1. "Different societies have different truths" – no, they have different beliefs
2. A belief is not "true" when it matches reality – it is "pretty accurate". See also Examples of type error: a belief is a cons cell binding a proposition to a number between zero and one, never exactly zero and never exactly one, and certainly not a boolean. If your belief put 70% on something later shown real, the belief's accuracy was log₂(70%) = -0.51 bits. If your belief put 30%, you did worse: log₂(30%) = -1.73 bits. Accuracy is measured in negative numbers up to zero, never hitting exactly zero.

1. "I want my sister to live" – terminal value
2. "I want to administer CPR on her" – instrumental value

If #1 disappears, you will not continue wanting to do #2 for itself.

Yet, English uses the word "want" for both – leads to confusion in discussions about AI.

Take a thought experiment, where there's a species with a sex ratio of 10/90 for some reason – maybe some researchers edited the genes – that is, 10% of the time a sperm meets an egg, the result is male, and 90% of the time, the result is female.

Will it stay on 10/90 in the long run?

Well, what happens for an individual who makes a bit more males than females? That individual's genes will be more successful. It's the same old story. The only stable state is 50/50, which is where the ratio ends up after some number of generations. But why 50/50, not, say, 60% males 40% females?

I want examples. I think some species make a skewed adult sex ratio by eating some of their young? Or?