People do this with pets, cars and weather. With AI it happens even faster, because chatbots write in fluent first-person language. A customer might thank a support bot, apologize to it or assume it “remembers” them, when the system is generating text from patterns and whatever data it was given.
Some human words are useful shorthand. Saying a model “thinks” or “hallucinates” can be convenient. The risk comes when the metaphor changes what people believe the system can do, for example trusting it like a colleague with judgment and responsibility.
Why it matters
Anthropomorphic framing affects trust. Users can over-rely on a system that sounds caring or certain, and builders can misdiagnose failures by asking what the model “wanted” instead of what in the data, prompt or design produced the output.
On the podcast
In episode 54, Dan Klein says the word “hallucination” comes partly from “anthropomorphizing these models” at 29:24 and argues the field should be more precise about what causes those errors. That is his view of the term. Galileo was the presenting sponsor of episode 54; the sponsor read is at 2:53.
Related terms
AI hallucination is the most common anthropomorphic label in AI. Large language model explains what is actually generating the text.
Hear it from the guest
“Anthropomorphizing these models, that when it tells you something that's not true, you're like, oh, it's hallucinating.”
Quotes lightly edited to remove filler words.
Sources
- Epley, Waytz and Cacioppo: On seeing human, a three-factor theory of anthropomorphism — Psychological Review paper (2007) that defines anthropomorphism as attributing humanlike characteristics, motivations or emotions to non-human agents and explains when people do it; checked October 24, 2026.
