Document Code: GRL-T1-013-EN
Track: Track I — Standards & Problem Definition
Category: Collective Judgment & Pressure (집단 판단과 압력)
Series: Definition / Conditions
Author: Gungri Research Lab / Jung Yuna
Published: June 16, 2026
Version: v4.1
Abstract
The alignment of a crowd in one direction has already been described, across sociology, economics, and crowd dynamics, in a number of stage models (Smelser 1962; Blumer 1939; Kindleberger–Minsky; Bikhchandani, Hirshleifer & Welch 1992; Granovetter 1978). This article does not propose a new stage division. Instead, it consolidates the stage structure these literatures share, and then identifies the state of individual judgment at each stage — unmet conditions, the withholding of one’s own judgment-output (deferral/HOLD or active yielding), the limit of sustaining it, and acceptance of an external standard. It then asks why judgment transfers outward, into how many types that delegation divides, how the transfer can be observed, why accountability is deferred in the process, and what consequences accumulate. The contribution is not the stages but the judgment-state layer placed over them, from which one testable prediction is derived. This layer does not replace the established mechanism of any stage. It offers no prescription or conclusion.
This document does not provide conclusions or recommendations. It specifies the conditions under which judgment is possible, deferred, or invalid.
Definitions
| Term | Definition |
|---|---|
| Withholding of Judgment-Output (deferral/yielding) | A state in which one withholds one’s own judgment-output. It includes two modes — (a) deferral (HOLD), recognizing that conditions are unmet, and (b) active yielding, where even a sufficient own signal is set aside because the group’s or system’s signal is taken to be better. |
| Limit of Sustained Deferral | The point at which time pressure, loss, or arousal makes the deferral state impossible to hold. |
| Acceptance of an External Standard | When deferral cannot be held, taking the conclusion of another, the group, or a system into one’s own place of judgment. = judgment delegation. |
| Accountability Dispersion | A state in which the acceptance of an external standard goes unrecorded, making it impossible to locate the judging subject after the fact. |
These terms are not a new mechanism. They are descriptive tools for the state of individual judgment over the established stage models below.
§1. The Phenomenon — No One Decided, Yet Everyone Moved the Same Way
A decision is made in a meeting room. Afterward, no one says, “I judged that.” Each person says, “that was the mood,” “everyone leaned that way.”
After an organization adopts an automated system, asking why a flawed outcome occurred returns the answer, “the system judged it so.” The subject who decided to trust the system is nowhere to be found.
Faster forms exist too. A withdrawal run spreading through a financial institution; a crowd surging one direction in a dense space; a cascade of sell orders in a market; a wave of online opinion rushing one way. All four share a feature: individuals did not fully judge for themselves, yet the group moved in one direction, and afterward there is no subject who decided that direction.
This article addresses that shared structure. Not “why do people get swept along,” but through what stages judgment travels from one person to the next, and how accountability disappears in the process.
§2. The Stages the Existing Literature Shares
Collective alignment is not a newly discovered phenomenon. It has been described, in different vocabularies across fields, in a strikingly similar stage structure.
| Framework (field) | Stages |
|---|---|
| Smelser 1962 — value-added theory (collective behavior) | structural conduciveness → structural strain → generalized belief → precipitating factor → mobilization → social control |
| Blumer 1939 — circular reaction (crowds) | individual restlessness → milling → collective excitement → social contagion |
| Kindleberger–Minsky (financial crises) | displacement → boom → euphoria → distress → revulsion/panic |
| Bikhchandani, Hirshleifer & Welch 1992 / Banerjee 1992 — information cascades | sequential observation → cascade onset (threshold) → propagation → fragility/shattering |
| Granovetter 1978 — threshold models (riots, opinion) | threshold-0 instigators → threshold accumulation → chain ignition |
Read across these five models, a common arc recurs regardless of field. In plain terms, it traces this process. First, conditions accumulate — tension and uncertainty build while the individual has not yet established a basis for their own judgment (①). Then people observe one another — having no standard of their own, they take what others do as a cue (②). Something pulls the trigger — a shock, deadline, or loss arrives that makes watching no longer tenable (③). At that moment one person’s action moves the next, and it spreads fast — each joins in the same direction by looking at the one before (④). Finally it stops through external intervention or exhaustion — control arrives or the momentum dies, and the alignment ends or reverses (⑤).
