This post addresses the subject of the appropriate human data compression format. Though an interesting idea, I think that the proposed method is too low in resolution. You acknowledge the lossiness, but I think it’s just going to be too much.
Although the method you advocate is probably the best thing short of cryonics, I doubt there is any satisfactory compression method that can make anyone that’s more similar to you than a best friend or a family member who gets stuck with your memories. It’s better to have too much data than too much.
Because we share the same evolutionary past as all of our conspecifics, the biology and psychology of our brains is statistically the same. We each have our quirks of genetics and development, but even those are statistically similar among people who share our quirks. Thus with just a few bits of data we can already record most of what makes you who you are.
I’m not confident in this part. Although a large percentage of human biology and psychology are identical, the devil is in the details. From a statistical perspective, aren’t humans and chimps practical identical also? Percentage similarity of traits is probably the wrong metric, since small quantitative differences can have large qualitative impact.
Your idea of a generic human template, with various subtemplates for quirks, is also interesting, but still too low resolution.
Under what metric do we say that you and I have the “same” quirk, even if our phenotypes look superficially similar? How much data is discarded by the aggregation?
Even if we assume that the notion of a generic human template is meaningful, there are almost as many ways that people can deviate from the generic human template as there are people, and there would have to be that many quirky subtemplates. It’s possible that we could compress human phenotypic deviation into a lower number of templates than the number of people, but I don’t think we are anywhere near having a satisfactory way to do so. In the least, storing the deltas of human phenotypes might cut down on the data we all have in common.
The problem with lossy measures of phenotype such as memories and our current ability to measure quirky deviations from the average is that they discard too much information: the genotype, and other low-level aspects of phenotype.
Let’s start with the genotype problem. In the future, we synthesize a human with the same phenotype as our crude records of you (your memories, and your quirks according to the best psychometric indexes currently available). We will call this phenotype set X. Yet since multiple genotype can converge towards the same phenotype (particularly for crude measures of phenotype), the individual created is not guaranteed to have the same genotype as you, and probably won’t. Due to having a different genotype, this individual could end up having traits outside X that you didn’t have. They will have the same set of phenotypic traits as you that were recorded, but they may lack phenotypic traits that weren’t recorded (because you method of recording discarded the data), and they may have extra phenotypic traits that you didn’t have because keeping those traits out wasn’t in the spec.
Fundamentally, I think it’s problematic to try to reverse-engineer a mind based on a record of its outputs. That seems like trying to reverse-engineer a computer and its operating system based on a recording of everything it displayed on its monitor while being used. Even if you know how computers and operating systems work in general, you will still be unable to eliminate lots of guesswork in choosing between multiple constructions that would lead to the same output.
If you know a system’s construction, you may be able to predict its outputs, but knowing the outputs of a system doesn’t necessarily allow you to work backwards to its construction, even when you know the principles by which systems of that type are constructed.
I think the best reconstruction you will get through this method won’t be substantially more similar to you than taking your best friend (who has highly similar personality and interests) and implanting him with your memories. At best, you would get someone similar enough to be a relative of yours.
We really need to preserve your genotype; otherwise, future scientists could make an individual with all your memories and crudely measured personality traits, but a different personality and value system (in ways that weren’t measured), who wakes up and wonders what they heck you were doing all your life. We would need a solution that has all the phenotypic traits recorded for you, with the constraint that the individual created has the same genotype as you.
Yet even such a solution would still be discarding too much information about biological traits that influence your phenotype yet are not recorded in your genetic code, such as your prenatal development. It’s been shown that prenatal factors influence your brain structure, personality, and interests. So we need to record your prenatal environment to be able to create a meaningful facsimile of you. Otherwise, we could end constructing someone with the same genotype, memories, and psychometric measures as you, who nevertheless had a different brain structure; such a person would probably be less like you than a twin of yours who was implanted with your memories, because your twin shares a similar prenatal environment to yours, while your copy does not. A different brain structure would create a similar problem to having a different genotype: a different brain that has the same recorded phenotype as you will differ from you in unrecorded aspects of phenotype.
