If the BASE problem is presented, they are extremely helpful. But if they have not been told what the problem is, or the problem has not been identified or told to them in the correct way, then they have some, but not a lot, of value.

AI has shown itself to not be aware of humanity. It is unfair to ask it to be aware of humanity since it is a machine. So in the search for a perfect answer, it may not take humanity into consideration.

"But", you say, "can't we program it to do that?"

Perhaps, but it will always be a mechanical abstraction to the AI, not true human characteristics.

If I asked AI how to stop A from destroying B, it may give me a concrete example that may not take into consideration all of the human elements of the problem, since it tries to make everything a simple binary choice.

We could ask it to take into consideration a list of human traits, but they are 'face value' traits to a machine. It does not have empathy, compassion, or morals.

Much like the people who are creating AI.