If you think about it, we already do by the recommendation engines of social media platforms*(is YT social media?)

I mean, we had an elderly President with Alzheimer's taking advice from his wife's astrologer. That isn't much worse.

If leaders use LLMs even a fraction as often as I do each day, we can say with 99.99% certainty that major decisions are already heavily influenced by them. IMHO, that's a significant improvement over human intervention, which often comes with biases or even ill intentions.

LLMs are not free of biases (and, because there are comparatively few major LLMs in use, and with very similar training, far less diversity in biases than humans that might be consulter.)

A credible source said that DHS was using AI software to identify immigrants to arrest, and may have been responsible for the incorrect charges/warrant for the Columbia student leader who is a permanent resident (they thought he was on a student visa, iirc).