# The Algorithmic Revolution: From the Quantification of War to the Struggle over Attention, Behaviour and Power

A soldier films a trench from the first-person perspective; a teenager repeats a gesture in a public square for an audience assembled by a social platform; a political movement circulates a meme that can be understood without watching a speech; a worker is evaluated by software that neither party can readily explain; an intelligence officer receives ranked information from a system whose errors may be difficult to trace. These scenes are historically distinct. Their common feature is that information about human activity passes through technical systems that classify it, predict its relevance or consequences, and redistribute it according to operational objectives.

The claim that algorithms "control society" is simultaneously provocative and imprecise. An algorithm is a specified procedure; it is not necessarily an artificial-intelligence model, a commercial recommendation system or a political actor. People continue to choose, resist, distrust, create and organise independently of platforms. Yet the distribution of choices is affected by who controls their informational environment, the criteria of selection, access to resources and opportunities, and the visibility granted to particular conduct. An algorithm does not need to issue an order to become part of social power. It may alter which people or messages are noticed, which applications are considered, which actions become worth filming, and which representations of a conflict travel farthest.

Three transformations account for the historical significance of this shift. First, the ambition to make complex processes calculable moved from statistics and administrative organisation into digital infrastructure. Second, behavioural data became a continuous input for ranking, pricing, monitoring and recommending, with effects that feed into the next round of observed data. Third, a growing share of social life acquired a prospective audience: a person may act while anticipating a video, a metric, a reaction or a possible recommendation. These transformations can improve coordination, widen access to knowledge, uncover abuses and reduce costs; they can also concentrate decision-making, reward distorted proxies, shape perceptions and redistribute risks without an equally visible system of accountability.

Human agency increasingly operates within institutional environments organised by mathematical criteria of selection. Commercial platforms, employers, governments and military organisations pursue objectives distinct from those of the people affected; their indicators influence visibility, access to resources and the conditions of decision-making. Experiments, observational studies, institutional documents and philosophical theories provide different kinds of evidence. An effect on emotional expression does not amount to determining a vote, just as the circulation of conflict imagery does not establish that the imagery caused the conflict.

## 1. A problem before the platforms: what can administration actually measure?

Long before a user opened a personalised feed, states used inventories, population registers, taxes, probabilities and classifications to administer territories and allocate resources. The industrial age added engineering measurements, production planning, accounting, insurance risk calculations and the systematic evaluation of labour. These methods were not a single unified ideology. They became powerful because formal measurements let organisations coordinate actions across distances, compare alternatives and document performance. They were also selective: a register records categories recognised by administrators; a productivity measure records the product that its designers decide to count.

The Second World War accelerated the development of operational research, including mathematical assessment of logistics, radar, allocation and combat decisions. Postwar computing and information theory expanded the ability to represent problems formally. Cybernetics studied communication and control in machines and organisms; statistical decision theory and operations management offered ways to make uncertainty tractable. There were important differences between Wiener, Shannon, Simon and subsequent engineers. A rigorous historical genealogy cannot convert these distinct disciplines into a direct line of descent leading inevitably to TikTok. Their shared contribution was to make prediction and feedback organisationally practical.

Quantification creates a temptation when a measure is easier to improve than the underlying objective. A company can raise the number of customer interactions without improving satisfaction. A security agency can count incidents detected without establishing whether a territory is safer. A school can optimise examination scores without measuring broader understanding. This is a general institutional vulnerability: people adapt to the measures by which they are assessed, and measurements may lose their meaning precisely because they have become targets. Whether that happens depends on governance, independent evidence and how the metric is designed.

A present-day feed produces a distinctive variant. Its measures are not only retrospective accounts. Views, skips, likes, follows and completion rates enter feedback loops that affect future exposure. People can then adapt what they produce to attract those measures. The shift is from **describing behaviour to participating in the conditions under which behaviour is rewarded**. It does not erase the earlier history of bureaucracy; it extends it into everyday interaction.

## 2. Robert McNamara and *The Fog of War*: when a metric obscures a war

Errol Morris's documentary *The Fog of War* (2003) is a useful opening to the history of the algorithmic society because it examines how a highly capable administrator confronted the limits of technical rationality in military and political decision-making. Its subject, Robert S. McNamara, had worked with quantitative management and later directed the US Department of Defense during the Kennedy and Johnson administrations. The film revisits the Second World War, the Cuban Missile Crisis and Vietnam through interviews and archival material. It is not a history of social-media algorithms, nor does it demonstrate that contemporary online systems descended directly from the Pentagon. Its relevance is the problem of treating a quantitative representation as though it were the problem itself. [1]

A study by the US research organisation CNA identifies three particularly significant systems associated with McNamara-era operations assessment: enemy body counts, the Hamlet Evaluation System and Southeast Asia Analysis Reports. The designers sought comparable indicators of counterinsurgency progress. But the data could not adequately reflect the political legitimacy, local allegiances, historical context and changing conditions that shaped the conflict. The CNA account concludes that these assessments proved unreliable as measures of progress. That is a historical finding about a particular institutional system, not a rejection of statistics in military affairs. [2]

The distinction between *measurement of an action* and *measurement of a desired outcome* is fundamental. Counting casualties can describe one military variable without answering whether a government is gaining durable legitimacy. Similarly, counting how long a video remains on screen may help predict a platform interaction without establishing whether it makes viewers better informed. A highly accurate prediction of clicking or watching is not the same as an ethically or socially satisfactory result. Nor does a failure in one dimension imply that all models are useless.

A second parallel is organisational. McNamara's systems were embedded in hierarchies that chose which information reached commanders and political leaders. Contemporary feeds and decision systems are embedded in firms and institutions that choose objectives, constraints, training data and reporting formats. In both cases, the central political question is where responsibility sits when a model appears more definite than the underlying uncertainty. The analogy is bounded: one system assesses an armed conflict within a state hierarchy; the other distributes information within commercial and social networks. Their historical causes, consequences and degrees of coercion differ radically.

Management dashboards offer an appearance of completeness rarely matched by the complexity of political and social processes. Standardised categories can be reproduced precisely while the interpretation of the whole remains mistaken. In Vietnam, apparently precise military indicators coexisted with an inadequate understanding of the conflict's political character.

## 3. Cybernetics, statistics and the migration of decision systems

The twentieth-century development of computers allowed measured events to be stored, compared and processed at scales not feasible for manual administrations. Banks used credit models, insurers calculated risks, factories scheduled production and public agencies classified populations. Search engines later made ranking central to finding information; online advertising made targeting and measurement central to the financing of internet services. Recommender systems combined these streams with models of individual interaction.

The transition was not simply from analogue repression to digital freedom or from freedom to universal surveillance. Earlier social orders already shaped information through editors, schools, churches, political parties, family networks and state broadcasting. What changed was the granularity of measurement, the speed of feedback, the capacity to update an individual's informational environment and the commercial value of knowing how that person responds. The shift has produced possibilities for emancipation and surveillance, distributed unequally across institutions and social groups.

