Episode summary
Andrew Huberman interviews Caltech neuroscientist Ralph Adolphs about how to define and study emotions across humans, animals and—tentatively—AI systems. Adolphs argues emotions are best treated as functional control states, characterised by properties such as behavioural priority, scalability and temporal persistence, rather than reduced to single “centres” in the brain or equated with conscious feelings. He cites work with amnesic patients to suggest sadness can persist even when people cannot remember the stimulus that triggered it, and revisits his lab’s studies of the amygdala lesion patient known as SM, including findings that reduced fear responses to external threats can coexist with panic during carbon dioxide inhalation.
The discussion turns to emotion perception and social cognition. Adolphs critiques classic “basic emotion” facial-expression tasks associated with Paul Ekman, saying posed actor expressions and multiple-choice labelling inflate real-world performance; he points to a review he co-authored with Lisa Feldman Barrett. He also describes research directions on dynamics and context in emotion inference, and reports his lab is using webcam-based gaze tracking during Zoom-style videos to quantify individual differences related to autism traits.
On AI, Adolphs relays claims that large language models can infer Big Five personality traits from a page of written text at clinician-level performance, raising privacy and assessment questions. The episode ends with a personal reflection: after a serious health diagnosis, he says he leaned on interpersonal emotion regulation and a neuroscience-informed view that the perceived world is constructed by the brain and ultimately “lost” at death.