| Common stage | Correspondence in each literature |
|---|---|
| ① Condition formation | conduciveness/strain / displacement / individual restlessness |
| ② Signal convergence | milling / sequential observation / generalized belief |
| ③ Trigger | precipitating factor / cascade onset, threshold / distress |
| ④ Transfer/propagation | social contagion / cascade propagation / panic, revulsion |
| ⑤ Termination | social control / cascade shattering / market reset |
▸ Methodological note. The five stages above are not a division newly devised by this article. They are a consolidation and alignment of the stage structure that recurs across the collective-behavior, crowd, financial-crisis, and information-cascade literatures (Smelser 1962, Blumer 1939, Kindleberger–Minsky, Bikhchandani et al. 1992, Granovetter 1978). This article’s contribution is not the creation of these stages, but the identification of which judgment state sits over each one (§3).
§3. The Judgment Mechanism at Each Stage
The established stage models describe what the group does (observe, mobilize, propagate, shatter). But the state of the individual’s judgment at each stage has not been separately named. Adding that layer yields the following.
| Common stage | Judgment mechanism in operation |
|---|---|
| ① Condition formation | Unmet conditions for judgment accumulate. The individual has not yet established a basis for their own judgment. |
| ② Signal convergence | Withholding of judgment-output arises — deferral (HOLD) from unmet conditions, or active yielding even with a sufficient own signal. Recognizing “I cannot judge yet” or “the group beats my signal,” the person observes others’ behavior to search for a standard. |
| ③ Trigger | The limit of sustained deferral is reached. Time pressure, the cost of inaction, or arousal exceeds the cost of holding. |
| ④ Transfer/propagation | Acceptance of an external standard (judgment delegation). Unable to hold the deferral, the individual takes the conclusion of another, the market, or a system into their own place of judgment, and that conclusion becomes the next person’s input — chaining onward. |
| ⑤ Termination | External intervention (social control) or exhaustion resets the conditions for judgment, or the accepted standard collapses and a reverse alignment begins. |
3-1. Why the Transfer Happens
The transfer is not weakness of will but the result of three conditions overlapping. First, at ① the individual never established a basis for their own judgment — there is no internal standard to lean on from the start. Second, at ③ the cost of holding rises — there is no time, inaction means loss, or arousal is too high. Third, right beside them is an already-made external output (the prior person’s action, the market price, the system’s result), and adopting it wholesale costs almost nothing.
When these three overlap, the individual chooses “borrow another’s conclusion” over “build my own judgment.” The moment the cost of holding exceeds the cost of accepting, with no internal standard in place, the transfer becomes the path of least resistance. So ④ is not a matter of will but a consequence of the cost structure.
3-2. Delegation Is Not of One Kind
The route into ④ — the mechanism that drives acceptance of an external standard — is not single. It divides into at least five types, and more than one can overlap in a single event.
| Type of delegation | What drives the acceptance | Representative domain / basis |
|---|---|---|
| ① Inference-driven | One rationally infers that others’ information beats one’s own signal and sets one’s signal aside. | information cascades (BHW 1992; Banerjee 1992) |
| ② Payoff-driven | Matching others’ action is optimal for one’s own payoff, so it is accepted. | bank runs / strategic complementarity (Diamond–Dybvig 1983) |
| ③ Conformity-driven | The mere fact that enough others have already acted crosses a threshold and pulls one in. | thresholds / social proof (Granovetter 1978; Asch 1956) |
| ④ Collapse-driven | Arousal or fear instantly breaks the deferral itself, copying the nearest action. | crowd panic (Helbing et al. 2000) |
| ⑤ Authority/system-driven | The output of an authority, expert, or automated system is taken as the standard. | expert reliance / automation (→ §4) |
These five are cut along one axis — “what drives the acceptance”: rational inference, payoff, numbers, arousal, authority. The list is not exhaustive; one could, for instance, add group identity (accepting because the group is one’s in-group) as a separate type. What matters is that different drivers reach the same outcome (④). The point of activation also differs by type — the inference- and conformity-driven types pass fully through the observation of ② before crossing the threshold at ③; the collapse-driven type nearly skips ② and fires immediately at ③; the authority/system-driven type, with an external standard always on standby, activates quickly anywhere from ② to ③. Which mode of withholding operates at ② also differs by type — the inference-driven type is ‘active yielding’ (the own signal is intact but the group is judged better), while the collapse-driven and unmet-condition cases are ‘HOLD’ (judgment does not form at all). The payoff- and conformity-driven types, though they look alike, are distinct — payoff-driven is equilibrium logic (matching others is optimal for one’s own payoff), while conformity-driven is the numbers themselves (the fact that enough have acted is social proof). As in a bank run, the two can operate together.