I would worry that a record of every single thought and behavior you have, both from yourself and observers, would still not be enough to reverse-engineer “you” in any meaningful way. Adding the constraint of matching your genotype would help, but we still don’t have a way to compress biological factors other than genotype, such as prenatal development. We have no MP3 format for humans.
The best record of your prenatal development that we have is your body and brain structure, so these would have to be stored along with your memories. Preferably at a rather cold temperature.
How good a record is good enough? In truth, I don’t think we even know enough to get the order of magnitude right. The best I can offer is that you need to record as much as you are willing to. The more you record, the more there will be to work with, and the less chance there will be of insufficient data. It may turn out that you simply can’t record enough to create a good restoration of a person from writing, but this is little different from the risk in cryonics of not being well preserved enough to restore despite best efforts. If you’re willing to take the risk that cryonics won’t work as well as you hope, you should be willing to accept that writing yourself into the future might not work as well as you hope.
I think you’re equating the relative risks here. If I wrote about myself for an hour a day for the rest of my life, I would rate chances very low that I could be reconstructed in contrast to cryonics. Your chances of being reconstructed increase with the amount of information present. One of the arguments for the safety of vitrifying the brain is that because the brain has a lot of redundancy (structure), we might be able to reconstruct damage.
I worked for a while in cryptography, where we try to recover original data wholly or partially from encrypted data. Based on those experiences and talking with Peter de Blanc, I looked into this reconstruction problem for a couple of hours at one point. Off the top of my head here’s some tips which might make it easier for me to reconstruct your body, brain and memories if I’m alive:
Record lots of data. Speech is better than writing. Include at least one photo. Videoblogging should record at a ate of least ten to a hundred times faster than writing, and if storage stays as cheap as it now it’ll survive.
Freeze a DNA sample (cheap, but riskier) or record one (expensive because you have to scan it, but less likely to be destroyed). This should allow one to reconstruct a physical twin at minimum.
I’m going to stick my neck out here and say don’t censor yourself if you’re recording audio. If I want to reconstruct your brain, the first step is probably to reverse-engineer your thoughts, so the more freely you talk the easier to deduce how you arrive at what you say. For example, I would learn more about someone if I watch them solving a cryptogram for ten seconds, than if they just gave me the answer. I have no idea if free association is equally as likely to work as essays. In general, the closer to the source the better in reverse-engineering: I’ve heard an estimate I buy that a minute of high-quality video could replace DNA in a pinch, but would still pay to freeze it, myself.
A friend of mine made a design for a $50 EEG cap, but the bandwidth is low enough that I doubt it’s worth the cash for most people. If you have money to spare it can’t hurt.
I wouldn’t guess I would be able to exactly reconstruct so that you’d notice no difference, which is my personal standard and why I’m not doing any of this myself.
Ciphergoth, I’d be interested in what you have to say about this.
At best, you would get someone similar enough to be a relative of yours.
Even if that’s all the better we can do, that’s much better than the nothing that will befall those who would otherwise have been totally lost because they didn’t sign up for cryonics.
that’s much better than the nothing that will befall those who would otherwise have been totally lost
I’m curious to know why you make this judgment. I imagine future people choosing between making a new person and making an as-similar-as-a-relative copy of a preserved person. In both cases, one additional person gets to exist. In both cases, that person is not somebody who has ever existed before. In neither case case does a future person get to revive a loved one, because the result will only be somebody similar to that loved one. Reviving the preserved person is better for the preserved person, I guess, but making a new person is better for the new person. Once you’ve lost continuity of identity, you’ve lost any reason why basing new people on recordings is better than making new people the old fashioned way.
Put another way, the nothing that will befall the totally lost feels exactly as bad to me as the nothing that will befall the future unborn whom they displace.
I know that ethical reasoning about potentially-existing people is hard, so I’m not too clear on this, so I’d like to know why you feel the way you do.
This post addresses the subject of the appropriate human data compression format. Though an interesting idea, I think that the proposed method is too low in resolution. You acknowledge the lossiness, but I think it’s just going to be too much.
Although the method you advocate is probably the best thing short of cryonics, I doubt there is any satisfactory compression method that can make anyone that’s more similar to you than a best friend or a family member who gets stuck with your memories. It’s better to have too much data than too much.
