I don’t think the important thing here is continuity. After all, a person can ‘continuously die’ from dementia or ‘discontinuously survive’ after brain surgery to remove a tumour. Surely what matters is the persistence of the information that ‘makes you who you are’ in some conscious mind somewhere.
The view which many people seem to hold but I regard as ‘obviously wrong’ is the one which believes in a thread of subjective identity, irreducible to the functional activity and information content of your mind, which might be ‘cut’ if you do something drastic like change substrate. That even if you (somehow) knew for sure that your copy would structurally isomorphic to and “Turing-test indistinguishable” from you, there would still an epiphenomenal ‘extra fact’ about whether the copy is ‘really you’.
The more interesting part of my statement is that psychological continuity is not important (per se) to substrate-independent self-modifying agents.
I think what underlies that ‘obviously wrong’ view in many cases is a recognition of the fact that in practice, we rely on continuity to establish identity.
A great many optical illusions and magic tricks depend on this: if entity A is here at T1 and entity B at T2, and I don’t notice any T1/T2 discontinuity, I’m likely to behave as though the same entity had been here throughout.
Of course, generalizing from those intuitions to a more fundamental notion of some kind of epiphenomenal identity is unjustified, as you say.
Then again, claiming that what makes me who I am is functional activity or information content is problematic, also. It isn’t clear, for example, that amnesia or brain damage makes me somebody else. Nor is it clear that if someone else is able to emulate me well enough to pass the equivalent of a Turing test, that developing that skill makes them me.
Mostly, I think is a composite notion, like . That is, we judge that identity is preserved by evaluating a close-enough match along many different axes, and there’s no single property or set of properties that is both necessary and sufficient to establish identity.
Then again, claiming that what makes me who I am is functional activity or information content is problematic, also. It isn’t clear, for example, that amnesia or brain damage makes me somebody else.
In case you thought I was implying that, let me clarify that the whole point is to deprecate binary oppositions such as “being someone else” vs “being the same person” and “still being oneself” vs “no longer existing”.
So of course it’s “not clear” that, say, frontal lobe damage leaves you the same person and it’s “not clear” that it leaves you as a different person.
Nor is it clear that if someone else is able to emulate me well enough to pass the equivalent of a Turing test, that developing that skill makes them me.
Only if you’re talking about identity in the loose, everyday sense, which like ‘furniture’ is a mishmash of many concepts. On the other hand, if you’re talking about whether the mental state of the copy is qualitatively identical to your own (or ‘as similar as makes no difference’), then I don’t think it’s remotely problematic to say that structural and functional isomorphism (or ‘as near as makes no difference’) guarantees this. Do you?
(This just boils down to “aren’t you a functionalist?”)
I think “as near as makes no difference” is not sufficiently well defined for the Turing-test-equivalent scenario I’m quoting. The question of “makes no difference to whom?” becomes important.
This is a problem for the traditional Turing test, as well… a great deal depends on who the auditor is; some people turn out to be surprisingly undiscriminating. (Or, well, it surprises me.)
But yes, if I don’t take the Turing test bit that I quoted literally, and instead think more abstractly about a sufficiently precise and reliable functional test, then I agree with you.
Actually, I don’t consider structural isomorphism necessary in and of itself; functional isomorphism is adequate for my purposes. (Though that said, I do think that an adequately functionally isomorphic system will tend to demonstrate a high level of structural isomorphism as well, although that’s not a well-thought-through assertion and my confidence in it is low).
I’m just not sure what such a test might comprise in practice. That is, if I’m in charge of QA for Upload, Inc. and it’s my job to make sure that the uploads we produce are adequately functionally isomorphic to the minds of the organic originals to avoid later lawsuits, it’s really not clear to me what tests I ought to be performing to ensure that.
I don’t think the important thing here is continuity. After all, a person can ‘continuously die’ from dementia or ‘discontinuously survive’ after brain surgery to remove a tumour. Surely what matters is the persistence of the information that ‘makes you who you are’ in some conscious mind somewhere.
The view which many people seem to hold but I regard as ‘obviously wrong’ is the one which believes in a thread of subjective identity, irreducible to the functional activity and information content of your mind, which might be ‘cut’ if you do something drastic like change substrate. That even if you (somehow) knew for sure that your copy would structurally isomorphic to and “Turing-test indistinguishable” from you, there would still an epiphenomenal ‘extra fact’ about whether the copy is ‘really you’.
That’s a good point.
I think what underlies that ‘obviously wrong’ view in many cases is a recognition of the fact that in practice, we rely on continuity to establish identity.
A great many optical illusions and magic tricks depend on this: if entity A is here at T1 and entity B at T2, and I don’t notice any T1/T2 discontinuity, I’m likely to behave as though the same entity had been here throughout.
Of course, generalizing from those intuitions to a more fundamental notion of some kind of epiphenomenal identity is unjustified, as you say.
Then again, claiming that what makes me who I am is functional activity or information content is problematic, also. It isn’t clear, for example, that amnesia or brain damage makes me somebody else. Nor is it clear that if someone else is able to emulate me well enough to pass the equivalent of a Turing test, that developing that skill makes them me.
Mostly, I think is a composite notion, like . That is, we judge that identity is preserved by evaluating a close-enough match along many different axes, and there’s no single property or set of properties that is both necessary and sufficient to establish identity.
I also don’t think it matters very much.
In case you thought I was implying that, let me clarify that the whole point is to deprecate binary oppositions such as “being someone else” vs “being the same person” and “still being oneself” vs “no longer existing”.
So of course it’s “not clear” that, say, frontal lobe damage leaves you the same person and it’s “not clear” that it leaves you as a different person.
Only if you’re talking about identity in the loose, everyday sense, which like ‘furniture’ is a mishmash of many concepts. On the other hand, if you’re talking about whether the mental state of the copy is qualitatively identical to your own (or ‘as similar as makes no difference’), then I don’t think it’s remotely problematic to say that structural and functional isomorphism (or ‘as near as makes no difference’) guarantees this. Do you?
(This just boils down to “aren’t you a functionalist?”)
I think “as near as makes no difference” is not sufficiently well defined for the Turing-test-equivalent scenario I’m quoting. The question of “makes no difference to whom?” becomes important.
This is a problem for the traditional Turing test, as well… a great deal depends on who the auditor is; some people turn out to be surprisingly undiscriminating. (Or, well, it surprises me.)
But yes, if I don’t take the Turing test bit that I quoted literally, and instead think more abstractly about a sufficiently precise and reliable functional test, then I agree with you.
Actually, I don’t consider structural isomorphism necessary in and of itself; functional isomorphism is adequate for my purposes. (Though that said, I do think that an adequately functionally isomorphic system will tend to demonstrate a high level of structural isomorphism as well, although that’s not a well-thought-through assertion and my confidence in it is low).
I’m just not sure what such a test might comprise in practice. That is, if I’m in charge of QA for Upload, Inc. and it’s my job to make sure that the uploads we produce are adequately functionally isomorphic to the minds of the organic originals to avoid later lawsuits, it’s really not clear to me what tests I ought to be performing to ensure that.