What if AI caused World War III?
What if I told you that last week, world peace hung by a thread because of erroneous information generated by artificial intelligence for the U.S. government? The U.S. military was already preparing to intercept and board a Chinese ship when officials discovered that the chatbot used to analyze data submitted by U.S. Special Operations Command had mistakenly assumed that the Chinese ship was carrying nuclear weapons. This news was reported by journalist Kyle Orland of arstechnica.com, who covers topics related to hardware, virtual reality, and generative AI.
Created by Bing Image Creator
Everyone already knows that AI isn’t 100% reliable. A term has even been added to the Cambridge Dictionary to describe what AI does when it doesn’t have an appropriate answer to a problem posed by the user: it simply makes up an answer. When this happens, we say that the AI “hallucinated.” It sounds like teen slang, but it’s a serious matter. And everyone who uses AI frequently is aware that this happens.
Despite widespread knowledge of AI’s tendency to hallucinate, scandalous cases continue to surface in the media where professionals pass off AI-generated texts to clients without performing any quality control, as if artificial intelligence were error-proof. In Brazil, lawyers for telecommunications companies submitted AI-generated petitions containing false case law to the court; in Canada, Air Canada’s chatbot invented nonexistent discount policies, and Ernst & Young produced a cybersecurity report in which more than 70% of the references were false; in the Middle East, the consulting firm PricewaterhouseCoopers advised governments and companies on the future of electric vehicles and automation based on false data. The number of cases like these is immense. However, it has been three years since instances of AI hallucination earned their own term in dictionaries.
Despite the risks involved, the widespread use of AI continues to grow by leaps and bounds. The U.S. Department of Defense recently announced that it had used Google’s Gemini and SpaceXAI’s Grok as the foundation for building its military generative AI platform, which is used by more than 1.5 million users in the armed forces.
In 2023, the State Department had already warned that the military should ensure the presence of a human expert to evaluate material produced by AI, in order to prevent inappropriate biases in data interpretation from leading to accidents. Despite the warnings issued by the U.S. State Department, there are reports that NATO has, in the recent past, used autonomous, AI-controlled drones in its attacks during the Russia-Ukraine conflict. Cases like this have led Anthropic to refuse to collaborate on the development of AI models for the Department of Defense’s control of autonomous weapons, given the possibility of lethal accidents.
As can be seen from these few lines, the risk of using AI in matters of war is currently immense; however, this has not prevented its use. Common sense would suggest that there is still a long way to go before artificial intelligence is capable of adequately performing all the tasks for which it has been programmed. Until that happens, we must ask ourselves how long we will continue to accept that leaders treat deaths caused by technological failures as mere collateral risks. Technology should exist to protect and improve our lives—not the other way around.