The difference between a rule-based program and a machine-learning recommender matters. A deterministic rule may display the latest article in reverse chronological order; a learned model estimates the probability of a viewer engaging with several candidate articles. Both are algorithms, but they generate different incentives and explanations. The chronological rule is transparent at one level, although publication frequency still affects visibility. A recommendation system may discover interests more effectively, although the reasons for a particular ranking can be difficult to inspect. Neither design guarantees viewpoint diversity, accuracy, welfare or neutrality.

This development also transformed the politics of classification. When a machine uses an employment history to recommend an applicant, it inherits both the useful information and possible historic inequalities in the data. When a platform models past watch time, it may reproduce habits that a user would not deliberately choose as long-run goals. Statistical correlation becomes an input into allocation, while allocation changes the next observations. The system must therefore be evaluated as a dynamic institution, not merely a static prediction.

## 4. How recommendation systems organise the visible world

A common architecture for large recommender systems separates *candidate generation* from *ranking*. A 2016 technical paper by Google engineers Paul Covington, Jay Adams and Emre Sargin described a deep-learning YouTube recommender with these two stages. The first estimates which videos may interest the user from a very large catalogue; the second ranks the smaller candidate set. The disclosure is a documented historical architecture, not a claim that every current YouTube decision still follows exactly the 2016 design. [3]

TikTok's user documentation provides a complementary description of today's recommendation signals. The company identifies interactions such as watched, skipped, liked, shared or commented content; information about the video, including sound and hashtags; and user or device context. It says interactions and watching time are generally more influential than certain other factors for many users. This is the platform's public account, not an independent audit of its weights or business objectives. The public feed remains personalised, and an individual user's feed cannot be reduced to a fixed formula known outside the company. [4]

The economic loop begins when a user produces signals, but the system's operating goals matter. Prediction may be optimised for watch time, expected interest, relevance, safety, freshness, customer experience or a combination of metrics. The exact optimisation scheme is proprietary and changes over time. It is therefore inaccurate to say that all social platforms optimise only anger or addiction. It is equally inaccurate to pretend that the selection of measurable goals has no social effects.

The source of power is partly a selection binding constraint. An immense catalogue is practically inaccessible without intermediaries. Search rankings, subscriptions, recommendations and social ties each filter information. A designer controls not only the algorithm but also permitted content, enforcement rules, advertising policy, creator monetisation and the experimental process used to test changes. The relevant institution consists of the entire system of choices, not merely its mathematical model.

```image
src: /assets/images/reports/2026-11-03-algorithmic-revolution-attention-power-social-conflict/recommendation-feedback.svg
origin: original
kind: editorial-illustration
width: 1600
height: 900
alt: Four linked stages represented as information modules, showing a conceptual selection and feedback process.
caption: Original conceptual illustration of the passage from measured signals to predictions, exposure and subsequent response. The drawing does not reproduce a proprietary platform system.
credit: LOGV / Marginal Thinking
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
```

```flow
Observed actions → Behavioural signals → Candidate selection and ranking → Exposure to selected information → Human response → New observed actions
Content-production incentives → Choices of format and intensity → Material available for ranking → Audience reaction → Revised creator incentives
```

The relationships in the diagram depend on decisions by users, creators and platform operators. Exposure is not agreement, and a recorded reaction does not establish lasting preference change. An audience selected because it already agrees may produce results resembling those of an audience persuaded by a message, although the processes have different causal implications.

## 5. The economics of attention: what is actually being sold?

Attention is limited by waking hours, competing obligations and human capacity. Digital content, by contrast, can be copied and distributed at low marginal cost. Platforms capable of organising a large audience may sell targeted advertising, charge subscriptions, intermediate purchases or earn from associated services. Their business models differ, but the distribution of attention is economically consequential across them.

Additional time on a platform produces revenue only through particular commercial mechanisms. Advertising load, ad prices, advertiser demand, retention, moderation costs, infrastructure and subscriptions alter the financial result. Material capable of raising immediate engagement may also reduce advertisers' trust or increase regulatory exposure. Profit therefore does not necessarily rise in proportion to viewing time.

The distribution of value is uneven. A platform owns its ranking infrastructure and the rules of participation. A creator may invest time and equipment without controlling distribution. Advertisers purchase probabilistic access to audiences; consumers receive entertainment, information and services while surrendering time and sometimes personal data. An emerging creator can suddenly gain a national audience, but the same arrangement can produce dependence on unpredictable changes in the feed. These are forms of market power that can exist even when users pay no monetary subscription fee.

Platforms operate markets involving several interdependent groups: users, creators, advertisers, merchants, developers and payment providers. Economic value depends on each group's ability to reach the others and on the cost of moving to another service. Creators unable to transfer their audiences face weaker bargaining power when distribution rules change. Competition analysis based solely on subscription prices overlooks much of this dependence.

Algorithmic selection also creates an externality problem. The owner of a feed may measure clicks and sales directly while the effects on public debate, sleep or institutional trust appear elsewhere. These consequences are difficult to price and sometimes difficult to attribute. Regulation must address that evidence gap without assuming that every regrettable social event was caused by a profitable algorithm.

## 6. From long-form video to short-form cultural grammar

TikTok, Instagram Reels, YouTube Shorts and comparable services made vertical, mobile-native and rapidly editable video formats central to digital culture. The transformation is not merely that videos became shorter. Music, captions, templates, remixes, repeated gestures, camera movements and immediate access to the next clip create a recognisable grammar. A segment can circulate without its original surrounding context, and the same visual joke or heroic montage can be reused across countries and political communities.

Short-form production has lowered barriers to creative expression. A small business can demonstrate a service without a production studio; an artist can reach a dispersed audience; a student can learn a technique in an accessible sequence; a civilian can document an abuse. Yet highly shareable formats also encourage compression. A twenty-second excerpt can communicate outrage or admiration efficiently while excluding qualifying evidence. That is a property of the communicative environment, not proof that every short video misleads.

The history of technology adoption is crucial. Bolsonaro's 2018 digital mobilisation cannot be described as an election driven by Instagram Reels or YouTube Shorts, which became prominent later. WhatsApp, Facebook, YouTube, images and shareable video clips were important then. The 2022 ecology was different: formats produced on TikTok or Kwai could be downloaded and redistributed inside WhatsApp groups. The same trend can therefore combine algorithmic discovery at one site with voluntary forwarding through an interpersonal messaging network elsewhere. [11][14]

Different layers of circulation have different mechanisms. A ranking system determines exposure within a feed; a group administrator may shape participation; a peer chooses to forward a message; a person with social authority may interpret it; a broadcaster may amplify the controversy. Calling the entire chain 'the algorithm' conceals the actors and causal steps that must be evaluated separately.

