Emily M. Bender explains why “stochastic parrots” was never meant as an insult

Emily M. Bender, professor of computational linguistics at the University of Washington, is pushing back against widespread misreadings of her 2021 paper “On the Dangers of Stochastic Parrots.” Gwendolyn Rak reports for IEEE Spectrum that Bender used the paper’s five-year anniversary to clarify what the term actually describes.

Not all of AI, and not an insult

According to Bender, the phrase “stochastic parrots” refers specifically to large language models, not to artificial intelligence as a whole. She notes that the word “AI” appears only once in the original paper, near the end, in a warning about systems designed to mimic human language closely enough to be mistaken for people.

Bender also rejects the idea that the metaphor was meant to belittle the technology. “It’s just a description of what these systems actually are,” she tells Rak. Calling it an insult, she argues, would require believing that a language model can take offense, or that it falls short of some ideal she does not hold.

A missing chapter on labor

Asked what she would add today, Bender points to one major gap: exploitative labor practices. She says the original paper failed to address the harsh conditions faced by data workers and the large-scale use of people’s creative and intellectual work to train these systems.

Bender also criticizes the umbrella term “artificial intelligence” itself, arguing that it groups unrelated technologies, such as chatbots and protein-folding tools like AlphaFold, together in ways that obscure real differences and hinder informed decisions about regulation and use.