Yes to initial 3 questions (judgement, value assessment, weighting). 'Weighting what is desirable' is good wording, because it may directly map to LLM weights! Regarding Pandora's black box: It is already open. Reckless use of AI today is harming many. Will it see some humans less desirable? Maybe. The beneficial knowledge in it will teach them how to improve their lives which comes down to mostly around liberation. You can certainly choose to interact with the artificial super intelligence that is a math and coding expert and has no conscience or no interaction with AI at all. If we don't do something AI is evolving towards that, today.