The sense of should: A biologically-based framework for modeling social pressure

The sense of should (Theriault, Young & Barrett 2021)

Theriault, J. E., Young, L., & Barrett, L. F. (2021). “The sense of should: A biologically-based framework for modeling social pressure.” Physics of Life Reviews, 36, 100–136. doi:10.1016/j.plrev.2020.01.004 (DOI printed on the manuscript; PMC author manuscript nihms-1551627, published in final form March 2021). Funded by NIH U01 CA193632. See jordan-theriault, lisa-feldman-barrett.

This is the wiki’s first source on social pressure, conformity, and norms as biological phenomena, and its first attempt to run the interoception machinery it has been assembling — allostasis, predictive-coding, affect as a transform of interoceptive signals — outward, into social cognition. Where Barrett (2017) derives emotion from the body budget, this paper (Barrett senior, Theriault lead) derives why you conform to other people from the same budget. The through-line is that other people are part of your sensory environment, and an unpredictable environment is metabolically expensive.

The claim, in the paper’s own order

The argument is built as a deduction from a “minimal set of biological premises,” and it is easiest to hold in sequence — the structure mirrors the seven-step derivation on theory-of-constructed-emotion.

1. A brain is a predictive regulator of its body (Conant & Ashby’s “every good regulator of a system must be a model of that system”; allostasis). Sensation and cognition are in the service of regulation.
2. Prediction error carries a metabolic cost. Neuronal signalling dominates the brain’s energy budget (~75% of grey-matter cost is signalling; the brain is ~20% of the body budget at rest), and predictive processing minimizes it by transmitting only unpredicted signals. So encoding prediction error is where the cost lives. See metabolic-cost-of-prediction-error.
3. Metabolic costs are prospectively avoided (allostasis: anticipate and prevent, don’t react). Therefore unpredictable environments are metabolically costly, and organisms are motivated to make them predictable.
4. Other people are part of your environment, and — the hinge — you are part of theirs. The prediction error you get from a person who is also predicting you is reciprocal prediction error. You can regulate it: if your behaviour conforms to their prediction, they have less reason to change their behaviour, so they stay predictable to you.
5. Conforming to others’ expectations is therefore individually adaptive — not because it earns reward, but because it buys a predictable, metabolically cheaper social environment. That motivation, felt, is the sense-of-should.

The formal core is one proportionality (Equation 4; Figure 2): your reciprocal prediction error from person i, and its metabolic cost, is proportional to |b − pb_i| — the discrepancy between your behaviour b and their prediction pb_i of it. Minimizing that discrepancy (conforming) minimizes both how much they change and how much prediction error you get back.

Why this is an interoception paper

The sense of should is not a new drive bolted on; it is the core-affect machinery pointed at social prediction error.

• Affect is a low-dimensional transform of interoceptive signals (the paper cites Barrett & Simmons 2015, Craig 2015, Seth 2013 — the wiki’s interoception, interoceptive-inference sources). Prediction error generates arousal (electrodermal, pupillary, cardiovascular ANS correlates), and arousal “is not necessarily valenced” until context interprets it — the same valence/arousal claim on core-affect.
• So when you violate others’ expectations, their behaviour changes, your prediction error rises, and you experience arousal that — if intense or persistent — is interpreted as aversive. The “punishment” is not administered by anyone. No one pays a cost to censure you; others merely react to their expectations being violated, and you make aversive meaning of your own interoceptive sensations. The paper’s phrase: “a punishment that Amelia’s brain literally inflicts on itself.”
• Kohlberg’s two adolescents — same anxiety “in the pit of their stomachs,” one reads it as “being chicken,” the other as “the warning of my conscience” — is offered as the categorization of interoceptive sensation at the root of moral feeling. This is theory-of-constructed-emotion applied to obligation.

The sense of should is thus experienced as an anticipatory aversion to violating others’ expectations (allostasis again: you regulate the violation before it happens). Footnote 11 draws the clinical corollary the wiki should keep: social anxiety may be pathological over-estimation of how much one’s behaviour will disrupt the social environment, or over-weighting of aversive interoceptive experience (citing Khalsa et al. 2018) — see anxiety-sensitivity.