## 7. Dopamine, reward prediction and the scientific limits of an easy metaphor

In popular discussion, dopamine is often presented as a chemical pleasure button that short videos manipulate until the brain is damaged. This description is misleading. Neuroscientific work associated with Wolfram Schultz and colleagues examines dopamine neurons as signals related to rewards and deviations from expected rewards. Those findings are part of a larger field involving learning, motivation, cognition, context and multiple neurotransmitter systems. They cannot be translated into the claim that a particular genre of video depletes a person's dopamine or causes permanent harm. [9]

The behavioural mechanisms are nevertheless serious research questions. Novelty, variable rewards, uncertain timing, social comparison and frictionless access can encourage repeated checking. An individual's behaviour depends on stress, social ties, mood, available alternatives and pre-existing tendencies. Some users may develop problematic patterns of use; others engage intentionally and experience benefits. To establish a causal effect, researchers must separate exposure from self-selection, account for baseline difficulties and measure outcomes over an appropriate period.

A 2025 systematic review and meta-analysis in *Psychological Bulletin* synthesised 71 studies with 98,299 participants. It reported associations between greater short-form video use and poorer cognitive indicators, with correlations of approximately -0.38 for attention and -0.41 for inhibitory control; mental-health associations were generally weaker and heterogeneous. These are effect-size correlations across a literature with substantial methodological variation, not probabilities of mental illness, reductions in IQ or estimates of permanent damage. Many included studies were observational; reverse causality remains possible. [8]

A plausible policy concern is not the existence of dopamine but the design of an environment that makes stopping difficult, especially for vulnerable users. A service can offer user controls while implementing default experiences that remain effortless to continue. The European Commission's February 2026 *preliminary* findings against TikTok addressed infinite scroll, autoplay, notifications and personalised recommendation in relation to possible failures to assess and mitigate risks under the Digital Services Act. The findings were provisional; they do not establish a final legal judgment or a universal medical outcome. [20]

Total screen time captures only one part of exposure. Content, time of day, purpose of use, offline relationships, prior conditions and available alternatives influence outcomes for sleep, attention, work and social life. The same duration of use may reflect deliberate learning, communication with relatives, creative work or repeated consumption without a defined purpose.

## 8. Music, fast cuts and cinematic intensity

A striking feature of short-form culture is the use of a rapid sequence of editing cues to give ordinary actions the expressive force of cinema: heavy bass, accelerated electronic music, phonk, funk or trap, slow motion, first-person camera, abrupt cuts, synchronized subtitles and sharply contrasting images. These choices can heighten arousal, direct attention or give a scene a particular emotional interpretation. Their effects depend on the viewer, prior knowledge, context and social meaning.

A montage of a person entering a room may portray confidence; the same gesture with different music can become comedy or menace. The edit changes the narrative frame without changing the underlying event. This is especially consequential when footage is detached from its date or location. The soundtrack can produce an impression of triumph even when the event had severe human consequences. In political conflicts, cinematic editing may make the moment of confrontation seem like the entire history of a dispute.

Musical genres, heavy beats and rapid cuts can shape audiovisual interpretation but do not independently determine aggression or physical courage. Their expressive effect emerges from their combination with images, cultural conventions and viewers' previous experience. Intensified emotion in an edit may alter the perception of an event without changing why that event occurred.

The same techniques may perform socially beneficial functions: make complex scientific ideas accessible, allow witnesses to communicate danger, facilitate artistic experimentation, or offer humour and connection. Their dual use is precisely why banning an aesthetic style would be a poor substitute for studying the underlying behaviours and incentives.

## 9. The camera as a second audience: acting while anticipating visibility

In Erving Goffman's sociology, social life involves performances and the management of impressions before others. Cameras did not invent that human tendency. Networked platforms, however, make the audience potentially persistent, distant and difficult to predict. An individual can act before people physically present and simultaneously imagine a much larger audience that may view an edited recording later.

This changes the role of the camera. At one level it documents an action; at another it enables a person to address a later audience; at a third it may become part of the reason the action is performed. Distinguishing these situations matters. A protester may film police violence for accountability. A performer may stage a harmless gesture to entertain friends. An offender may seek notoriety. The existence of a recording alone does not establish which motivation applies.

The mechanisms are social as well as technical. Comments, view counts and imitation offer imperfect information about group esteem. A creator may anticipate future recognition and adapt a performance; viewers may reproduce a gesture because it signals membership in a community. These possibilities are best described as *audience-conditioned action* rather than a claim that a platform removes agency. Agency remains active precisely because participants interpret and use the system.

```image
src: /assets/images/reports/2026-11-03-algorithmic-revolution-attention-power-social-conflict/camera-performance.svg
origin: original
kind: editorial-illustration
width: 1600
height: 900
alt: Conceptual drawing of a person, a digital recording interface and separate audience signals.
caption: A networked camera can document conduct and also create an anticipated audience before whom conduct is performed. No real individuals or events are represented.
credit: LOGV / Marginal Thinking
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
```

An experiment by Lauren Sherman and colleagues presented 32 adolescents with a simulated Instagram-like interface while using brain imaging. Participants were more inclined to endorse photographs shown as having many likes; the responses implicated social influence and neural networks related to reward and attention. The design supports a bounded mechanism of peer endorsement; it does not establish that social-media metrics cause violent behaviour or that the same effects hold across all ages. [10]

Recognition among peers, economic payment and the obligation to maintain a public reputation create different incentives. Friends' approval produces local prestige; monetisation may convert attention into income; sustaining a persona can increase pressure to publish material resembling earlier posts. Wider exposure changes the relationships between these forces without making every performance commercially motivated.

## 10. From reputation to a score: the symbolic economy of “aura”

The expression *aura farming* transfers the logic of accumulation from video games into digital social life. In gaming, farming means repeatedly performing actions to acquire resources or points; on social platforms, *aura* loosely describes a quality recognised by an audience—presence, confidence, charisma, style or the ability to create a memorable scene. Together, the words make prestige speakable as a kind of score. The value does not reside in the gesture alone but in its interpretation by people who recognise a shared cultural convention.

Competition for reputation long predates the internet. Public displays of courage, self-presentation and the pursuit of recognition are familiar features of social life. Platform infrastructures make small performances reproducible, editable and distributable to audiences far beyond the performer's immediate group. A gesture performed in front of three friends may later be judged by thousands of strangers. Recognition no longer depends exclusively on physical presence, although relationships, humour and group conventions still influence it.

The vocabulary of aura also reflects the convergence of cinematic conventions, game culture and ordinary life. Walking towards a camera, standing still, directing a deliberate look or assuming a fictional character's posture can acquire a strikingly different meaning with music, framing and editing. Commenters may celebrate the scene, ridicule it or treat it as deliberate parody. A recording can produce opposing interpretations; distribution does not equal collective approval.

