The persistence of harm at scale produces a familiar conviction: that recognition must eventually force correction. If enough people see, if enough outrage accumulates, if exposure becomes unavoidable, then intervention is expected to follow. This conviction is high-adoption and rhetorically convenient. Structurally, it is unreliable. Recognition does not contain, within itself, a mechanism of conversion.
A conversion mechanism is valid only if it specifies all of the following:
The node at which cost lands,
The enforcement instrument that applies it,
The incentive gradient altered by enforcement, and
Why replication cannot outpace the intervention.
Absent any of these, recognition remains a detection event without consequence.
Collapse occurs only when a specific node is both causally necessary for continued circulation at scale and subject to durable, enforceable cost that alters amplification and replication economics.
The following do not constitute attribution closure and are always insufficient on their own: visibility, condemnation, post-distribution takedown, account bans, policy statements, transparency reports, advisory boards, and awareness campaigns.
Recognition is an informational event. It increases awareness. It does not, by default, impose cost. Conversion occurs only when recognition is coupled to enforceable consequence. That coupling is not automatic. It depends on additional conditions that are frequently absent in large-scale digital systems: attributable responsibility, jurisdictional authority, liability that cannot be displaced, and intervention capacity that precedes escalation rather than follows it.
In systems where these conditions are weak or fragmented, recognition accumulates without resolution. The signal intensifies, but the system does not cross a threshold that would force structural response. Instead, recognition becomes a stable input. It is absorbed, processed, and redistributed as part of normal operation.
This distinction, between recognition and conversion, is central. Many analyses collapse the two, assuming that awareness carries an implicit force. In practice, awareness is inert unless it is translated into obligation. Without translation, recognition functions as noise or fuel depending on system design. In engagement-driven environments, it reliably functions as fuel.
This explains why repeated exposure does not exhaust the circulation of harm. Exposure increases saturation. Saturation increases habituation. Habituation reduces the probability that any given instance will trigger decisive response. The system adapts not by correcting itself, but by normalizing intensity. Recognition becomes background variance.
Condemnation follows the same trajectory. Moral judgment does not interrupt circulation unless it alters routing conditions. In systems that privilege amplitude, condemnation and endorsement are processed as equivalent engagement signals. Both increase interaction density. Both extend lifecycle. Both reinforce visibility.
Moral polarity does not alter this routing logic. It alters subjective experience, not systemic behavior.
This is the point at which many readers experience resistance. It feels counterintuitive to suggest that condemnation can stabilize what it opposes. The resistance arises from a category error: confusing moral opposition with structural opposition. Moral opposition expresses values. Structural opposition imposes cost. Only the latter forces reconfiguration.
When condemnation remains decoupled from enforceable consequence, it becomes compatible with persistence. It can even become necessary to it. Outrage signals salience. Salience feeds ranking systems. Ranking systems determine visibility. Visibility maintains circulation. The system does not need outrage to agree with harm. It needs outrage to register that harm matters in ways that sustain attention.
At this stage, the system begins to externalize resistance as labor. Moderators absorb trauma while filtering content. Activists expend energy amplifying alarms. Investigators chase attribution across fragmented infrastructures. Journalists document recurrence. None of this labor lands as internal cost to the system that carries the signal. Instead, it dissipates externally. The system remains operational precisely because resistance is metabolized elsewhere.
This externalization produces a characteristic exhaustion pattern. Individuals cycle out. Institutions issue statements. Platforms adjust policies at the margins. The underlying circulation remains intact. Burnout is treated as an individual failure of resilience rather than a systemic output. This misidentification is functional. It preserves the boundary across which cost is displaced.
Recognition also fails because it is temporally mismatched to intervention. Digital signals propagate at speeds that legal, institutional, and infrastructural systems cannot match. By the time recognition stabilizes into consensus, the signal has already mutated, migrated, or been replaced. Reactive governance is always late. The system does not need to evade intervention entirely. It needs only to outrun it.
Attribution compounds this delay. Large-scale harm rarely resolves to a single actor. Responsibility is distributed across producers, platforms, intermediaries, and audiences. Each node is causally involved but individually insufficient. This diffusion is not accidental. It is a stabilizing property. When no node is both causally necessary and structurally accountable, recognition has nowhere to land. Enforcement stalls. Delay becomes indefinite.
This is the conversion failure in its simplest form: recognition increases without triggering attribution closure. Without attribution closure, cost cannot be internalized. Without internalized cost, the system has no reason to reconfigure. Recognition becomes a pressure that the system can absorb rather than a force that compels change.
For some there comes a retreat to psychological explanations. “Perhaps people do not care enough?” “Perhaps empathy collapses under scale?” “Perhaps moral progress was an illusion?” These conclusions feel plausible because they describe the subjective experience of saturation. Structurally, however, they add nothing. They do not identify an intervention surface. They do not specify a mechanism by which recognition could convert into consequence under current conditions.
