Loading page…
In a new WIRED interview, the researcher argues that safety debates should examine the institutions behind AI systems and the harms they may already be enabling.
Listen to this story
In a new WIRED interview, Timnit Gebru argues that AI accountability should center on the people and institutions designing, testing, approving, and deploying systems—not on speculation about machine sentience or extinction. She highlights autonomous weapons, job displacement, environmental costs, and chemical or biological weapons as immediate concerns.
Gebru’s bridge analogy shifts failure analysis toward designers, testers, operators, and regulators rather than autonomous systems themselves.
She identifies autonomous weapons, workforce displacement, climate effects, and chemical or biological weapons as present-risk areas.
Gebru criticized OpenAI’s handling of a disputed million-dollar mathematics problem, arguing mathematicians should have reviewed the claim first.
In a new WIRED interview, AI researcher Timnit Gebru argued that warnings about AI-driven human extinction can pull attention from harms already tied to the technology’s use. She pointed instead to autonomous weapons, workforce displacement, climate effects, and chemical or biological weapon risks—and to the companies and institutions making choices about those systems.
Gebru previously worked at Google evaluating bias in its AI tools. She left after Google rejected a paper she coauthored about risks from probabilistic language models.
Her position is not that safety questions should be ignored. She argues that attention should move away from whether a model is ethical, sentient, or independently rogue, and toward the humans and organizations that create, test, approve, and deploy it.
Can a bridge decide to collapse?
Timnit Gebru, speaking to WIRED
In her comparison, investigators would not ask whether a collapsed bridge made an ethical decision. They would examine who designed and built it, whether it was tested, and who permitted it to operate. Gebru argues AI-system failures deserve that same chain of scrutiny.
She said companies’ attention to programming, chess, and mathematics can turn narrow successes into a larger claim that they have solved intelligence. Her concern is the promotional leap from a specific result to a sweeping story about a system’s capabilities.
Gebru applied that concern to OpenAI’s handling of a disputed million-dollar mathematics problem. She said the claim should have received fuller review from the mathematics community before it was presented as a breakthrough.
She also said anticipated initial public offerings have made AI researchers more competitive and less collaborative, contrasting that pressure with earlier cross-company collaboration she experienced while at Google. Her critique leaves a practical next question: whether companies’ testing, review, and public claims will keep pace with the systems they put into use.
Loading discussion...
Join the conversation
Share what questions companies should answer first.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.