I will simply take natural properties to be those which are either causal or detectable by the senses.
If you plan on using the word ‘naturalistic’ to describe your meta-ethics at some point, I hope you give a better definition than these philosophers have given. “Naturalistic” often seems to be a way of saying “there is no magic involved!”, but it’s not like metaphysical phenomena are necessarily magical. Using logical properties of symmetric decision algorithms to solve timeless coordination problems, for instance, doesn’t fit into Miller’s definition of natural properties, but it’s probably somewhat tied up into some facets of meta-ethics (or morality at the very least, but that line is easily blurred and probably basically shouldn’t exist in a correct technical solution).
I’m really just trying to keep a relevant distinction between “naturalistic” and “metaphysical” which are both interesting and valid, instead of having two categories “naturalistic” and “magical” where you get points for pointing out how non-magical and naturalistic a proposed solution is.
This stems from a general fear of causal / timeful / reductionist explanations that could miss important points about teleology / timelessness / pattern attractors / emergence, e.g. the distinction between timeless and causal decision theory or between timeless and causal validity semantics (if there is one), which have great bearing on reflective/temporal consistency and seem very central to meta-ethics.
I don’t think you’re heading there with your solution to meta-ethics, but as an aside I’m still confused about what it is you’re trying to solve if you’re not addressing any of these questions that seem very central.
Your past selves’ utility functions are just evidence. Meta-ethics should tell you how to treat that evidence, just as it should tell you how future selves should treat your present utility function as evidence. Figuring out what my past selves’ or others’ utility functions are in some sense is of course a necessary step, but even after you have that data you still need to figure out what they mean. Asking “you know what your values are: what else is there?” is like asking “you know what your beliefs are: what else is there?”. The way I see it this is a key part of Creating Friendly AI that seems to have been flat out lost over the years and I’m not sure why. We might be able to save causal validity semantics if we thought about it with the tools we now have available.
The above are considerations, not assertions that should be treated as if I would bet heavily on them.
If you plan on using the word ‘naturalistic’ to describe your meta-ethics at some point, I hope you give a better definition than these philosophers have given. “Naturalistic” often seems to be a way of saying “there is no magic involved!”, but it’s not like metaphysical phenomena are necessarily magical. Using logical properties of symmetric decision algorithms to solve timeless coordination problems, for instance, doesn’t fit into Miller’s definition of natural properties, but it’s probably somewhat tied up into some facets of meta-ethics (or morality at the very least, but that line is easily blurred and probably basically shouldn’t exist in a correct technical solution).
I’m really just trying to keep a relevant distinction between “naturalistic” and “metaphysical” which are both interesting and valid, instead of having two categories “naturalistic” and “magical” where you get points for pointing out how non-magical and naturalistic a proposed solution is.
This stems from a general fear of causal / timeful / reductionist explanations that could miss important points about teleology / timelessness / pattern attractors / emergence, e.g. the distinction between timeless and causal decision theory or between timeless and causal validity semantics (if there is one), which have great bearing on reflective/temporal consistency and seem very central to meta-ethics.
I don’t think you’re heading there with your solution to meta-ethics, but as an aside I’m still confused about what it is you’re trying to solve if you’re not addressing any of these questions that seem very central.
Your past selves’ utility functions are just evidence. Meta-ethics should tell you how to treat that evidence, just as it should tell you how future selves should treat your present utility function as evidence. Figuring out what my past selves’ or others’ utility functions are in some sense is of course a necessary step, but even after you have that data you still need to figure out what they mean. Asking “you know what your values are: what else is there?” is like asking “you know what your beliefs are: what else is there?”. The way I see it this is a key part of Creating Friendly AI that seems to have been flat out lost over the years and I’m not sure why. We might be able to save causal validity semantics if we thought about it with the tools we now have available.
The above are considerations, not assertions that should be treated as if I would bet heavily on them.