During 2026, examples of this cultural grammar appeared in documented in-person meetings and contests in Latin America, including Brazil. The movement from a digital trend to public squares illustrates how a shared idiom can coordinate people without elaborate institutional organisation. Online audiences supply models; physical gatherings provide a place to enact them; subsequent recordings return the event to the distribution cycle. The result is an exchange between public space and digital reputation, not simply the replacement of face-to-face interaction by screens. [24]

The economic incentive is often indirect. For most participants, the reward is symbolic—laughter, prestige among friends, recognition or a sense of belonging. For creators with substantial audiences, circulation can also translate into followers, commercial opportunities and advertising income. Visible metrics quantify a judgment whose social meaning remains ambiguous. A view count measures reach and distribution, not an objective quantity of charisma, yet it can shape what future participants regard as successful performance.

The camera changes the temporal relation between an act and its evaluation. Performance occurs in the present but may be oriented towards reactions expected later. Its circulation depends on the audience's familiarity with the trend, the timing of publication, editing, initial social networks and distribution systems. Repeated imitations reveal the importance of learned conventions: apparently spontaneous performances often draw from an established repertoire.

These incentives do not produce a single social consequence. Documented aura events have included play, dance, parody and nonviolent community interaction. Elsewhere, the expectation of notoriety may combine with peer rivalry and willingness to take risks. No automatic progression connects the two. The relationship between audience and hazardous conduct depends on context, motives and the rewards attached to particular actions. A similarly cinematic montage can depict a dance, an athletic contest, a demonstration or an actual confrontation without the underlying events sharing a cause.

**Platform metrics and cultural codes have become everyday categories of reputation.** Individuals need not calculate their own behaviour numerically to inhabit an environment where recognition is constantly counted, compared and displayed. Aura farming is one expression of that arrangement: people still confer prestige, while technical systems organise much of the circulation that gives prestige its scale.

## 11. Adolescence, identity and the apprenticeship problem

Adolescence is a period in which peer relationships, identity and recognition often have particular salience, but young people are not a homogeneous psychological category. The social meaning of a platform can differ sharply between a teenager seeking friendship, an adult working as a creator, a student seeking information and a person experiencing social isolation. Even within the same demographic group, prior mental health, family context, school environment and available offline relationships may change outcomes.

An environment that offers continuous comparison can intensify feelings of inadequacy in some contexts, while providing minority communities or geographically isolated users with social support in others. Research quality must therefore distinguish problematic use from ordinary participation and separate self-reported symptoms from independently measured health outcomes. Aggregate correlations in the short-video literature should not be treated as clinical diagnoses of individuals. [8]

Skills develop through practice, mistakes, feedback and increasing responsibility. Recommendations that reproduce past preferences can make navigation more comfortable while reducing encounters with unfamiliar material, depending on system design. Search, tutoring and recommender tools can also widen access to new knowledge. Educational outcomes depend on the substance of the activity, not merely on the interface through which it is encountered.

The social dimension is as important as cognition. A person may become expert at producing a recognisable online persona while having limited opportunities to gain professional recognition, stable income or influence in formal institutions. This can be understood as a mismatch between digital prestige and material power. The mismatch may create frustration, entrepreneurship, civic participation or disengagement: no single outcome follows.

## 12. Facebook experiments: influence has been demonstrated, total control has not

The 2012 study *A 61-million-person experiment in social influence and political mobilisation* used randomised messages displayed on Facebook around the 2010 US congressional elections. The researchers found effects on political self-expression, information seeking and real-world voting; transmission through close social ties was especially significant. The experimental result demonstrates that the design of a platform message can affect specified behaviours. It does not show that a recommendation algorithm can freely choose a person's political preference or determine an election result. [5]

In a 2014 *PNAS* experiment involving 689,003 Facebook users, researchers adjusted the prevalence of positive or negative posts displayed in feeds and studied changes in subsequent emotional expression. The reported effect was evidence of emotional contagion through an online environment. The study also prompted controversy about consent and research ethics, and its effect sizes and outcome definition limit any extrapolation to clinical mental health or mass political manipulation. [6]

The experimental findings also have significant limits. A 2023 *Science* study experimentally assigned consenting Facebook and Instagram users to reverse-chronological feeds rather than the default algorithmic ranking during the 2020 US election campaign. Exposure and activity changed substantially, but the study did not find significant changes in central measures of political polarisation and attitudes over its three-month observation period. This does not prove that feed design has no longer-run effects; it disproves the assumption that every substantial change in exposure must rapidly transform political opinion. [7]

| Study | Intervention or observed exposure | Result supported by design | What is not established |
| --- | --- | --- | --- |
| Bond et al., 2012 | Randomised Facebook mobilisation messages among 61 million users | Specific political participation and network-transmission effects | Control over political identity or election outcomes |
| Kramer et al., 2014 | Variation in emotional exposure for 689,003 users | Changes in aggregate written emotional expression | Clinical disorder or general behavioural obedience |
| Chronological-feed experiment, 2023 | Feed ordering changed during 2020 US election | Significant platform-experience changes without significant short-run changes in key attitudes | Absence of all long-run or population-specific effects |
| Nguyen et al., 2025 | Meta-analysis of observational and other short-video studies | Associations with attention, inhibitory control and well-being | Universal cognitive damage or a single confirmed causal direction |

Social-media design can alter attention and specific behaviours, while the magnitude, duration, direction and social consequences vary with context. Existing experiments identify bounded effects without establishing unrestricted platform control over political preferences.

## 13. The Brazilian political case: an existing social conflict meets a new communications infrastructure

The rise of Jair Bolsonaro from a long-serving federal deputy to president of Brazil in 2018 is one of the most consequential cases for studying the interaction between political grievances and networked communication. Its causes cannot be reduced to memes. Brazil entered the election after the 2013 demonstrations, economic crisis, corruption scandals, polarised debates over institutions, the impeachment of Dilma Rousseff and contestation around political representation. A study based on the 2018 Brazilian Electoral Panel Study identifies anti-Workers' Party sentiment, conservative cultural preferences, strict law-and-order views and economic liberalism among the factors associated with Bolsonaro support. The technological environment provided mechanisms of organisation and expression; it did not manufacture all of these preferences. [13]

Bolsonaro's communication was also distinctive. Supportive communities circulated ironic images, heroic representations, excerpts from interviews, denunciations of adversaries and symbolic shorthand capable of communicating political affiliation without extended argument. A short phrase or a picture could serve simultaneously as humour, persuasion and a badge of belonging. Viktor Chagas's research on memes from the early months of the administration analyses their nationalist, ideological and rhetorical characteristics. That corpus concerns a particular period and sample, not the totality of 2018 campaign material, but it documents the political functions of images circulated among supporters. [12]

The 2018 architecture was multiplatform rather than a uniquely TikTok-driven election. In a peer-reviewed study, Victor Piaia and Marcelo Alves examined **194,710 messages in 21 pro-Bolsonaro WhatsApp groups**, collected between 5 September and 11 November 2018. A relatively small number of participants accounted for a large share of posts; links to YouTube and Facebook appeared prominently; and some organised campaign activities were identifiable. The sample describes monitored groups, not every supporter, every Brazilian WhatsApp group or the probability that an individual voted because of a message. The study's greatest contribution is to show how interpersonal forwarding and public platforms operated as an interconnected communication system. [11]

There were different reasons for participation. Some users shared content as political conviction; others sought community, opposed a particular party or followed trusted family networks. An influencer might see a market for attention while a group participant regarded a forward as civic duty. Political mobilisation thus involved both genuine supporters and strategically organised activity; the two are not mutually exclusive. Evidence of some coordinated accounts does not prove that all supporters were manipulated, and evidence of genuine convictions does not exclude organised amplification.