(The specific variables and measurement criteria underlying this classification are part of a proprietary analytic framework and are not disclosed in this document.)
3-3. One Case, End to End
We instantiate the abstract scaffold in a single domain and walk it through (a generalization with identifying information removed).
A team faces an unfamiliar kind of decision. ① No one yet has their own standard for judging this matter (unmet conditions). ② In the meeting, people read one another and stare at the score from the newly deployed automated recommendation system (deferral, signal search). ③ A deadline looms and pressure arrives — “if we don’t decide, we’re on the hook” (limit of sustained deferral). ④ One person says, “let’s go with the system’s score,” and the others fall in with “that sounds right” — the system output is accepted as the external standard, and that acceptance becomes the next person’s input (the chaining of authority/system-driven delegation). ⑤ When the outcome turns out badly, “the system saw it that way” is what ends up in the minutes (terminated with accountability transferred to the system).
At no node is there a record reading “I judged this, on this basis.” Judgment clearly traveled, yet the subject who performed it is fixed nowhere.
3-4. How to Observe It — Signatures
How does one distinguish, from the outside, “deferred then accepted an external standard” from “judged for oneself”? The following signatures serve as proxy indicators.
| Signature | When it is acceptance of an external standard |
|---|---|
| Post-hoc account | Asked the reason for the decision, the person answers with an external pointer rather than their own grounds — “everyone did,” “the system,” “the mood.” |
| Timing | The conclusion aligns immediately after an external signal arrives. There is no time interval for one’s own deliberation. |
| Record | No deferral rationale or unmet conditions are recorded anywhere. If present, self-judgment; if absent, suspect acceptance. |
| Reversibility | It reverses easily on a small contrary signal. A borrowed basis is weaker than one’s own. |
These signatures are a starting point for identification, not precise measurement. But they make the judgment-state layer observable — if all four are negative, the layer’s account does not apply to that case.
3-5. Why Accountability Is Deferred
Having accepted an external standard at ④, the person does not perceive themselves as the one who decided. Two layers overlap — feeling less responsibility (subjective) and being unholdable to account (objective) are different things, and in delegation both happen at once.
Subjective — a decline in felt responsibility. When others who could also act are visible, an individual’s sense of obligation declines. Diffusion of responsibility research has reported this repeatedly (Darley & Latané 1968). That research originally addressed inaction in emergencies, so its application here is an analogical extension (see §5). Still, the common mechanism — “the presence of others reduces one’s own share of responsibility” — is taken to apply to the active acceptance of an external standard as well. When an automated system is involved, the perception that “the system is in charge” amplifies this decline (moral buffer, Cummings 2004).
Objective — the impossibility of attribution. An outcome to which many each contributed a little has no single subject to hold responsible — the “problem of many hands” (Thompson 1980). Originally aimed at public organizations, the concept maps directly: a delegation chain produces the same structure — dispersed contribution → no single subject. Moreover, since the acceptance itself is unrecorded (the record signature of §3-4), there is not even a basis on which to locate that node afterward.
When subjective non-responsibility and objective non-attribution overlap, accountability is not something someone evaded — it is something whose place has disappeared.
3-6. Consequences — A Single Vicious Loop
Once this structure has run, the consequences do not stand apart; they close into a loop that feeds itself.
