Lying to the ghost in the machine
Charlie Stross,
Charlie's Diary,
Apr 02, 2021
Two things of note here. The first is the author's main argument, which is this: "ready-trained NNs like GPT-3 or CLIP are often tailored as the basis of specific recognizer applications and then may end up deployed in public situations... This is the future of security holes in our internet-connected appliances." Quite right. The second point is a question I would pose readers in general: is it unethical to lie to artificial intelligences? Clearly that could depend on the purpose to which the AI is put, but of course, we seldom know what that purpose will be. It would also depend on our perception of the risk the AI poses to ourselves, but does our perception of risk outweigh the potential harm lying to an AI could cause? Interesting questions, all round.
Today: 5 Total: 1150 [Direct link] [Share]