Bolsonaro's national ascent took place amid anti-Workers' Party sentiment, institutional conflict, economic crisis and distrust of established parties. Digital networks changed the circulation of political language and symbols by lowering organisational costs and enabling persistent sharing among dispersed groups. Their particular contribution to the election result remains distinct from preferences that existed before the campaign: observational data describe networks and associations, not the outcome of an alternative election without platforms.

## 14. From 2018 to 2022: cross-platform video and the persistence of an identity

Between the elections of 2018 and 2022, short-form video became a more prominent part of the Brazilian communication ecology. Clips from podcasts, speeches, interviews and street encounters travelled across platforms; sound and subtitles reframed fragments for new audiences. A single recording could serve as primary political material, commentary, parody or evidence depending on who cut, titled or distributed it. The intermediary between an original event and its final audience was frequently a sequence of creators, recommenders and group members.

An investigation by Agência Pública using data from the Eleições Sem Fake project at the Federal University of Minas Gerais examined highly interactive videos circulating in pro-Bolsonaro WhatsApp groups during July and August 2022. It reported that 41.3% of the sampled high-interaction videos originated on TikTok or Kwai; of 133 such videos examined, 108 originated on TikTok, 24 on Kwai and at least one carried marks from both applications. These are **sample-dependent findings**, not percentages of all election videos, all Brazilian WhatsApp traffic or causal effects on voting. The original reporting also identified misleading claims in some items. [14]

```chart
type: bar
title: Origin platform marks among 133 short videos analysed in Brazilian pro-Bolsonaro WhatsApp groups, July–August 2022
unit: videos in the investigated 133-video subset, not all election content
TikTok | 108
Kwai | 24
Both marks | 1
```

Source: Agência Pública / UFMG Eleições Sem Fake investigation, 12 September 2022 [14]. Platform identity here describes the editing/distribution provenance of a selected subset. Do not extrapolate the bars into vote shares or national reach.

The result reveals an important division of labour. A video can be discovered through ranking on one platform and then distributed through WhatsApp by people who know one another. That second stage may draw authority from interpersonal trust rather than from a feed's hidden formula. A shared political meme can also travel in the opposite direction, from a messaging group to a public creator's video. The resulting communication network does not have one central decision-maker.

Political movements opposed to Bolsonaro also used memes, videos, influencers and messaging networks. Their content, organisational models and audiences differed, but no claim that digital politics is exclusive to a particular ideological position is supported by the underlying technology. A comparative study would need examine matched samples from multiple camps rather than simply accumulating the most striking examples from one.

The long-run political consequence is the survival of identity outside formal election periods. A set of phrases, symbols and shared histories can persist across changes in parties and public office. This persistence may mobilise lawful participation, protest or other activities. Understanding which outcomes occurred requires attention to actual messages and institutions, not the presumption that a digital identity is inherently undemocratic or inherently emancipatory.

## 15. January 2023 in Brasília: communication, coordination and the boundary of attribution

The attacks on the institutions of Brazil's three branches of government on 8 January 2023 make it necessary to examine the passage from online claims and coordination to a physical event. The relevant mechanisms include circulation of narratives contesting the electoral result, mobilisation logistics, transport, social ties, calls for action and the decisions of particular participants. These should not be collapsed into a claim that a recommendation algorithm directly caused the attacks.

InternetLab's report on Telegram material associated with the period collected **592,644 publications between 1 and 8 January 2023**, including text, images, audio and video. The researchers analysed repeated messages, important imagery and patterns of communications and organisation. The corpus is substantial but has clear boundaries: it is not the universe of private conversations, not an individual-level map of motives, and not proof of criminal participation by anyone who was present in a monitored group. [15]

The sequence from digital circulation to physical action involves distinct stages: encountering a statement, endorsing it, forwarding it, coordinating an activity and participating in an event. Each transition has its own conditions. Sharing false information is not equivalent to committing an assault, and message volume alone cannot yield a conversion rate into demonstrators. Conversation records, journeys, testimony and judicial findings document different aspects of organisation but cannot substitute for one another.

Private moderation and public responsibility occupy different institutional domains. Platforms can remove posts, limit their reach or close accounts; state institutions remain responsible for public security, lawful enforcement and individual guarantees. Measures against incitement may limit harm, while opaque moderation can also produce errors and overreach. Transparency, appeals and judicial review matter particularly when digital circulation intersects with political conflict.

## 16. France in 2023: when events are organised, filmed and remediated

The French disturbances following the killing of Nahel Merzouk by a police officer in Nanterre on 27 June 2023 provide a distinct case. Their proximate trigger, history of police-community relations, spatial inequalities and political context cannot be replaced by a theory of TikTok or 'dopamine'. Digital circulation became part of the **form, timing, visibility and coordination** of some subsequent unrest without replacing the social and political conditions that brought it about.

A 2024 French Senate inquiry examined the unrest and the role of platforms. It described different roles for public networks, interpersonal messaging and location-aware functions. According to figures from a sample examined in a joint inspectorate report cited by the Senate, Snapchat was mentioned with a frequency of **36%**, compared with X at 5%, TikTok 4%, Instagram 2%, Telegram 2%, WhatsApp 1% and Facebook 0.6%. These are rates of mention in a particular sample. They are not a distribution that sums to 100%, platform market shares or the proportions of all rioters using each service. The relative prominence of Snapchat is particularly important: the most relevant logistics platform in a specific event need not be the platform whose content is most visible to distant observers. [16]

```chart
type: bar
title: Platforms mentioned in the French inspectorate sample on 2023 unrest, cited by Senate in 2024
unit: % of references in the sampled material; not an exhaustive platform share
Snapchat | 36
X (Twitter) | 5
TikTok | 4
Instagram | 2
Telegram | 2
WhatsApp | 1
Facebook | 0.6
```

Source: French Senate, report on June 2023 unrest, 2024 [16]. The denominator is the inquiry's specific sample; unmentioned channels, overlaps and non-response prevent interpreting these values as a complete distribution.

