Raw LLM Responses

There are two problems: technical and social. Technically, previous systems have all had control transferred to a human in the trigger loop. I don't know about KAIST, but the Korean Defense Department (ADD) is funding both robotics and AI. The way I heard it phrased during an explanation of their current research directions was "but of course we cannot actually hook it up because of the ethical issues". Right, like once the hardware and software are both developed they will be strongly firewalled. Socially, my guess would be that either the KAIST guys were bragging at an international conference, or they were submitting military-funded conference papers about distinguishing types of clothing with AI (or something else suitably suspicious) and other researchers put 2 and 2 together. Money is on the former: most professors are not shy about talking shop, because talking leads to international collaborations. There is another possible social aspect to this: most countries don't publish research papers on actual weapons development. Koreans see it as low-hanging fruit, and so they do. As in, they will see from PR demos that the US developed a system, but didn't publish papers on it, so they re-develop the idea, and publish papers themselves. Could be very problematic when the technology is uniform-distinguishing AI, and the paper contains pseudocode...