AI Avoids Technology Criticism

LLM-generated content rarely includes critical or skeptical attitudes toward technology, an absence that users recognize as a tell.

Study participants noted that when personas expressed concerns about technology (“very unsure about technology,” “scared of making mistakes”), this signaled human authorship. Participants reasoned that AI “would not describe technology as something negative.”

This reflects a structural bias: LLMs trained on internet text (which skews toward tech-positive, early-adopter perspectives) are less likely to generate technology-skeptical viewpoints. The absence isn’t random, it’s a predictable artifact of training data composition.

For design contexts, this matters: personas representing technology-hesitant users may require deliberate human input rather than AI generation.

Related: 05-atom—llm-stereotype-defaults, 07-atom—dq-dimensions-ai-training-data, 05-atom—flaws-as-authenticity-signals