The Senate described an initial phase of circulation of video and calls to gather, followed by more varied logistical uses of messaging services, including features associated with Snapchat. Platform representatives reported different timelines of activity. The institution also considered imitation, visibility and competition for online attention, while maintaining that social circumstances and urban conditions mattered. This is much more specific than the assertion that all unrest is 'cinematic' because young people enjoy edits. [16]

Recordings of conflict can preserve evidence, expose abuse, mobilise support or display courage. These motivations may coexist in varying degrees and cannot be inferred from cinematic framing alone. When anticipated recognition becomes part of the decision to act, an additional incentive enters a physical confrontation. Evidence of that mechanism rests on the sequence of events, participants' communications and social context; a dramatic video uploaded after unrest cannot establish its origin.

The French and Brazilian cases illuminate different pathways. One concerns long-term political identity followed by an institutional assault; the other concerns a sudden protest and unrest triggered by a police killing. Both involved networked communication, but their histories and mechanisms remain distinct. Treating them as equivalent expressions of an algorithmic tendency would ignore the very social structures that an adequate explanation must recover.

## 17. International comparison: the same devices, different social mechanisms

Brazil, France, Ukraine and the United States contain different institutional configurations despite widespread use of similar digital devices. In Brazil, the circulation of political symbols and group messages intersects with electoral competition; in France, communication networks were used amid unrest triggered by a police killing; in Ukraine, first-person video participates in war documentation, propaganda and fundraising; in the United States, experiments have measured effects of specific feed interventions. The same technologies provide common capacities while historical and social conditions shape their uses.

```map
title: Different transmission mechanisms across researched settings
Brazil | Election-oriented memes, interpersonal forwarding, political identity and separately documented mobilisation | WhatsApp studies from 2018 and 2022; Telegram research on January 2023
France | Police killing followed by unrest, public video circulation and logistical messaging | Senate 2024 inquiry; platform-use observations have sample limitations
Ukraine | War documentation, first-person testimony, advocacy and platform-native entertainment forms | Scholarly analysis of 97 Ukrainian creator videos; no proven causal recruitment rate
United States | Randomised platform interventions and measured political or emotional responses | Facebook experiments from 2012 and 2014; 2023 chronological-feed counterevidence
```

The events documented in these four countries are not stages of a single process and do not represent uniform responses across their populations. Differences between events and platform functions limit direct cross-national comparison. The same function—video sharing—can support electoral persuasion, eyewitness documentation, public accountability, fundraising or unlawful coordination, depending on the situation.

Documents establish that a particular message or recording existed. Platform data describe interactions within access limitations; interviews reveal stated motivations but may reflect selective memory; experiments estimate effects under defined interventions; administrative and judicial records document outcomes and accountability. Attributing large-scale social change to algorithmic mechanisms requires convergence across these forms of evidence rather than reliance on the emotional intensity of footage.

## 18. Ukraine: witnessing, propaganda and the first-person perspective of war

The Russian invasion of Ukraine in February 2022 created an environment in which civilians, soldiers, journalists, institutions and propagandists produced and circulated images of conflict. Mobile recording and social platforms offered opportunities to show previously inaccessible aspects of daily survival, evacuation and combat. Those same images could be extracted from context, staged for communication, manipulated or used to construct selective narratives. Authentication and dating are central because a visceral first-person recording may still be misleading about place, timing or responsibility.

Tom Divon and Moa Eriksson Krutrök's 2025 study of **97 TikTok videos from Ukrainian creators** identifies three styles: point-of-view testimony, memes used to communicate wartime experiences, and playful political entertainment associated with engagement and fundraising. The authors examine how platform-native forms mediate conflict visibility and can support civic engagement. Their sample concerns selected creators and videos, not all wartime communication, all combatants or recruitment outcomes. [17]

Research on Ukrainian soldiers' TikTok self-representation likewise shows how soldiers perform national and gender identities through audiovisual conventions. The proximity of the mobile camera can make distant combat seem psychologically immediate, yet a clip rarely conveys the full logistical, ethical or human cost of the underlying events. A montage featuring music and a confident camera perspective may make a soldier appear heroic without measuring the viewer's subsequent willingness to enlist. [18]

Military enlistment is shaped by factors rarely visible in combat montages: patriotism, legal obligations, family connections, material conditions, official campaigns and security circumstances. Admiring comments under POV videos record reactions to an image, not actual decisions to enlist. The transition from audiovisual exposure to military participation remains undetermined without comparable recruits' viewing histories, individual accounts and administrative records.

War communication also exposes a moral asymmetry. Some images may document atrocities or help civilians receive assistance; others can glorify harm or create operational security risks. The same distribution system that supports testimony can spread propaganda and false imagery. Treating every military clip as dangerous would undermine documentation; treating every spectacular clip as authentic would reward manipulation.

## 19. From a video of war to algorithm-assisted decisions in war

Images of trenches are only one interface between algorithms and conflict. Militaries also employ computational systems for logistics, sensor fusion, image classification, threat assessment, simulation and decision support. Their consequences differ in scale and responsibility from a social feed. A system that prioritises possible targets or signals may influence life-and-death decisions even when a human formally retains the final command.

Military systems move through stages of development and application with different evidentiary status: capability *claimed* by manufacturers, *demonstrated* in bounded tests, *deployed* in operations and *verified as effective at scale*. A promotional drone video does not establish operational autonomy, correct target discrimination or stable performance under adverse conditions. Dated documents, independent assessments and observed operating conditions determine the significance of a technical claim.

The second risk is compressed decision time. Systems that deliver more alerts and faster recommendations may improve defence while increasing pressure on operators to defer to automated classifications. Imperfect training data, adversarial deception, uncertain attribution and unclear accountability can produce severe errors. Speed may be useful precisely when there is time to verify; reducing time without improving verification may worsen decisions.

This returns to McNamara without collapsing history. The Vietnam-era problem involved interpreting quantitative indicators of a broad political-military process. Current automated military systems may act within technical chains whose inputs and intermediate decisions are opaque to participants. The shared institutional vulnerability is overconfidence in a representation that does not incorporate the whole context. The technologies and legal stakes differ, but both require meaningful human responsibility and independent auditing.

## 20. Algorithms beyond the feed: work, income, credit and public administration

The most consequential algorithm in an individual's life may never recommend a video. A credit-scoring system can affect access to finance; recruitment software can filter applicants; automated scheduling can alter shifts; fraud detection can interrupt a payment; navigation software can determine which neighbourhood receives traffic. Such systems allocate opportunities, time and resources. Their impact on society may be more immediate than their cultural visibility suggests.