The failure is not located in feeling. It is located in design. Systems that allow harm to circulate without forcing attribution closure will treat recognition as input rather than interruption. Systems that externalize the cost of opposition will stabilize around persistence even when condemnation is widespread.
Recognition alone does not resolve harm. Where amplification parameters remain unchanged, recognition increases circulation without producing conversion. It can normalize it. It can exhaust those who respond to it. Without a mechanism that converts recognition into enforceable consequence, exposure remains compatible with escalation.
Desire is not moral. It does not tend toward the good, the just, or the humane, nor does it orient itself toward repair or reconciliation. Desire is used here as a shorthand for observable circulation dynamics: repetition, connection, and intensification of behaviors that persist independently of ethical judgment or normative intent. [34]. In this sense, desire is not an interior psychological state but an operational driver that organizes behavior at scale.
This distinction matters because most contemporary responses to cruelty rely on an incorrect premise: that cruelty represents a deviation from an otherwise stable moral order. Under conditions of contemporary digital mediation, this premise fails. Cruelty can emerge as a selected outcome within specific configurations of desire and constraint. When social, technical, and affective conditions align, cruelty is not an exception to desire but one of its reproducible modes of operation [35].
Deleuze and Guattari’s critique of desire-as-lack is instructive here. Harm does not persist because something is missing (for example empathy, education, or awareness) nor because subjects have failed to internalize norms. Desire does not wait to be corrected. It produces.
Under conditions of alienation, anonymity, and deterritorialized circulation, behavioral patterns do not collapse: they redistribute along available affordances [36].
Digital platforms provide these pathways. They do not introduce cruelty as content, but they remove friction from its circulation. They abstract consequence from action, decouple visibility from responsibility, and compress repetition into continuous feedback loops. Within such environments, desire does not require belief, ideology, or justification. It requires only continuity: another upload, another reaction, another confirmation that an effect has been registered by the system [37].
Cruelty, in this configuration, is not organized around hostility toward the victim. It can function as a behavior that produces high-amplitude reaction under conditions of low attribution and asymmetric cost. Acts are retained insofar as they register disturbance within the system. Attention, particularly intense or hostile attention, serves as confirmation of efficacy: proof that an action has produced an observable disturbance within the system [38].
This dynamic explains why moral condemnation frequently fails as an intervention within engagement-optimized systems. Moral address presumes a subject oriented toward ethical recognition and capable of reflexive recalibration. Platform-mediated desire does not operate at that level. When behavior is organized around circulation rather than meaning, condemnation functions as an accelerant. Outrage increases visibility. Visibility increases repetition. Repetition sustains continuity [39].
What appears externally as transgression is often internally compliant with system incentives. Platforms optimized for engagement without semantic discrimination register intensity, not intent. As a result, the system cannot distinguish between protest and promotion. Cruelty circulates whether it is condemned or celebrated, provided it produces interaction [40].
In this context, cruelty is no longer anchored to ideology. It does not require narrative coherence, doctrinal belief, or justificatory frameworks. It persists as a format: repeatable, legible, and interoperable within platform grammars. Desire flows not toward belief, but toward amplification [41].
This also clarifies the structural misalignment of empathy-based appeals. Empathy presupposes relational proximity, temporal pause, and interpretive depth. Platform architectures collapse distance, accelerate time, and privilege quantity over reflection. Within such systems, empathy is not suppressed by malice: it is bypassed by design [42].
Cruelty is not usefully modeled as an agent-level deviation (psychological abnormality, moral failure). Those framings may describe individual cases, but they do not explain persistence, scale, or reproducibility. Cruelty is not explained by psychological abnormality or moral failure at the system level: persistence follows from routing, reinforcement, and attribution diffusion. Cruelty must be understood as a systemically available outcome within architectures that reward shock, repetition, and affective disruption. The relevant analytical question is not why some individuals desire cruelty, but why specific systems so reliably select for it under defined conditions [43].
This reframing does not absolve perpetrators. It relocates responsibility. Individual acts occur within environments that differentially enable, amplify, or constrain behavior. When systems repeatedly select for cruelty, moral outrage alone cannot alter outcomes. Only structural intervention can [44].
The implication is unavoidable: cruelty is not external to contemporary digital systems. It is one of their outputs. Until desire is analyzed as machinic rather than moral, responses will remain reactive, symbolic, and ineffective [45].
Information Failure Modes in Networked Violence
Elijah Morgan
© 2026 Elijah Morgan
This work is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
License: https://creativecommons.org/licenses/by/4.0/
DOI: https://doi.org/10.5281/zenodo.19599231
Version 1.1
Published 2026-04-15