Klaas Enno Stephan
The most-cited name in this wiki without a page, until now. “Stephan et al. 2016” appears on computational-psychiatry, allostasis, interoceptive-inference, interoceptive-psychopathology, sickness-behaviors, visceromotor-areas, anterior-cingulate-cortex and core-response-network — always as the source of allostatic self-efficacy, always at one remove through Khalsa’s roadmap or Quadt et al. Created under the load-bearing-figure convention, now that the wiki has a paper he is senior author on.
The idea the wiki runs on
Allostatic self-efficacy is a two-step argument:
- If the brain’s business is controlling bodily states, then chronic failure to control them produces persistent unresolved interoceptive surprise.
- A higher level of the hierarchy reads that persistence as evidence about itself — a metacognitive belief that one cannot regulate one’s own body — and the psychological expression of that belief is fatigue and depression.
What makes it distinctive among the wiki’s computational proposals is where it puts the pathology. Barrett & Simmons’s EPIC locates illness in the precision structure of the prediction/error loop itself; Stephan locates it one level up, in a belief about the loop’s performance. Depression becomes, in part, a verdict rather than a mood. computational-psychiatry holds the mechanism; Quadt et al. give it the wiki’s most concrete narrative, with chronic cortisol disrupting the belief-updating machinery and locking the verdict in.
The same move recurs in Petzschner et al.’s requirement that models of interoception treat consciousness and metacognition as separate variables: “modeling an individual’s response to internal states and their appraisal may be as important as modeling the signals themselves.” That sentence is this idea generalized into a modelling constraint.
The TNU posture
Petzschner’s page describes the group’s programme — generative modelling as a clinical instrument, Marr’s levels kept apart, computational biomarkers as the deliverable — and that is Stephan’s programme as its director. Two features are visible in the 2021 study and worth recording as a signature:
They pre-register, and it costs them. The analysis plan specified a protected exceedance probability above 90% for model selection. No model reached it. The plan’s fallback rule (take the simplest) is what licensed the Rescorla-Wagner analysis — so the paper’s entire computational-neural section rests on a model the data declined to prefer, reported as such, because the rule was fixed in advance. The wiki should read that as the pre-registration working rather than failing.
They test their own framework and report the misses. The paper’s central negative result — no anxiety effect anywhere in prediction error, contradicting Barrett & Simmons, Brewer et al., and Paulus & Stein — is a null against the family of hypotheses this group belongs to. Stephan et al. (2016) is itself cited in the paper as the source of the proposal that error is propagated to metacognitive areas, which is the one part that did find support (FDT metacognitive performance ↔ aIns error activity, r = 0.42).
Placement
The architect of the wiki’s computational-clinical thread, as Petzschner is its modeller, Khalsa its convener and Paulus its clinician. He supplies the hypothesis the clinical pages lean on hardest and, through his group, the first instrument capable of testing any of it.
Held at one remove
Stephan et al. (2016) is not in raw/, so allostatic self-efficacy is still known here only through others’ summaries of it — an odd situation for a claim this load-bearing. Nor are the HGF papers or the Bayesian model-selection papers, all of which the wiki now describes in methodological detail via breathing-learning-task.