The Organisation for Economic Co-operation and Development reported in December 2025 an employer survey on algorithmic management across several advanced economies. It found at least one such tool reported by **90% of managers surveyed in the United States**, compared with an average of **79% in the four surveyed European countries** and **40% in Japan**. Those values describe firms represented by a particular employer survey and the adoption of at least one management tool. They are not the proportion of workers fully managed by autonomous AI, nor a measure of how many decisions were automated without human supervision. Respondents reported perceived benefits and concerns about accountability, transparency and worker health. [19]

The economic mechanism differs from advertising. A manager may seek more consistent allocation of tasks, fraud control or productivity. Yet a worker can face a rating whose data sources and correction routes are not visible. When employers share external assessment tools, errors or biases can spread across multiple hiring decisions. When every worker is evaluated by a narrow measure, people may focus on that measure even at the expense of service quality or safety.

The governing questions are practical: can the affected person find out that automation was used, challenge an error, obtain a human review, inspect relevant evidence and obtain a remedy? Is the model accurate across the populations it serves? Are performance gains verified or merely believed by managers? These rights and tests are often more informative than a general promise of ethical AI.

The employment and public administration cases also reveal a limit to the word *control*. A ranking system need not control someone's beliefs to influence their access to an interview, a bank account or a public service. The mechanisms of power can be distributive, procedural and material rather than psychological.

## 21. Peter Thiel, René Girard and the political theology of technological power

Peter Thiel's 2025–2026 discussions of the Antichrist belong in the intellectual history of contemporary technological power, not as evidence that he secretly coordinates the recommendations on social platforms. Reporters documented private lectures in San Francisco in 2025 and Rome in March 2026 addressing apocalyptic theology, the danger of a global political authority and fears about attempts to prevent catastrophic technologies. Contemporary reporting also records strong criticism of his interpretation by religious thinkers. These positions are attributed accounts of his statements, not independently verified prophecies or an agreed philosophical framework. [22][23]

René Girard's theory of mimetic desire and rivalry influenced Peter Thiel's thinking. In that account, social models participate in the formation of desire and competitive relationships. Repeated online behaviours, prestige competition and visible approval offer a possible setting for applying the interpretation. Girard's influence on a technology investor does not mean that technical recommendation systems were engineered according to his philosophy.

Thiel's concerns also expose a governance paradox. A rhetoric of existential danger can justify greater central control; a rhetoric of limitless innovation can justify strong private discretion. Neither position resolves who has legitimate authority, who bears risk, and what forms of evidence should constrain a decision. These questions intersect with accelerationist debates about technological speed and state power, but do not reduce to an opposition between progress and fear.

Guy Debord described social relationships mediated by representation; Erving Goffman examined self-presentation and everyday performance. Their work offers ways to interpret circulating images and anticipated audiences, but does not establish the causal effect of a recording on political unrest or military conflict. That effect varies according to institutions, participants and events that can actually be observed.

## 22. What regulators can govern without pretending to govern the mind

Regulatory choices range from transparency and independent risk assessment to rights of appeal, limits on certain profiling practices, age-appropriate design and specific obligations for very large platforms. These tools address different harms. A requirement to disclose why a person saw a recommendation does not give outsiders the ability to reproduce all ranking decisions. A restriction on autoplay may reduce frictionless consumption while affecting accessibility or user convenience. A moderation intervention against explicit incitement must be distinguished from a restriction on lawful dissent.

The European Commission's February 2026 preliminary findings about TikTok's interface referred to infinite scroll, autoplay, push notifications and high personalisation under the Digital Services Act. The document is a dated regulatory allegation with detailed risk concerns; TikTok retained the right to respond, and the announcement expressly did not prejudge the final outcome. Conflating preliminary findings with a final sanction would distort the evidence. [20]

In Brazil, Law 15,211 of 2025, the *ECA Digital*, became effective on **17 March 2026**, alongside institutional changes affecting the national data protection authority. The framework establishes obligations concerning the protection of minors in digital environments. Its legal operation, enforcement capacity and measurable effects must be assessed over time, rather than assumed from the existence of the statute. [21]

Concentration of regulatory power creates its own risks. Large corporations may be better able to finance complex compliance, while small entrants may face barriers. Governments can invoke safety to impose excessive surveillance or suppress opposition. Effective governance therefore requires independent courts, proportionate obligations, published methodologies, participation by affected groups and reviewable decisions. Auditing a system should not become a licence to monitor every private conversation.

Informational autonomy, competition, employment rights, child protection, public accountability and operational security require different instruments. Disclosure of classification criteria enables scrutiny but does not replace an individual's right to challenge a decision; risk auditing is not a mandate for indiscriminate surveillance. Legitimate governance depends on the ability to review consequential decisions by private firms and the state.

## 23. The future configuration of algorithmic power

The next stage of algorithmic society cannot be inferred from a straight-line projection of watch time or model capability. Three interpretations deserve attention.

**Diffused assistance and pluralism.** Lower-cost models and transparent interfaces may expand access to education, public information, creative tools and productive capabilities. Users can choose different ranking methods, creators can switch distribution channels and workers can challenge automated decisions. Supporting evidence would include broad entry, affordable alternatives, measurable learning gains, lower switching costs and improvements in due process. The scenario weakens if infrastructure concentration rises and affected people cannot escape opaque allocation.

**Concentrated selection and dependency.** A small number of infrastructure providers may control important gateways to information, work and economic participation. Platform rules then become quasi-institutional conditions for participation. Evidence would include durable concentration, declining interoperability, unstable creator income, inability to challenge consequential rankings and observed reductions in independent discovery. The spread of open standards, new competitors and effective rights of appeal would reduce that dependence.

**Fragmented and securitised information environments.** States may intensify control over data, platforms and AI systems because of military, security and sovereignty concerns. Different populations could experience separate informational environments under distinct regulatory and technical regimes. Evidence would include expanding cross-border restrictions, state-specific model access, content controls and verification burdens. International interoperability, verifiable governance and independent information exchange would constrain fragmentation.

These paths can coexist: education may become more accessible while employment algorithms become more opaque; content moderation may improve safety while state surveillance grows elsewhere. No numerical probability is assigned because the relevant causal model and uncertainty ranges are not established. The future of algorithmic power is a question of institutional choices, not an inevitable consequence of faster computation.

