Artificial intelligence can trick partisans into trusting news they hate, but it has no idea why its own tricks work.
Large language models can reframe toxic headlines into versions that both Democrats and Republicans find trustworthy. Most people assume that if a machine is smart enough to manipulate human emotions, it must understand human psychology. These models consistently fail to predict which specific people will respond to the debiasing. The AI is essentially a master of social engineering that is completely blind to the actual minds it is influencing. This means we are building tools that can change public opinion without any internal compass for how they are doing it.
Can AI Debias the News? LLM Interventions Improve Cross-Partisan Receptivity but LLMs Overestimate Their Own Effectiveness
arXiv · 2605.01006
Partisan news media erode cross-partisan trust, but large language models (LLMs) offer a potential means of debiasing such content at scale. Across two pre-registered experiments, we tested whether LLM-generated debiasing of liberal news headlines could improve conservative readers' trust-relevant judgments. Study 1 found that subtle lexical debiasing (replacing emotive words with more moderate synonyms) had no effect on any outcome. Study 2 found that a more substantive reframing intervention s