AI doesn't need with consciousness or a sense of agency to be dangerous. AI doesn't need to be implemented entirely, or even partly, with neural networks either

AI doesn't need with consciousness or a sense of agency to be dangerous. AI doesn't need to be implemented entirely, or even partly, with neural networks either. I'd highly recommend Bostrom's book Super Intelligence if you have the time and inclination to read a book that makes this argument more clearly.