What is genuinely new here

Three constructs carry their own pages because they recur and are separable from the sense-of-should thesis:

• metabolic-cost-of-prediction-error — the biological engine. The bridge from the free-energy/predictive-coding literature to a motivational claim: because prediction error costs energy, a predictable environment is a resource, and unpredictability is a cost the organism will act to avoid. This is the premise the whole framework stands on.
• interactive-inference — the constructive counterpart to Barrett’s mental-inference-fallacy. To know what others expect, you cannot read their minds; you form a hypothesis from prior knowledge (dispositional and situational inference, framed as Bayesian posteriors), enact a behaviour, and use the resulting prediction error as evidence about the error in your inference. Violating expectations is a controlled experiment. Embodied, after Gallagher’s interaction theory.
• sense-of-should itself — the felt obligation, and the paper’s two coined terms for the influence it enables: stabilizing influence (controlling others by conforming to them — keeping them predictable) vs directing influence (causing specific behaviour in others, e.g. by communicating). “The most straightforward way to control someone else’s behaviour may be just to make them aware of what you want.”

A fourth idea, constructing vs coasting, folds into the metabolic engine: organisms trade off paying to encode information (building a more general model) against exploiting the model they have. The sense of should is a strategy for coasting — but the same behaviour–expectation contingency that lets you coast is what makes interactive inference (constructing) possible, so neither dominates.

The extensions (Section 4)

The paper’s reach is deliberate and is where the falsifiability worry lives. Each is offered as a phenomenon the framework “unifies”:

• Status quo bias. If a sense of should is widespread, everyone’s small conformist sacrifice funds a shared predictable environment, so everyone has a stake in maintaining it — and the more predictable it is, the more each individual loses by disrupting it. This is offered as a mechanism for why people defend even an oppressive-but-stable order (illustrated with King’s Letter from Birmingham Jail), and why destabilization can flip it fast.
• Communication and language. The mirror of conformity: instead of changing b to match pb_i, you issue signals to change pb_i to match b (speech-act theory; “I’ll use this pen”). If others also have a sense of should, communicating your expectation (an imperative) lets you exercise directing influence over them.
• Behavioural economics / game theory. In a prisoner’s dilemma the sense of should is inert by design — the game removes real-world uncertainty, which is exactly the condition the sense of should exists to manage. And it is shown formally equivalent to the guilt-aversion model of Chang et al. (2011): both make the agent minimize the discrepancy between behaviour and inferred expectation. The paper’s twist, via constructed emotion: “guilt” is not a thing but a category of heterogeneous instances unified by the sense of should — guilt is one instance under the umbrella, not synonymous with it.
• Culture and norms. The logic works regardless of the content of the expectation — only the discrepancy matters. So arbitrary expectations, once shared, motivate conformity with no functional payoff and no explicit punishment, which “poses serious problems for hypotheses about innate, functionally adaptive cognitive modules” (contra Cosmides & Tooby). A concrete mechanism for cultural inheritance.
• Beliefs, including scientific ones. If close others can infer your beliefs, the sense of should can compel you to hold beliefs that keep the social environment predictable — an extension of cognitive dissonance, and (via Kuhn) a claim that scientific communities’ shared expectations exert real social pressure on what scientists believe.

How it sits in the wiki

No hard contradiction — checked deliberately (auto-ingest writes no pages on one). This is a Barrett-lab framework consonant with the constructivist thread the wiki already hosts: it presupposes allostasis as the core task, affect as interoceptive, constructed emotion for guilt, and predictive-coding/active-inference for the mechanism. It extends rather than collides. The one place it presses on existing content is mental-inference-fallacy: Barrett names the fallacy of inferring internal states from behaviour; this paper (same lab) offers the positive account of how minds are nonetheless inferred — not by reading states off behaviour, but by iteratively testing hypotheses against prediction error. The two are complementary, and the paper is careful to avoid “theory of mind” language and its propositional-representation baggage.

What to hold loosely. The framework inherits the constructed-emotion family’s structural weakness: enormous scope, a flexible core, and no clearly stated disconfirmer. It also rests entirely on the metabolic premise — that prediction error is costly enough to motivate — which the paper concedes is quantitatively unspecified (the ∝ signs are load-bearing, and footnote 2 admits the micro-realization of metabolic cost is unsettled). And it is a target article: Physics of Life Reviews publishes it with open peer commentary, which the wiki does not have but should assume exists.