Source Body Text

Amid a barrage of news stories warning about superintelligent machines rendering humanity extinct, a CNN story describing the opposite scenario – one in which the US military’s reliance on brittle chatbots almost brought the US into war with China – went mostly unnoticed by the public. The biggest international AI news of the past three weeks was the Anthropic engineer Jacob Coxon’s resignation. According to him, OpenAI and Anthropic are “racing straight towards self-improving superintelligence and gambling with our lives”. Coxon’s description of a “terminator” scenario, a machine becoming much smarter than humanity and deciding to wipe us out, captured the public’s, journalists’ and politicians’ attention, stirring waves of fear-inducing, credulous reporting, and policy actions. The California governor, Gavin Newsom, issued a mandate to create an “AI kill-switch”; New York’s governor, Kathy Hochul, is similarly exploring the possibility of having “AI kill-switches”; and Senator Bernie Sanders and Congressman Greg Casar introduced legislation to “ban artificial superintelligence”. While so many of us were focused on fictional superintelligent machines that we were told could hypothetically cause human extinction, however, there was a more serious story being reported. A series of events almost got us to armageddon, not caused by “rogue, superintelligent AI”, but by government reliance on error-prone systems that are marketed by their manufacturers as being close to superintelligent. A CNN report published on 18 September, which hasn’t been verified by other major outlets, describes how the US “almost started a war” with China, based on an intelligence report that claimed a Chinese ship was transporting components of nuclear weapons. Military personnel made plans to intercept the Chinese vessel, complete with aircraft and soldiers ready to board the ship. A military skirmish of this kind between China and the US could quickly escalate and spiral into world war three. It was only moments before executing on the plan that American officials further investigated the intelligence report and found that it had been generated with the use of a chatbot and contained erroneous information about the vessel’s contents. Unlike the terminator, “rogue AI”, scenario, this incident didn’t galvanize lawmakers into announcing calls for regulation so as to avoid potential human extinction. Media outlet after media outlet didn’t warn the public that we almost triggered a war against a nuclear superpower because of our military’s reliance on error-prone chatbots. We describe the large language models (LLMs) powering these chatbots as stochastic parrots, that is, models that are designed to regurgitate the patterns of the data they are trained on. News coverage describing them as “powerful” “rogue” models that could render humanity extinct all on their own, bolsters perceptions of tools based on these models being infallible enough to merit usage in stakes as high as warfare. Even CNN’s more critical coverage describes the chatbot used by the US military as presenting “profound risks” because it is “powerful, new and relatively poorly understood technology”. In fact, today’s LLMs and the chatbots they power are not poorly understood, nor are they powerful in the sense of being effective tools for this use case. They are systems for generating plausible-looking text, built out of enormous, haphazardly collected datasets and then “fine-tuned” (that is, further trained) to be especially appealing to their intended users. CNN wasn’t able to learn which chatbot product was used in this case, but the LLM powering the product was probably fine-tuned to output text with the stylistic hallmarks of intelligence reports. Both the basic functionality and the risks of such a model have been well-understood and well-documented for years, including in our 2021 paper On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? Claims that system behavior is mysterious arise from deliberate mystification on the part of those selling those systems, and a misapprehension of what their output is by those procuring and using it. Unlike the unscientific and fantastical claims of OpenAI and Anthropic CEOs and those who uncritically repeat their talking points, the harmful, and well-documented, impacts of these companies’ products have arisen because people believe that the products are superintelligent, not because they actually are. From “AI” scribes used by hospitals erroneously classifying patients as illicit drug users, to governments killing children after relying on “intelligence analysis” systems which mistake schools for military facilities, we are seeing the predictable harms that come from misplaced faith in unreliable automated software. Meanwhile, the marketing campaigns from the companies selling this software warn of impending superintelligence, directing our attention away from the real risk to their fantasies of doom. It is high time for us to regulate usage of systems sold as “AI”, not because they are magical machines, but because they are error-prone products that shouldn’t be used in high-stakes scenarios. Any automation used in military, medical or other life-and-death situations must be thoroughly evaluated within its use case and paired with usage norms that keep accountability with people who have actual power to make decisions. As the former FTC chair Lina Khan has repeatedly reminded us, there is no exemption from the law when it comes to AI: there are existing laws that federal agencies and other regulators can enforce to protect the public against the AI industry’s unsafe and unscientific practices of pushing unsound products while marketing them as “magic intelligence in the sky”. Going along with company executives and describing their products as beings with agency allows these executives to evade accountability while deceiving the public. We call on journalists and lawmakers to hold companies accountable rather than parrot their marketing talking points. Timnit Gebru is executive director of Dair and author of the forthcoming book Deep Unlearning: The Radicalization of a Tech Idealist, which is available for preorders now and set to publish on 16 February. Emily M Bender is professor of linguistics at the University of Washington and co-author of The AI Con.