AI Influence on Human Thought and Action: A Lexicon and Taxonomy of a Fragmented Literature

Abstract

Human oversight of artificial intelligence is often justified by having a person who can reject the system's recommendation. Yet AI may already have shaped what that person sees, notices, and thinks. This article brings together terms from several disciplines to classify these influences and examine their ethical implications. It groups established findings and emerging concepts by how AI selects information, directs attention, shapes understanding, takes over mental work, affects judgment, and feeds earlier responses into later decisions. It also distinguishes influence through recommendations from influence through the way choices and evidence are presented. The central argument is that giving a human the final say is not enough. To protect people from AI errors, reviewers need time, information, and authority to check the reasoning, consider overlooked options, and challenge how the system presents the problem. Designers and organizations share responsibility for making that review possible. The article also explains why human agreement after AI exposure cannot be assumed to provide independent validation. Its proposed safeguards require testing in practice. The aim is to preserve people's ability to judge AI assistance, question it, and act on their own conclusions. Version 2.0 (September 24, 2026) corresponds to author-approved manuscript v0.6. It expands the ethical analysis and simplifies the language. The accompanying supplementary lexicon contains 22 terms and their supporting sources. This is an unreviewed preprint; it has not been peer reviewed.