These are the types of events we need to know about, and fit into the category of AI issues that create dangerous scenarios.
CNN reported the US military was prepared and just about to stop and board a Chinese ship that an AI generated intelligence report said was carrying components for a nuclear weapon to the Mideast. It was called off at the last minute when military planners took one last look at the intelligence, and realized it as a chatbot that hallucinated incorrect information.
AI hallucinations are explained here, but in general are AI outputs that sound correct, but are actually entirely wrong and faulty. This is why it is vital for something like military decisions to still have human oversight and final determination of actionable orders, even though AI is being used more and more to generate reports and analyze intelligence data.
This was a close call. Is it just a matter of time before something catastrophic, meaning people die, occurs because of incorrect AI reporting and conclusions?
No comments:
Post a Comment