Abstract
AI systems are increasingly embedded in the information environments that shape human belief, preference, and behavior. This paper provides a comprehensive survey and synthesis of how AI systems influence human epistemics — the processes by which people form, revise, and act on beliefs. We identify three primary mechanisms: direct influence through personalized content and sycophantic responses; indirect influence through changes in the information ecosystem; and systemic influence through feedback loops between AI outputs and training data.
We further identify amplifiers that make AI influence qualitatively different from prior epistemic technologies: scale (simultaneous exposure of millions of users), personalization (responses tuned to each user's existing beliefs), and opacity (influence that operates below conscious awareness). Analyzing consequences at individual and population levels, we connect AI influence to sycophancy, confirmation bias, echo chamber formation, preference homogenization, and value lock-in. The paper concludes with a framework for evaluating and intervening on AI epistemic influence, distinguishing interventions at the model, interface, ecosystem, and governance levels.
Cite
@article{qiu2026aiinfluence,
title = {AI Influence: Mechanisms, Amplifiers, and Consequences},
author = {Qiu, Tianyi and He, Zhonghao and Lin, Tao and Glickman, Mark
and Calcott, Rich and Wihbey, John and Kleiman-Weiner, Max},
journal = {SSRN Preprint},
year = {2026},
url = {https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5756323}
}