A. D. (Bud) Craig
Neuroanatomist (Barrow Neurological Institute) whose account of interoception and the insula is the reference framework the modern field is built on — and that Seth both extends and departs from. See his two foundational papers craig-2002-interoception and craig-2009-anterior-insula.
The anatomy: interoception as the afferent side of homeostasis
Craig’s core contribution is the lamina-i-spinothalamocortical-pathway: small-diameter (Aδ/C) afferents from all tissues → lamina I of the dorsal horn → the dedicated thalamic relay VMpo → the dorsal posterior insula. This is the “long-missing afferent complement of the efferent autonomic nervous system” — the sensory limb of homeostasis. Its consequence is a reclassification of pain, temperature, itch, and sensual touch as interoceptive (feelings from the body) rather than aspects of exteroceptive touch.
The insular hierarchy and the emotion dyad
A posterior→mid→anterior gradient: primary objective interoceptive maps (posterior) → integration (mid) → subjective re-representation (“how you feel”) in the anterior insula (AIC), right-lateralized in humans. Craig pairs the insula (limbic sensory cortex, feeling) with the ACC (limbic motor cortex, motivation/agency): together they constitute an emotion. See insular-cortex.
Awareness (Craig 2009)
Craig extends the hierarchy to consciousness: the AIC is proposed to “engender human awareness” via the global-emotional-moment model, with von-economo-neurons as the substrate and the salience-network as the functional home. He grounds James’s and Damasio’s (somatic-marker-hypothesis) intuitions in specific anatomy, and thus in embodied-selfhood (“the material me”).
Point of divergence from Seth
Craig’s architecture is an ascending re-representation hierarchy with a comparator at the top (AIC) — feelings are read out from integrated afferent signals. Seth adopts the same anatomy but recasts its computation as top-down predictive inference (predictions descend, prediction errors ascend). Notably Craig 2009 already gestures at prediction (the AIC represents “predictions of future feelings”; anxiety as “distorted interoceptive predictions”), so the divergence is about architecture (ascending-integration vs descending-generative), not whether prediction exists at all. See feedforward-vs-predictive-interoception.
A change of emphasis, not of fact (Berntson & Khalsa 2021)
The wiki’s afferent anatomy has been Craig’s, and on his pages the lamina I → VMpo → dorsal posterior insula route reads as the substrate of interoception. Berntson & Khalsa (2021) present it as the best-characterized channel among several — vagal, cranial V/VII/IX, sacral, humoral, and direct central chemoreception by neurons and astrocytes, the last two reaching behaviour with no afferent fibre at all.
This is not a contradiction and is recorded as none: Craig’s claim concerned a dedicated, modality-labelled, phylogenetically recent pathway, never exclusivity, and one of the additional routes (a direct NTS → insula/SII vagal-activated pathway in macaque) is sourced to Strigo & Craig 2016 — his own later work. What it corrects is an emphasis the wiki had absorbed from reading Craig first.
The phylogeny claim gets a two-sided update. Berntson & Khalsa cite Craig’s human-distinctiveness argument as one of the two reasons for the “chasm between animal and human studies,” and report that there are no animal models of interoceptive awareness — Craig’s boundary, restated as a missing method rather than missing hardware. But they also report mouse insular cortex anticipating expected satiety (Livneh et al. 2020), i.e. an animal generating and updating interoceptive predictions. Whether that counts as feeling is exactly what cannot be asked. See can-we-know-animal-feelings, insular-cortex.