Emily M. Bender is a computational linguist, Professor of Linguistics at the University of Washington, and faculty director of the UW Computational Linguistics Master's Program. She is among the more prominent academic critics of frontier-AI hype and a co-author of two papers central to that critique. She was profiled, alongside her debating partners, in Heaven's MIT Technology Review feature (July 2024).
Background and roles
Bender holds a professorship in linguistics at the University of Washington and directs the faculty side of the UW Computational Linguistics Master's Program. With sociologist Alex Hanna of the DAIR institute, she cohosts the podcast Mystery AI Hype Theater 3000, which reviews public claims made by AI executives. A frequent rhetorical device on the show is reading such claims aloud while substituting "mathy math" for the word "AI"; Bender uses the reframing to argue that anthropomorphic language ("AI feels," "AI knows," "AIs are individuals") is a category error.
Anchor papers
In "Climbing towards NLU" (ACL 2020), written with Alexander Koller of Saarland University, Bender introduced the octopus thought experiment: a model trained only on text learns the form of language but not its meaning, because meaning consists of words plus the reasons they were uttered. The paper is a recurring reference point for the argument that LLMs mimic linguistic form without understanding (see Stochastic Parrots (Bender, Gebru, McMillan-Major, Mitchell, 2021) and the Octopus Test).
"On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜" (FAccT 2021), written with Timnit Gebru, Angelina McMillan-Major, and Margaret Mitchell, described LLMs as "haphazardly stitching together sequences of linguistic forms ... without any reference to meaning" and popularized the "stochastic parrots" characterization as a term used against LLM hype. An internal Google conflict over the paper led to Gebru and Mitchell leaving the company.
Positions and statements
Bender treats the term "AI" itself as a Big Tech buzzword that distracts from associated harms, and declines to use it without scare quotes.
On methodology, she argues that behavior is not evidence of understanding. This is the basis of her critique of Bubeck's Sparks paper, which she called "a fan fiction novella," and of benchmark-driven claims about model capabilities more broadly.
Bender endorses the TESCREAL diagnosis (TESCREAL Framework (Gebru and Torres)), crediting Gebru and Émile Torres with the lens. She has described the stakes of "tangling with those people" as "more here than just academic ideas. There's a moral code tied up in it as well."
Blaise Agüera y Arcas of Google has called critics such as Bender "AI denialists." Bender's response is that extraordinary claims require extraordinary evidence, which she argues is not available.
Her recurring debating partners include Sébastien Bubeck of Microsoft Research (Sparks of AGI versus stochastic parrots), Blaise Agüera y Arcas of Google (the "AGI is already here" debate), Geoffrey Hinton (the claim that neural networks can do everything), and Sam Altman and Mustafa Suleyman on frontier-lab marketing language.
Bender's arguments against extrapolating from current models to AGI sit in dialogue with AGI Timelines, Sparks of Artificial General Intelligence: Early experiments with GPT-4, Scaling Laws, General-Purpose AI (GPAI), LLM Fallacy, and Mechanistic Interpretability. She is associated with the critique grouping that includes Gebru, Marcus, and the DAIR and AI Now Institute circles.
Relationships
- related: Stochastic Parrots (Bender, Gebru, McMillan-Major, Mitchell, 2021) and the Octopus Test — co-author of both anchor papers
- related: TESCREAL Framework (Gebru and Torres) — sister critique she endorses
- contradicts: Sparks of Artificial General Intelligence: Early experiments with GPT-4 — primary public critic of the paper's methodology
- contradicts: AGI Timelines (in part) — challenges the extrapolation premise
- related: Timnit Gebru — co-author on Stochastic Parrots; allied DAIR figure
- related: Sébastien Bubeck — recurring debating partner
- related: LLM Fallacy — sister critique
- supports: What is AI? — MIT Technology Review (Heaven, 2024) — primary biographical source in the wiki