Missing record → failure to learn → repetition/accumulation → non-attribution → no recurrence-prevention → (back to condition formation)
Because the deferral is unrecorded (missing record), one cannot tell afterward at which node judgment was empty. Unable to tell, one cannot detect or correct the same failure (failure to learn). Uncorrected, the same conditions reproduce the same transfer each time, and mistaken acceptances pile up without a corrective signal (repetition/accumulation). For the accumulated outcome too, contribution is dispersed and there is no place to assign responsibility (non-attribution). With no one located, there is no place for a design that prevents recurrence (no recurrence-prevention). And so the conditions form again, and the same structure rises in the same place.
The key here is that what disappears is not the decision but “the record that judgment was deferred.” And the absence of that record makes attribution structurally difficult — because the node to point to does not exist as data, so that even where post-hoc reconstruction from timestamps and speaking order is possible, its cost rises sharply. Accountability dispersion proceeds together with this absence, and exactly to its extent.
A testable prediction. Here, the “deferral record” refers not to the post-hoc suspicion signature of §3-4 but to an independently existing audit field (a prior log of deferral rationale and unmet conditions) — separating the classifying variable from the predictor variable avoids tautology. If this layer has explanatory power, the following should hold — delegation chains in which deferral is unrecorded should, relative to chains in which it is recorded, show (a) a higher recurrence rate of the same failure, (b) a longer time to correction, and (c) a sharper reversal under external shock. Conversely, if the presence or absence of a deferral record is observed to be unrelated to recurrence rate, correction speed, and reversal magnitude, then this layer adds no explanatory power over the existing stage models. This is the falsification condition the article stakes.
§4. AI and Social Extension — Automation Accelerates the Transfer and Hides It
Among the five types in §3-2, the last — authority/system-driven delegation — expands explosively in the age of automation. As the system’s output becomes an external standard always on standby, it can step in immediately at the place where the individual should defer. The arrival of AI does not create a new phenomenon; it scales up one delegation route that already existed.
How it actually appears. When an algorithmic feed amplifies a particular claim, people accept it as their standard; automated trading fires in unison at a threshold; the output of an automated decision system in hiring, credit, or content moderation is accepted wholesale by the operator. And public discourse about the outcome often converges on “the algorithm did it,” “the AI judged.” That discourse is itself evidence that delegation occurred — the system is named as the judging subject, and the individual drops out of the chain. The very habit of attributing responsibility to the system is the after-trace of an accepted external standard (the post-hoc-account signature of §3-4).
Two mechanisms make the delegation easier and less visible. One is automation bias — people over-rely on system output and tend not to search for information that contradicts it (Skitka et al. 2000). The other is the moral buffer — the perception that “the system is in charge” weakens the individual’s sense of moral agency and responsibility (Cummings 2004). The first makes acceptance easy; the second intensifies the accountability dispersion of §3-5.
So the pattern of change accelerates. Automation does three things at once. It compresses the ②→④ transition (skipping the deferral interval), scales the chain (one system output becomes millions of inputs), and renders the transfer node invisible (state transitions leave no trace in the audit trail). Faster, larger, and unseen — the vicious loop of §3-6 turns faster and at greater scale.
Where these transformations overlap, what disappears in collective judgment in the age of automation is not the decision itself. What disappears is the record that judgment was deferred. If there is no record of the stage at which judgment was handed outward, there is no record of where accountability should be. (The structure of this missing record is treated separately in “What Should a Judgment Deferral Log Look Like?”)
§5. Limitations
This document does not address the following.
No claim of stage originality: The stages in §2 are a consolidation of existing literature, not a proposal of this article. The contribution is limited to the judgment-state layer in §3.
Nature of the interpretive layer: The judgment-state layer is an interpretive identification placed over existing mechanisms, not a unified formal model. The delegation types of §3-2 and the signatures of §3-4 are proxy indicators that require further operational definition and measurement.
Status of borrowed concepts: The diffusion of responsibility (Darley & Latané) and the problem of many hands (Thompson) in §3-5 come from the contexts of inaction and public organizations respectively; their application to judgment delegation is an analogical extension.
