Emily Bender Clarifies "Stochastic Parrots" Metaphor and AI Terminology

Emily M. Bender, a professor of computational linguistics at the University of Washington and lead author of the "On the Dangers of Stochastic Parrots" paper, has published a blog post to address common misunderstandings of the "stochastic parrots" metaphor five years after its initial publication. The paper, released in March 2021, argued that large language models (LLMs) generate text by statistically predicting word sequences rather than through genuine understanding, likening them to "stochastic parrots." Bender emphasizes that the metaphor specifically applies to LLMs and not to all systems labeled "artificial intelligence." She clarifies that the term "stochastic parrots" was not intended as an insult but as a descriptive metaphor for how these systems function. Bender also advocates for clearer terminology than "artificial intelligence," arguing that the umbrella term obscures the distinct capabilities of various technologies and hinders informed decision-making and regulation. She notes that "AI" has become conflated with chatbots and LLMs, overshadowing other AI applications like AlphaFold. Bender explains that the phrase "stochastic parrots" was chosen for its catchiness and to highlight the risks associated with increasingly large language models, even though the term itself appears only twice in the original paper. Her work, along with other metaphors like the "octopus test," aims to make the operational mechanisms of these systems more vivid to a broader audience.