We have built a planetary communications system in which an argument can remain profitable long after it has become socially ruinous. Every return offers another advertising opportunity, another behavioural trace and another reason to reproduce the conditions that brought us back. Peace, humanity, love and freedom can remain in circulation while the relationships that give them substance are weakened. The problem reaches beneath the vocabulary into the organisation of attention itself. The decisive issue is what the machinery selects for, and what its continued success costs everyone else.
Civilisation does not persist by standing still. Food systems, infrastructure and living bodies require continuing work and energy; trust and cooperation need time, attention and repair. Prigogine’s work on systems far from equilibrium describes organisation sustained through flows of energy and matter (Prigogine, 1977). Entropy makes maintenance unavoidable; it does not make extinction inevitable. The economic danger begins when the costs of sustaining cooperation can be displaced. A decision becomes cheap for one organisation because its consequences are paid by someone else, somewhere else, or later. The underlying dynamics of persistence are explored further in The Logic that Lives.
In a study published in 2025, Milli et al. found that Twitter’s engagement ranking amplified hostility towards political opponents, even though users did not prefer the political posts selected by that algorithm. Piccardi et al. (2025), experimenting independently with 1,256 X users, showed that increasing or reducing exposure to hostile and antidemocratic content shifted feelings towards the opposing party in corresponding directions. The measured changes were modest but causally attributable to the experimental intervention. Engagement records an action; it does not establish a considered wish for the world that action helps reproduce.
That distinction matters because the rewards alter what people subsequently produce. Brady et al. (2021) found that positive feedback for outrage increased its later expression. Opposing political positions become mutually sustaining: each supplies the threat through which the other recruits, organises and persists. Even resistance can feed the circuit. Amnesty International (2022) recounts how Facebook supported Myanmar’s ‘flower speech’ anti-hate stickers, only for activists to report that posting them beneath hateful material increased its visibility. An appeal to interrupt hatred became another signal to distribute it. The system absorbed the correction as an additional reason to continue. This recursive problem is examined in Disinformation Interdiction.
Restrictions on political expression are also documented. Human Rights Watch (2023) identified 1,049 instances of peaceful pro-Palestinian content being removed or unduly restricted on Facebook and Instagram among the cases it examined. A 2026 analysis by digital rights organisation 7amleh examined 3,520 reported cases from 2021–2025, identifying recurring restrictions, reduced distribution and opaque enforcement. These investigations document particular patterns of content restriction; they do not establish their overall prevalence or demonstrate a general commercial policy against peace. Their significance extends beyond missing posts. Freedom to speak depends in practice on being able to reach others, assemble a community and sustain relationships capable of changing anything.
The US Federal Trade Commission (2024) documented extensive collection and use of personal information by major social media and video streaming services, including for targeted advertising and automated systems. That establishes commercial extraction, without establishing a strategy of manufacturing conflict specifically to produce training material. Amnesty’s Myanmar investigation also examined internal warnings about harmful recommendation systems and repeated failures to respond adequately (Amnesty International, 2022). Responsibility does not disappear into an optimisation function. People decide what counts as success, whose losses remain outside the calculation and how much damage is tolerated before the model changes. A consequence can be unintended without the decision to tolerate it being innocent. The wider organisational problem is examined in Institutional Self-Preservation.
Alternatives already exist. A six-month experiment involving 9,386 desktop users, reported in a 2026 preprint, found small reductions in partisan hostility under alternative ranking systems. Engagement fell slightly on Facebook and Reddit and rose on X, while the study found no demonstrated improvement in wellbeing (Stray et al., 2026). These results suggest that reducing hostility need not entail a substantial loss of engagement, although the findings do not establish how such systems would perform across all users, platforms or longer periods. Governments matter, but concentrating authority at planetary scale cannot by itself correct incentives that reward self-preservation at the expense of shared survival. Change means making cooperation worth sustaining, bringing damage into the calculation and keeping power open to correction. We cannot preserve the whole by making its disintegration more rewarding than its maintenance. A bunker merely puts a reinforced door on the same dependencies.
Key takeaways
- Conflict can be profitable. Its commercial rewards can persist long after its social costs become destructive.
- Engagement is not endorsement. What attracts attention is not necessarily what people want.
- Opposition can reproduce what it opposes. Even attempts to interrupt hostility can reinforce its circulation.
- Peace requires different incentives. Cooperation must become more rewarding to sustain than conflict.
References
7amleh (2026) ‘New research reveals systematic suppression of Palestinian digital content across Meta platforms’, 9 June. Available at: 7amleh.
Amnesty International (2022) The social atrocity: Meta and the right to remedy for the Rohingya. ASA 16/5933/2022. Available at: Amnesty International.
Brady, W.J., McLoughlin, K., Doan, T.N. and Crockett, M.J. (2021) ‘How social learning amplifies moral outrage expression in online social networks’, Science Advances, 7(33), eabe5641. doi:10.1126/sciadv.abe5641.
Federal Trade Commission (2024) A look behind the screens: Examining the data practices of social media and video streaming services. Washington, DC: FTC. Available at: FTC report.
Human Rights Watch (2023) Meta’s broken promises: Systemic censorship of Palestine content on Instagram and Facebook, 21 December. Available at: Human Rights Watch.
Milli, S., Carroll, M., Wang, Y., Pandey, S., Zhao, S. and Dragan, A.D. (2025) ‘Engagement, user satisfaction, and the amplification of divisive content on social media’, PNAS Nexus, 4(3), pgaf062. doi:10.1093/pnasnexus/pgaf062.
Piccardi, T., Saveski, M., Jia, C., Hancock, J., Tsai, J.L. and Bernstein, M.S. (2025) ‘Reranking partisan animosity in algorithmic social media feeds alters affective polarization’, Science, 390(6776), eadu5584. doi:10.1126/science.adu5584.
Prigogine, I. (1977) ‘Time, structure and fluctuations’. Nobel Lecture, 8 December. Available at: Nobel Prize lecture (PDF).
Stray, J. et al. (2026) ‘The prosocial ranking challenge: Reducing polarization on social media without sacrificing engagement’, arXiv [Preprint], submitted 20 March. doi:10.48550/arXiv.2603.19626.