## 24. The extent and limits of algorithmic power

Algorithmic power becomes consequential when social outcomes depend on decisions about measurement, classification and allocation, particularly when affected people cannot inspect or challenge them. A demonstrated change in informational exposure may occur without a change in political preference; a credit score can limit access to resources without changing beliefs. The kind of power involved depends on the variable actually altered and the institution controlling the procedure.

| Mechanism | Appropriate evidence | Strongest competing explanation | Test that would weaken the claim |
| --- | --- | --- | --- |
| Recommendation changes political attitudes | Randomised exposure with comparable groups and repeated attitude measures | Users choose content because they already hold the view | No difference after verified exposure changes over adequate periods |
| Short-form media harm attention | Longitudinal or controlled study with reliable cognitive outcomes | People with existing concentration difficulties seek more short videos | Apparent effects disappear after baseline and confounder control |
| Public filming encourages risky performance | Dated actions, testimony on anticipated audience and comparison events | Existing peer competition or local conflict drives the act | Equivalent rates when audience incentives are absent |
| Memes organise political communities | Traceable circulation, identity surveys and coordinated-action evidence | Prior political identities explain the sharing | Network changes without matching identity or mobilisation changes |
| War videos cause recruitment | Recruit-level exposure and motives linked to administrative recruitment data | Patriotism, legal mobilisation and material incentives | Recruitment unchanged among appropriately matched exposed groups |
| Automated work decisions reproduce inequality | Audited selection outcomes, error correction and group-specific outcomes | Structural inequalities existing before algorithm adoption | Equalisation of outcomes after controlled adoption and correction |

The historical continuity reaches back to the dilemma illuminated by McNamara: a system can become extraordinarily good at measuring what it has been instructed to count while being much less capable of representing the full human problem. The new difference is that metrics increasingly operate inside feedback systems that expose, reward and allocate social opportunities before the next decision is made.

Algorithmic infrastructures have not abolished politics, desire, class, institutions or culture. They provide environments through which all of these are increasingly expressed and contested. Their significance lies in the conjunction of calculation with human expectations: people know that other people, machines and markets may see their actions, and some adjust accordingly. The crucial public question is therefore **who may define the criteria by which social life becomes visible, valuable and consequential—and who may refuse or contest those criteria**.

## References

[1] Sony Pictures Classics, *The Fog of War*, documentary by Errol Morris (2003): https://www.sonypictures.com/movies/thefogofwar

[2] Emily Mushen and Jonathan Schroden, CNA, *Are We Winning? A Brief History of Military Operations Assessment* (2014): https://www.cna.org/analyses/2014/are-we-winning

[3] Paul Covington, Jay Adams and Emre Sargin, *Deep Neural Networks for YouTube Recommendations* (2016): https://research.google/pubs/deep-neural-networks-for-youtube-recommendations/

[4] TikTok Help Center, *How TikTok recommends content* (consulted 2 November 2026): https://support.tiktok.com/pt_BR/using-tiktok/exploring-videos/how-tiktok-recommends-content

[5] Bond et al., *A 61-million-person experiment in social influence and political mobilisation*, *Nature* (2012): https://www.nature.com/articles/nature11421

[6] Kramer, Guillory and Hancock, *Experimental evidence of massive-scale emotional contagion through social networks*, *PNAS* (2014): https://pmc.ncbi.nlm.nih.gov/articles/PMC4066473/

[7] Guess et al., *How do social media feed algorithms affect attitudes and behavior in an election campaign?*, *Science* (2023): https://doi.org/10.1126/science.abp9364

[8] Nguyen et al., *Feeds, feelings, and focus: A systematic review and meta-analysis examining the cognitive and mental health correlates of short-form video use*, *Psychological Bulletin* (2025): https://pubmed.ncbi.nlm.nih.gov/41231585/

[9] Schultz, Dayan and Montague, *A neural substrate of prediction and reward*, *Science* (1997): https://pubmed.ncbi.nlm.nih.gov/9054347/

[10] Sherman et al., *The Power of the Like in Adolescence*, *Psychological Science* (2016): https://www.psychologicalscience.org/journals/psychological-science/0956797616645673/

[11] Victor Piaia and Marcelo Alves, *Abrindo a caixa preta: análise exploratória da rede bolsonarista no WhatsApp*, *Intercom* (2020): https://www.scielo.br/j/interc/a/JB3zHccN7KnHJXTwsRj8WjF/?format=html

[12] Viktor Chagas, *Meu malvado favorito: os memes bolsonaristas de WhatsApp e os acontecimentos políticos no Brasil*, *Estudos Históricos* (2021): https://www.scielo.br/j/eh/a/vXzQKJb4KJY4LV7ZXXGSzvH/?format=html&ilang=pt_BR

[13] *The Bolsonaro Voter: Issue Positions and Vote Choice in the 2018 Brazilian Presidential Elections*, *Latin American Politics and Society* (2020): https://www.cambridge.org/core/journals/latin-american-politics-and-society/article/bolsonaro-voter-issue-positions-and-vote-choice-in-the-2018-brazilian-presidential-elections/98B69A9646D6CCB18253339E5DAE5FF5

[14] Laura Scofield and Nathallia Fonseca, Agência Pública / UFMG Eleições Sem Fake, *TikTok e Kwai levam desinformação sobre urnas e Forças Armadas ao WhatsApp* (12 September 2022): https://apublica.org/sentinela/2022/09/tiktok-e-kwai-levam-desinformacao-sobre-urnas-e-forcas-armadas-ao-whatsapp/

[15] InternetLab, *Desdobramentos dos ecossistemas de (des)informação política ao longo dos períodos pré e pós eleitoral* (2023): https://internetlab.org.br/pt/pesquisa/desdobramentos-dos-ecossistemas-de-desinformacao-politica-ao-longo-dos-periodos-pre-e-pos-eleitoral/

[16] Sénat français, *Émeutes de juin 2023 : comprendre, évaluer, réagir*, report (2024): https://www.senat.fr/rap/r23-521/r23-521_mono.html

[17] Tom Divon and Moa Eriksson Krutrök, *The rise of war influencers: Creators, platforms, and the visibility of conflict zones* (2025): https://doi.org/10.1177/29768624251325721

[18] *War, social media and gender: an analysis of the use of TikTok by Ukrainian soldiers* (2025): https://www.tandfonline.com/doi/full/10.1080/23337486.2025.2472100

[19] OECD, *How widespread is algorithmic management in workplaces?* (19 December 2025): https://www.oecd.org/en/publications/how-widespread-is-algorithmic-management-in-workplaces_cda7a114-en.html

[20] European Commission, *Commission preliminarily finds TikTok's addictive design in breach of the Digital Services Act* (6 February 2026): https://digital-strategy.ec.europa.eu/en/news/commission-preliminarily-finds-tiktoks-addictive-design-breach-digital-services-act

[21] Brazilian Ministry of Justice, *ECA Digital* legislation and implementation (2026): https://www.gov.br/mj/pt-br/assuntos/sua-protecao/sedigi/eca-digital/eca-digital-1

[22] Reuters, *Thiel's secretive Rome conference draws Church attention* (15 March 2026): https://www.reuters.com/technology/thiels-secretive-rome-conference-draws-church-attention-2026-03-15/

[23] The Washington Post, *Inside billionaire Peter Thiel's private lectures: Warnings of the Antichrist and U.S. destruction* (10 October 2025): https://www.washingtonpost.com/technology/2025/10/10/peter-thiel-antichrist-lectures-leaked/

[24] Associated Press, *Aura farming trend inspired by video games is sweeping Latin America* (2026): https://apnews.com/article/5b82900616f7c4dc2a34566322ccc070