Emily M. Bender is a professor of linguistics at the University of Washington and a leading academic critic of overstated claims about what large language models understand or know.
Bender's academic background is in computational linguistics, and her research has long focused on the relationship between linguistic form and meaning, and on the risks of treating fluent language production as evidence of genuine comprehension. In a 2020 paper with Alexander Koller, "Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data," Bender introduced the "octopus test," a thought experiment in which a hyper-intelligent octopus, having only ever observed two people communicating by cable without any access to the physical world they describe, learns to produce statistically plausible responses without ever grasping what the words refer to. Bender used the analogy to argue that systems trained purely on the statistical patterns of text, however large, cannot be assumed to have acquired meaning or understanding in the way a human speaker does.
Stochastic Parrots
In 2021, Bender co-authored "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" with Timnit Gebru, Angelina McMillan-Major and Margaret Mitchell. The paper, written while Gebru and Mitchell were at Google, argued that scaling large language models carried environmental costs, risked encoding and amplifying social biases present in web-scraped Training data, produced text that was difficult to audit for factual accuracy, and encouraged users and even researchers to mistake fluent output for understanding, a critique that gave the field the widely used term Stochastic parrot. The paper's publication, and Google's objections to it, contributed to Gebru's high-profile departure from the company.
Public commentary
Bender has remained one of the most visible academic voices pushing back on anthropomorphizing language, describing chatbots and pushing for more precise, non-mentalistic vocabulary when describing what LLMs do, objecting for instance to describing models as understanding, knowing, or hallucinating in a literal sense, even while accepting terms like Hallucination (AI) as established shorthand within the field. She co-hosts the podcast "Mystery AI Hype Theater 3000" with sociologist Alex Hanna, in which the two critically examine media coverage and industry claims about AI capabilities.
Influence
Bender's linguistics-grounded skepticism has become a standard reference point in debates over Artificial general intelligence timelines and claims of emergent understanding in large models, frequently invoked alongside critiques from researchers such as Gary Marcus as a counterweight to more optimistic narratives from major AI labs.