When AI Turns Societies into Risk Objects

My latest paper, The Syntax of Digital Dehumanization: Subjugated Societies as Risk Objects in AI-Governed Discourse, examines a form of dehumanization that does not require slurs, explicit hatred, or openly discriminatory language. It can emerge through apparently neutral systems of classification.

AI-mediated discourse can keep a society highly visible while progressively stripping it of political subjecthood. Populations may appear as refugee flows, humanitarian burdens, extremist zones, sanction targets, crisis categories, security concerns, or moderation risks. They remain present in the text, but increasingly as objects to be managed rather than subjects capable of acting, demanding, resisting, negotiating, accusing, governing, or defining their own political position.

Dehumanization without hate speech

This distinction matters because contemporary AI governance is often audited through visible categories such as toxicity, hate speech, bias, misinformation, and explicit discrimination. Those categories remain important, but they can miss a subtler structural effect. A system may produce perfectly polite language and still reorganize the political status of the people it describes.

When a society repeatedly appears through security, humanitarian, sanctions, or crisis frames, the linguistic system may narrow the range of roles available to it. The dominant actor retains verbs of decision, strategy, intervention, enforcement, protection, deterrence, and governance. The subordinated actor is increasingly represented through exposure, suffering, instability, radicalization, displacement, need, or risk. The asymmetry is not simply ideological. It is grammatical.

Measuring political subjecthood

The paper proposes two analytical tools: the Political-Subjecthood Retention Rate, or PSRR, and the Digital Dehumanization Syntax Index, or DDSI. Their purpose is to move the argument from impressionistic criticism toward auditable linguistic structure. The question is not merely whether a population is mentioned, but whether the grammar preserves its capacity to appear as a political subject.

This framework is especially relevant to Palestine, Iran, and other societies positioned within asymmetric geopolitical systems. It also extends beyond any single conflict. The same method can be applied wherever automated systems summarize populations through administrative, security, humanitarian, financial, migration, or platform-governance categories.

The central claim

Digital dehumanization does not begin only when a machine uses openly hateful language. It can begin when a political community is grammatically converted into an object of management. The decisive question is therefore not only whether AI sees a society, but what kinds of agency the system allows that society to retain once it has been seen.

Full paper: https://zenodo.org/records/21903591