From tools to thieves: Measuring and understanding public perceptions of AI through crowdsourced metaphors

Public perceptions of AI are becoming increasingly human-like: Over a 12-month period, Americans' views of AI as human-like increased by 34% and perceptions of AI's warmth increased by 41%, suggesting a significant shift in how people mentally model AI systems.

Most Americans view AI as both competent and trustworthy: The study found that Americans generally perceive AI as warm and competent, and these positive perceptions strongly predict whether people will trust AI and adopt it in their lives (explaining 18-21% of variation in adoption willingness).

Metaphors reveal how people actually think about AI: By collecting over 12,000 open-ended metaphors rather than relying on traditional survey questions, the researchers captured 20 dominant mental models that better reflect how the public genuinely understands AI—moving beyond surface-level responses to reveal deeper conceptual frameworks.

Demographic groups have significantly different AI perceptions: Women, older individuals, and people of color show higher tendencies to anthropomorphize AI (attribute human qualities to it), which has important implications for understanding why trust and adoption rates may differ across demographic groups.

How we talk about AI matters for responsible deployment: The study demonstrates that implicit perceptions shaped by metaphors—not just explicit knowledge—drive real-world decisions about AI adoption, suggesting that public communication about AI should carefully consider what mental models are being reinforced.