Agustin V. Startari

Modern conflict reporting does not always erase suffering. Often, it does something more subtle: it shows suffering while weakening the grammar of responsibility. In AI-generated summaries, automated headlines, platform moderation outputs, and press rewrites, Palestinian civilians can appear as “killed,” “displaced,” “affected,” or “caught in conflict,” while the actor responsible for producing that condition disappears from the sentence.

This article explores how predictive systems displace responsibility by producing authority without subjects. It introduces accountability injection, a three-tier model (human, hybrid, syntactic supervised) that structurally reattaches responsibility. Case studies include the AI Act, DAO governance, credit scoring, admissions, and medical audits, offering a blueprint for legislators and regulators to restore appeal and legitimacy in predictive societies.

Large language models are trained to predict words, not to check facts.They are optimizers of plausibility, not validators of reliability. How AI Tricks Us Into Trusting It