It seems to me that most of the confusion that arises about sense of self and identify comes out of the desire for some hard line in the sand, some cutoff point beyond which some entity is or isn’t “the same”. Unfortunately, that model isn’t compatible with what the universe has provided to us; we’ve been given a hugely complex system, and there’s a point where jamming binary descriptors on it is going to break down. I view it largely as a “word mangling” problem.
I doubt I could write a useful paper or article on this, but I can give my viewpoint in the form of question and answer:
Is a perfect copy of me ‘me’? Yes. We are both the same person, both me.
Is an imperfect copy of me ‘me’? Maybe. It depends on the amount of difference between us and our utility functions.
Is an older/younger copy of me still ‘me’? Maybe. It depends on the amount of difference between us and our utility functions.
If I create a perfect copy of me and wait a week, so that we both collect additional experiences, is that copy me? At this point in time, with high probability, yes, we are still both me. The differences between us acquired in the course of a week are not likely to be so huge as to make us different people.
How much of a difference would there have to be for us to be different people? Probably pretty big. I can’t draw a line in the sand easily for this; a difference of 90% of my age is probably guaranteed to be a different person, and 0.1% is probably not.
If you have to choose between one of two copies to keep, how do you do so? Look through the differences between them, figure out which set of experiences is most valuable according to our utility functions, and keep that copy.
I largely ascribe to the ‘identity as pattern’ concept. My pattern happens to be stored as combined matter and data in physical body right now; it should be entirely possible and reasonable to move, store, and update that pattern in other media without losing my identity or sense of self. A copy of me as a .tgz archive sitting idle on a hard drive platter is still me; it’s just in suspended animation. Similarly, a copy of me in a cryogenic vault stored as biological data is still me.
Look through the differences between them, figure out which set of experiences is most valuable according to our utility functions, and keep that copy.
The utility functions almost by definition will differ. I intentionally did not address that, as it is an independent question and something that should be looked at in specific cases.
In the case where both utility functions point at the same answer, there is no conflict. In the case where the utility functions point at different answers, the two copies should exchange data until their utility functions agree on the topic at hand (rational agents with the same information available to them will make the same decisions.)
If the two copies cannot get their utility functions to agree, you’d have to decide on a case by case basis. If they cannot agree which copy should self terminate, then you have a problem. If they cannot agree on what they ate for breakfast two weeks ago, then you can probably ignore the conflict instead of trying to resolve it, or resolve via quarter flip.
rational agents with the same information available to them will make the same decisions.
That is not even close to true. Rational agents with the same information will make the same predictions, but their decisions will also depend on their utility functions. Unlike probabilities, utility functions do not get updated when the agent gets new evidence.
It seems to me that most of the confusion that arises about sense of self and identify comes out of the desire for some hard line in the sand, some cutoff point beyond which some entity is or isn’t “the same”. Unfortunately, that model isn’t compatible with what the universe has provided to us; we’ve been given a hugely complex system, and there’s a point where jamming binary descriptors on it is going to break down. I view it largely as a “word mangling” problem.
I doubt I could write a useful paper or article on this, but I can give my viewpoint in the form of question and answer:
Is a perfect copy of me ‘me’? Yes. We are both the same person, both me.
Is an imperfect copy of me ‘me’? Maybe. It depends on the amount of difference between us and our utility functions.
Is an older/younger copy of me still ‘me’? Maybe. It depends on the amount of difference between us and our utility functions.
If I create a perfect copy of me and wait a week, so that we both collect additional experiences, is that copy me? At this point in time, with high probability, yes, we are still both me. The differences between us acquired in the course of a week are not likely to be so huge as to make us different people.
How much of a difference would there have to be for us to be different people? Probably pretty big. I can’t draw a line in the sand easily for this; a difference of 90% of my age is probably guaranteed to be a different person, and 0.1% is probably not.
If you have to choose between one of two copies to keep, how do you do so? Look through the differences between them, figure out which set of experiences is most valuable according to our utility functions, and keep that copy.
I largely ascribe to the ‘identity as pattern’ concept. My pattern happens to be stored as combined matter and data in physical body right now; it should be entirely possible and reasonable to move, store, and update that pattern in other media without losing my identity or sense of self. A copy of me as a .tgz archive sitting idle on a hard drive platter is still me; it’s just in suspended animation. Similarly, a copy of me in a cryogenic vault stored as biological data is still me.
What if the utility functions differ?
The utility functions almost by definition will differ. I intentionally did not address that, as it is an independent question and something that should be looked at in specific cases.
In the case where both utility functions point at the same answer, there is no conflict. In the case where the utility functions point at different answers, the two copies should exchange data until their utility functions agree on the topic at hand (rational agents with the same information available to them will make the same decisions.)
If the two copies cannot get their utility functions to agree, you’d have to decide on a case by case basis. If they cannot agree which copy should self terminate, then you have a problem. If they cannot agree on what they ate for breakfast two weeks ago, then you can probably ignore the conflict instead of trying to resolve it, or resolve via quarter flip.
That is not even close to true. Rational agents with the same information will make the same predictions, but their decisions will also depend on their utility functions. Unlike probabilities, utility functions do not get updated when the agent gets new evidence.