Suffering Without Perpetrators: The Grammar of Responsibility Loss

Civilian suffering can remain completely visible while responsibility becomes grammatically optional. My paper Suffering Without Perpetrators: The Humanitarian Passive in AI-Generated Conflict Discourse develops this problem through the concept of the humanitarian passive.

The humanitarian passive is not limited to passive voice in the narrow grammatical sense. It includes nominalizations, agentless clauses, abstract causality, humanitarian vocabulary, security frames, and moderation-oriented reformulations that preserve the fact of harm while weakening the path from harm to the actor responsible for producing it.

When consequence survives and agency disappears

A formulation such as civilians were killed can be factually true, but it does not necessarily preserve who killed them. Access to medicine became limited can describe a real humanitarian condition while obscuring the actor or policy that imposed the restriction. Content was removed can report suppression while leaving the suppressing institution grammatically absent.

The paper focuses on Palestine, Iran, and platform moderation because these cases reveal how humanitarian visibility and responsibility attribution can diverge. A discourse may sound compassionate, balanced, and neutral while still reducing the traceability of political agency.

From bias detection to responsibility detection

To measure this process, I introduce the Responsibility Loss Index, or RLI. The aim is to evaluate whether AI-generated summaries, headlines, reports, and moderation notices preserve or erase the agents responsible for violence, sanctions, restriction, censorship, or humanitarian harm.

The larger argument is that AI ethics should not stop at asking whether a system is biased, toxic, or inaccurate. It should also ask whether the grammar of the system preserves the conditions under which responsibility can still be named.

Full paper: https://zenodo.org/records/20139961