Status of the case and prediction: The case in §3-3 is illustrative, not empirical; the prediction in §3-6 is an as-yet-unverified hypothesis.
No prescription: It offers no methods to prevent or mitigate collective alignment. It describes conditions and structure only.
Generalized cases: It does not analyze specific events, institutions, or markets. All domain examples are generalizations retaining structure only.
§6. FAQ
Q1. Did you invent these five stages?
No. They consolidate the stage structure shared across the models of Smelser, Blumer, Kindleberger–Minsky, Bikhchandani et al., and Granovetter. What this article adds is not the stages but the judgment state and the types of delegation operating at each.
Q2. Then how does this differ from rational herd behavior like information cascades?
It does not differ — it does not compete. The information cascade is one type of delegation in §3-2 (the inference-driven type). The judgment-state layer does not replace it; it identifies, over that mechanism, the point at which the individual handed their judgment outward.
Q3. Is this a testable claim?
§3-4 offers observable signatures and §3-6 a falsifiable prediction (presence/absence of a deferral record ↔ recurrence rate, correction speed, reversal magnitude). If the signatures are all negative or the prediction fails, the layer has no explanatory power for that case.
Q4. Does collective judgment always resolve into stage ④?
No. Where an individual’s conditions for judgment are met so that no deferral arises (② is skipped), or where deferral can be sustained, no acceptance of an external standard forms. The stages are a conditional progression, not a deterministic sequence.
Further Reading:
- Smelser, N. J. (1962). Theory of Collective Behavior. Free Press.
- Blumer, H. (1939). Collective Behavior. In R. E. Park (Ed.), An Outline of the Principles of Sociology. Barnes & Noble.
- Asch, S. E. (1956). Studies of Independence and Conformity. Psychological Monographs, 70(9), 1–70.
- Granovetter, M. (1978). Threshold Models of Collective Behavior. American Journal of Sociology, 83(6), 1420–1443.
- Bikhchandani, S., Hirshleifer, D., & Welch, I. (1992). A Theory of Fads, Fashion, Custom, and Cultural Change as Informational Cascades. Journal of Political Economy, 100(5), 992–1026.
- Banerjee, A. V. (1992). A Simple Model of Herd Behavior. Quarterly Journal of Economics, 107(3), 797–817.
- Diamond, D. W., & Dybvig, P. H. (1983). Bank Runs, Deposit Insurance, and Liquidity. Journal of Political Economy, 91(3), 401–419.
- Helbing, D., Farkas, I., & Vicsek, T. (2000). Simulating Dynamical Features of Escape Panic. Nature, 407, 487–490.
- Kindleberger, C. P., & Aliber, R. Z. (2005). Manias, Panics, and Crashes (5th ed.). Wiley. [Minsky–Kindleberger stage model]
- Darley, J. M., & Latané, B. (1968). Bystander Intervention in Emergencies: Diffusion of Responsibility. Journal of Personality and Social Psychology, 8(4), 377–383.
- Thompson, D. F. (1980). Moral Responsibility of Public Officials: The Problem of Many Hands. American Political Science Review, 74(4), 905–916.
- Skitka, L. J., Mosier, K. L., & Burdick, M. (2000). Accountability and Automation Bias. International Journal of Human-Computer Studies, 52(4), 701–717.
- Cummings, M. L. (2004). Automation Bias in Intelligent Time Critical Decision Support Systems. AIAA Intelligent Systems Technical Conference. [moral buffer]
Disclaimer. The cases in this article are generalizations of observed structural patterns and bear no relation to any specific individual, institution, or event. All identifying information has been anonymized.
This document does not provide conclusions or recommendations. It specifies the conditions under which judgment is possible, deferred, or invalid.
Jung, Y. (2026). When Judgment Travels — How Accountability Disperses in Groups (GRL-T1-013-EN). Gungri Research Lab.
Glossary: https://gungriresearch.com/glossary/
License: CC BY-NC-ND 4.0
© 2026 Gungri Research Lab. Published under CC BY-NC-ND 4.0.
궁리연구소 | GUNGRI RESEARCH LAB 
댓글 남기기