1:16 PM (6 hours ago) | ||||
Nuclear War could be Just Another HallucinationMilitary discusses how to use AI to launch
While we have recently been deluged with dire warnings that “rogue” AI agents might potentially conspire to wipe out humanity, we have also just discovered that earlier this year AI nearly caused a war between nuclear superpowers . Thanks to an amazing scoop by CNN, we learn that an AI chatbot in the service of Special Operations Intelligence in Hawaii flagged a Chinese merchant ship somewhere near Iran as carrying a cargo of nuclear weapons components. The relevant analyst unthinkingly (we assume) used a chatbot to lay out this “intelligence” as an official report and then disseminated it on a classified network worldwide. America’s warriors sprang into action, rapidly initiating plans to board and capture the ship. Troops assigned to the mission were actually in the air when one official thought it might be a good idea to look over the original report of the ship’s cargo. There were no nuclear components. The chatbot had simply had a hallucination and made up the information. So far, so worrisome. Not least because it confirms that the defense department is in not only inthe hands of a tehnical illiterate, it is staffed by, idiots who treat this sort of report as credible and eagerly rush to “go kinetic” regardless of the probable consequences. Hegseth, obedient to the dictates of the private equity oligarchy now helming Pentagon spending, has avidly promoted the introduction of AI into all Pentagon operations, with an appropriate shower of gold landing on Silicon Valley. In January he launched an “Artificial Intelligence Acceleration Strategy,” and festooned the Pentagon with pictures of himself in Kitchener-like pose urging universal military AI adoption It is always worth remembering that the technology does not confine itself to making up intelligence reports. According to the Financial Times,”the most popular AI models from ChatGPT, Claude, Copilot, Grok and Gemini provided wrong answers to financial queries 57 per cent of the time (my emphasis) on average, according to research from technology firm Saturn.” A Stanford study revealed that AI legal research tools hallucinate fictitious answers and case studies in as much as 34% of all queries. But there is a frightening subtext to this fiasco. Remember those scary cold war stories of the times when human or technical error almost brought on a U.S.- Soviet thermonuclear war? The time in 1980 when national Security Adviser Zbigniew Brzezinski was alerted that Soviet ICBMs were on their way, and was just about to wake up Jimmy Carter to discuss retaliation when another call reported it was all just a glitch caused by someone plugging in a training tape? Or the time in 1983 when a newly installed Soviet early warning system reported that five U.S. missiles were on their way? Luckily, a Soviet officer did not believe that anyone would start a war with five missiles and refused pass on the report to higher authority. In 1962 a guard at a US Air Force nuclear command post in Duluth shot at an intruder climbing the fence, setting off an alarm that had nuclear-armed bombers warming up on the runway. The intruder turned out to be a bear. The list of such brushes with disaster is long, and quite certainly longer than we have been told. Fortunately, human common sense has intervened in time to stave off Armageddon. But now, with the onslaught of AI, common sense is being relegated into obsolescence, as evidenced by the story of the Chinese ship. The details of AI integration into nuclear command and control are highly classified, but we do know that there has been a concerted effort to introduce machine learning into our early warning systems, as well as streamlining the process by which the president can order a nuclear attack. This system swings into action more often than is generally appreciated. Jon Wolfsthal, formerly director of global risk at the Federation of American Scientists, and currently the Fellow for Nuclear Policy at Pax Sapiens, spent years at the heart of US policymaking on nuclear security and non-proliferation. ... Keep reading with a 7-day free trialSubscribe to Spoils of War to keep reading this post and get 7 days of free access to the full post archives. A subscription gets you:
|