↖
back
More Work
→

Words Apart

Words Apart is a data visualization project comparing how people and language models reach for different words when answering the same questions. Built on a topic-matched corpus of advice posts - human replies and model replies to the identical prompts - it turns a linguistic difference into something you can look at and play with, including The Humanometer, which scores any word you type on how human or machine its use is.
scroll right
→
Process
01
Pairing the corpus
Most human-versus-AI language comparisons quietly compare topics instead of authors: one corpus explains photosynthesis, the other brings personal problems. The project is built on 50 advice posts answered by both people and a model, so every pair argues about the same thing and only the author changes.
02
Finding what actually differs
Measured on that paired set, the headline claim that AI leans on empathy words did not hold up. What did hold up is a move rather than a vocabulary: validation — “your feelings are valid”, “you’re not wrong” — appears in 4.0% of human answers and 19.3% of machine ones, and the gap survives cutting every answer to the same length. Imagination words run 1.67 times as often on one side.
03
Making it something you can hold
The findings are drawn as two gumball machines, a head and a robot, each loaded with the words its side reaches for. The Humanometer turns the same data into a tool: type any word and it reports, on an ECG readout, how human or machine that word’s use is.