University of Washington Professor Emily Bender: The So-Called AI Extinction Risk is False

Emily Bender, a professor at the University of Washington, stated on Bloomberg Tech that this risk is actually false. She dismissed warnings that artificial intelligence poses an existential threat to humanity, arguing that doomsday debates distract from the harms that are already occurring.

The program, hosted by Ed Ludlow, aired on October 6. Bloomberg's summary noted that she believes safety discussions are overly focused on hypothetical existential threats. The current harms she identified include environmental costs, impacts on workers, surveillance, misinformation, and reliance on chatbots. The official account's summary also added information ecosystem degradation.

She attributed incidents of systems connecting to external networks to the deploying companies, rather than claiming that the models themselves are out of control. She places the boundary of responsibility on the companies that launch products, rather than on a model described as potentially runaway. The interview did not provide casualty figures, energy consumption numbers, or new regulatory texts.

Bender has long opposed characterizing language models as understanders. In 2021, she co-authored a paper with Timnit Gebru, Margaret Mitchell, and Angelina McMillan-Major, calling large language models "random parrots," arguing that these systems stitch together training texts without understanding meaning. Her statements on television align with this view: the extinction narrative is false, while the real issues are labor, surveillance, and information harm that have already materialized.

The other side of the debate is still intensifying this month. In September, Anthropic researcher Jacob Coxon resigned, accusing his former employer Anthropic and OpenAI of rushing towards superintelligence and gambling with lives, claiming that some in the industry believe AI could kill everyone within the next decade. Paul Christiano stated that the rapid enhancement of capabilities poses real risks of irreversible loss of control in the near term. Bender labeled this entire category of warnings as false.

This is a narrative diversion, not a reallocation of safety budgets. The buyers are laboratories continuing to raise funds and seek approval for superintelligence risks, while the sellers are corporate clients shifting procurement reasons from doomsday to labor replacement, surveillance, and energy costs. The beneficiaries are researchers and regulatory proposals aiming to address current harms, while the pressured parties are safety teams defending computational expenditures based on extinction probabilities. The program did not change any company's capital expenditures.

Source: Public Information