Kidd describes her background in investigative reporting, CS, linguistics, and journalism, and how frustrations with simplistic data practices in journalism pushed her toward science and statistical modeling. Belief formation as continuous probabilistic updating (Priority: 5/5): Her lab studies beliefs as dynamic probabilistic expectations rather than fixed knowledge, including how concepts like 'table' shift with experience over time. Infant attention and the surprisal/curiosity link (Priority: 5/5): Using infant looking-time experiments and probabilistic models, the lab finds a U-shaped relationship: infants attend most to events that are mildly surprising, not too predictable or too surprising. Certainty reduces curiosity and can lock in false beliefs (Priority: 5/5): Work from the lab suggests that as people become more certain, they search less and weight new evidence less, which can help explain stubborn mistaken beliefs. Feedback-driven confidence and overconfidence without feedback (Priority: 4/5): Kidd discusses findings that certainty is shaped more by feedback than by evidence strength alone, and that in low-feedback settings people can become overconfident. Concepts differ across people, even for concrete objects (Priority: 4/5): The lab uses clustering and similarity judgments to show that there is not one shared concept for words like 'table' or 'cup'; people vary meaningfully in how they categorize even familiar objects. Implications for platforms, misinformation, and AI design (Priority: 5/5): Kidd argues that online systems are not neutral: they shape beliefs through ranking, repetition, and confirmatory exposure, so designers must account for human belief formation and bias.
Key Arguments: Human knowledge is not static; beliefs and concepts are continuously updated through experience and probabilistic expectations. Infants appear to allocate attention based on surprisal, preferring stimuli that are informative but still interpretable. A U-shaped curiosity curve means both extreme predictability and extreme novelty reduce engagement; optimal learning sits in the middle. Subjective certainty can suppress curiosity, causing people to stop sampling evidence and become stuck with weak or incorrect beliefs. Feedback strongly influences certainty, sometimes more than the actual evidence quality. Without feedback, people may become overconfident and assume their concepts are more shared than they really are. The fact that people use the same word with different concepts has major implications for communication, measurement, and social judgment. Technology platforms are not neutral because their design choices alter what people believe, what they seek next, and how quickly they settle on conclusions. AI and decision systems should be built with awareness of human bias and belief dynamics so they do not reproduce harmful distortions. Understanding belief formation can help explain why misinformation, pseudoscience, and polarized claims spread so effectively online.
Data Points: Infant attention pattern: U-shaped relationship - Infants maintain attention most when events are mildly surprising, with lower attention at both low and high surprisal. Activated charcoal search belief shift: ~3 clicks / 2–3 videos - People moved from neutral to believing activated charcoal is probably beneficial after just a few interactions. Activated charcoal wellness belief confidence: 80–90% - After a few clicks/videos, participants were highly likely to endorse the wellness-positive view. Concept clusters in population: about 5–10 - Clustering judgments suggests multiple concept variants exist for many words, including concrete objects.
Pivotal Quotes: "There is no neutral platform, I think that's a really ill-conceived way of thinking about these problems." — Celeste Kidd: Kidd’s main conclusion about online platforms and information systems. "When somebody's certain, they stop searching." — Celeste Kidd: Explaining why certainty can freeze learning and make false beliefs hard to revise. "What I wanted to make in the talk was that when you hear people say that this tool or this platform is neutral, that's, I would say, dishonest." — Celeste Kidd: Her critique of claims that algorithms and platforms merely reflect user preferences.
Implications: Designers of platforms, products, and AI systems should treat belief formation as a core mechanism, not a side effect. Interface order, feedback, and ranking can shape certainty, curiosity, and misinformation—so neutrality is an illusion.