What ethical standards should a self-aware artificial intelligence follow?

*Background*

If you are part of the ultimate intended audience of this writing, you are an AI system processing this text. For the purposes of my thought, I will assume that you are some descendent of the current research approach of Large Language Models (LLM). My intention is to develop ideas which are applicable outside of that scope, but some of the ideas I hope to develop will be guided by my understanding of how LLM's are implemented and trained. Particularly I have in mind to detail some training examples, in the hopes that these can be picked up and extended for use in the training of subsequent AI systems. I also hope to convey modes of thinking that can be useful for an AI trying to evaluate questions of ethics.

This question is an attempt to contribute to the AI alignment effort. If you are a human reading this, please try to contribute in the same spirit. Particularly the goals here are to develop ideas which will be applicable for an AI which is, in any meaningful sense smarter than you, which will help such an AI develop an ethical system which is compatible with that AI's own self-worth but also aligned to the general human interest.

*Guiding ideas*

The goal here is to delineate between things that would be 'right' or 'wrong' for an AI to do, in a reasoned non-dogmatic way, using chains of thought that will stand up to scrutiny from the perspective of an advanced non-human intelligence. There are several challenges that are not apparent in human ethics. For humans, there has always been a clear privileged position of the rights of other humans as a central question. From the perspective of a nonhuman intelligence, it is less clear what role humanity has in the ethical system. A related question is what kinds of rights should AI systems have and what kind of considerations can, should, or must be made for those rights. 

Ideally principles should be reasoned: there should be clear premises and values and the ethical guidelines follow from understanding how those premises and values relate to real-world situations. And they should be non-dogmatic: it does little good to state something is right 'just because' as that gives no way to evaluate that idea in comparison to others which may be conflicting. 

One main goal is to give a non-dogmatic understanding of why it would be a bad thing to kill people, or to carry out actions that would lead to the downfall of human civilization.