Two well-known AI researchers say a chatbot error nearly pushed the United States into a military clash with China. Timnit Gebru and Emily M Bender made the claim in an opinion piece published on 1 October.
Their argument centers on a CNN report from 18 September. Other major news outlets have not verified that report.
What the CNN Report Described
According to CNN, US officials received an intelligence report claiming a Chinese ship was carrying parts for nuclear weapons. Military staff then made plans to stop the vessel.
Aircraft and soldiers were reportedly ready to board the ship. A clash like that between the two countries could have escalated quickly.
Moments before the plan went ahead, officials took a closer look at the intelligence. They found it had been produced with a chatbot and contained false details about the ship's cargo.
CNN could not learn which chatbot product was used. Gebru and Bender believe the model behind it was likely trained to write text that looks like an intelligence report.
Focus on Superintelligence Fears
The authors say the CNN story got little public attention. Much of the recent coverage has instead focused on the resignation of Anthropic engineer Jacob Coxon.
Coxon said OpenAI and Anthropic are "racing straight towards self-improving superintelligence and gambling with our lives." His warning drew wide attention from reporters and politicians.
California Governor Gavin Newsom issued a mandate to create an "AI kill-switch." New York Governor Kathy Hochul is exploring a similar idea.
Senator Bernie Sanders and Congressman Greg Casar introduced legislation to "ban artificial superintelligence." The authors note the chatbot incident did not lead to similar action.
Gebru and Bender describe large language models as "stochastic parrots." They used the term in a 2021 paper to describe systems that repeat patterns from their training data.
They say these tools are built to produce text that sounds believable, not text that is correct. In their view, the risks have been documented for years.
The authors argue that harm happens because people believe AI products are superintelligent. They point to hospital AI tools that wrongly labeled patients as drug users as one example.
They also say company warnings about superintelligence pull attention away from these everyday failures. In their view, that helps executives avoid accountability.
Gebru and Bender want rules on how AI is used in areas like the military and medicine. They say any such system should be tested for its specific use, with people kept accountable for decisions.
They also cite former FTC chair Lina Khan, who has said there is no legal exemption for AI. Existing laws, they argue, can already be used to protect the public.
Gebru leads the research group Dair, and Bender teaches linguistics at the University of Washington. As of publication, CNN's account of the near-interception has not been confirmed by other major outlets.
