Was den Menschen zum Menschen macht

AI-Algorithmic Arrangement

5 Min. Lesezeit

🇩🇪 Deutsche Version: KI-algorithmische Anordnung

The AI-algorithmic arrangement is the personalized selection and ordering of content by recommender systems, ranking algorithms, and personalization engines. What the person gets to see, in what order, with what frequency, is the result of a behavioral prediction about herself. It is a subform of AI-arranged oblivion of personhood and the technically most effective member of the family.

What structurally happens

The person experiences an order — her own feed, her own search-results list, her own product recommendation — that she has not chosen. The architecture of this order remains hidden from her. What she sees is the result of a function whose inputs are behavioral traces of herself and of statistically similar persons, and whose output is a prediction of maximized dwell time, click probability, conversion rate.

Shoshana Zuboff (The Age of Surveillance Capitalism, PublicAffairs 2019) has named the economic logic precisely: the commodity is not the user, but the prediction about him. And the more manipulable the behavior, the more valuable the prediction. From this arises a structural incentive asymmetry in favor of manipulation: whoever sells predictions has a financial interest in keeping behavior steerable.

Foucault: pastoral power in the machine

In Sécurité, territoire, population (lectures 1977/78 at the Collège de France), Michel Foucault analyzed Christian pastoral power: the shepherd knows each sheep individually and leads it. This logic is the structural analogy to the algorithmic arrangement. The recommender knows the person better than any human pastoral instance before it — click history, dwell time, scroll depth, reaction patterns to a hundred variants of stimulus. And it leads her: not by command, but by what it shows her.

Foucault’s gouvernementalité (Naissance de la biopolitique, 1978/79) sharpens this: governing operates through the arrangement of incentives, not through prohibition. This is exactly what the algorithmic arrangement does. Important: Foucault himself is descriptive. His analysis is a tool, not a justification. The normative evaluation comes from another tradition.

Why this is oblivion of the person

Three reasons:

First, the algorithmic arrangement treats the person as a statistical profile, not as a someone. Spaemann (Persons, 1996, ch. 8): “Persons exist only in the plural” — personal reality constitutes itself in recognition as an unmistakable counterpart. Recommender logic addresses not a person, but the intersection of profile features.

Second, it undermines what Karol Wojtyła determined as the self-transcendence of the person (Osoba i czyn, 1969): the capacity to go beyond oneself, to turn to something that does not follow from one’s own prior history. Whoever lives in an optimized feed receives almost only what fits his own prior history. Self-transcendence is replaced by self-mirroring — what Shannon Vallor (The AI Mirror, Oxford UP 2024) has named a “backward-looking mirror.”

Third, it collides structurally with the truth: optimization is for dwell time and click, not for truth-faithful representation. Christopher Bail (Breaking the Social Media Prism, Princeton UP 2021) differentiates the simple filter-bubble thesis — the causal story is more complex than “algorithm radicalizes.” But that the algorithmic arrangement is not calibrated to truth remains undisputed.

Empirical situation 2024–2026

Frances Haugen’s Facebook leak (2021) documented internal acknowledgment that engagement optimization intensifies polarization. Salvi et al. (Nature Human Behaviour 2025) showed that GPT-4 with access to demographic information was more persuasive than human debate partners in 81.7 % of cases — the persuasion-scaling potential of the combination of microtargeting plus LLM is demonstrated. The effects in the 2024 elections (USA, EU, Slovakia) were present, but empirically less decisive than feared (Simon/Altay/Mercier, HKS Misinformation Review 2023).

Caution: “less decisive in 2024” does not mean “no manipulation risk.” The effect sizes can become nonlinear once the technology becomes cheaper and better.

Ontological classification

Sources: Generated by querying the Personhood ontology.

Further sources:

  • Foucault, Michel (1975): Surveiller et punir. Paris: Gallimard.
  • Foucault, Michel (2004): Sécurité, territoire, population. Cours au Collège de France 1977–1978. Paris: Gallimard/Seuil.
  • Foucault, Michel (2004): Naissance de la biopolitique. Cours au Collège de France 1978–1979. Paris: Gallimard/Seuil.
  • Zuboff, Shoshana (2019): The Age of Surveillance Capitalism. New York: PublicAffairs.
  • Bail, Christopher A. (2021): Breaking the Social Media Prism. How to Make Our Platforms Less Polarizing. Princeton: Princeton University Press.
  • Salvi, Francesco et al. (2025): “On the conversational persuasiveness of GPT-4”. Nature Human Behaviour 9, 1645–1653 (online 2024, print 2025). DOI 10.1038/s41562-025-02194-6. (arXiv preprint 2024: arXiv:2403.14380.)
  • Simon, Felix M.; Altay, Sacha; Mercier, Hugo (2023): “Misinformation reloaded? Fears about the impact of generative AI on misinformation are overblown”. Harvard Kennedy School (HKS) Misinformation Review 4(5).
  • Vallor, Shannon (2024): The AI Mirror. Oxford: Oxford University Press.
  • Spaemann, Robert (2006): Persons. The Difference between ‘Someone’ and ‘Something’, transl. Oliver O’Donovan. Oxford: Oxford University Press (German original: Personen. Stuttgart: Klett-Cotta, 1996).
  • Wojtyła, Karol (1979): The Acting Person, transl. Andrzej Potocki. Dordrecht: Reidel (Polish original: Osoba i czyn, 1969).

See also

Graphansicht

Backlinks