I’m not confident in this part. Although a large percentage of human biology and psychology are identical, the devil is in the details. From a statistical perspective, aren’t humans and chimps practical identical also? Percentage similarity of traits is probably the wrong metric, since small quantitative differences can have large qualitative impact.
Your idea of a generic human template, with various subtemplates for quirks, is also interesting, but still too low resolution.
Under what metric do we say that you and I have the “same” quirk, even if our phenotypes look superficially similar? How much data is discarded by the aggregation?
Even if we assume that the notion of a generic human template is meaningful, there are almost as many ways that people can deviate from the generic human template as there are people, and there would have to be that many quirky subtemplates. It’s possible that we could compress human phenotypic deviation into a lower number of templates than the number of people, but I don’t think we are anywhere near having a satisfactory way to do so. In the least, storing the deltas of human phenotypes might cut down on the data we all have in common.
The problem with lossy measures of phenotype such as memories and our current ability to measure quirky deviations from the average is that they discard too much information: the genotype, and other low-level aspects of phenotype.
Let’s start with the genotype problem. In the future, we synthesize a human with the same phenotype as our crude records of you (your memories, and your quirks according to the best psychometric indexes currently available). We will call this phenotype set X. Yet since multiple genotype can converge towards the same phenotype (particularly for crude measures of phenotype), the individual created is not guaranteed to have the same genotype as you, and probably won’t. Due to having a different genotype, this individual could end up having traits outside X that you didn’t have. They will have the same set of phenotypic traits as you that were recorded, but they may lack phenotypic traits that weren’t recorded (because you method of recording discarded the data), and they may have extra phenotypic traits that you didn’t have because keeping those traits out wasn’t in the spec.
Fundamentally, I think it’s problematic to try to reverse-engineer a mind based on a record of its outputs. That seems like trying to reverse-engineer a computer and its operating system based on a recording of everything it displayed on its monitor while being used. Even if you know how computers and operating systems work in general, you will still be unable to eliminate lots of guesswork in choosing between multiple constructions that would lead to the same output.
If you know a system’s construction, you may be able to predict its outputs, but knowing the outputs of a system doesn’t necessarily allow you to work backwards to its construction, even when you know the principles by which systems of that type are constructed.
I think the best reconstruction you will get through this method won’t be substantially more similar to you than taking your best friend (who has highly similar personality and interests) and implanting him with your memories. At best, you would get someone similar enough to be a relative of yours.
We really need to preserve your genotype; otherwise, future scientists could make an individual with all your memories and crudely measured personality traits, but a different personality and value system (in ways that weren’t measured), who wakes up and wonders what they heck you were doing all your life. We would need a solution that has all the phenotypic traits recorded for you, with the constraint that the individual created has the same genotype as you.
Yet even such a solution would still be discarding too much information about biological traits that influence your phenotype yet are not recorded in your genetic code, such as your prenatal development. It’s been shown that prenatal factors influence your brain structure, personality, and interests. So we need to record your prenatal environment to be able to create a meaningful facsimile of you. Otherwise, we could end constructing someone with the same genotype, memories, and psychometric measures as you, who nevertheless had a different brain structure; such a person would probably be less like you than a twin of yours who was implanted with your memories, because your twin shares a similar prenatal environment to yours, while your copy does not. A different brain structure would create a similar problem to having a different genotype: a different brain that has the same recorded phenotype as you will differ from you in unrecorded aspects of phenotype.
I would worry that a record of every single thought and behavior you have, both from yourself and observers, would still not be enough to reverse-engineer “you” in any meaningful way. Adding the constraint of matching your genotype would help, but we still don’t have a way to compress biological factors other than genotype, such as prenatal development. We have no MP3 format for humans.
The best record of your prenatal development that we have is your body and brain structure, so these would have to be stored along with your memories. Preferably at a rather cold temperature.
I think you’re equating the relative risks here. If I wrote about myself for an hour a day for the rest of my life, I would rate chances very low that I could be reconstructed in contrast to cryonics. Your chances of being reconstructed increase with the amount of information present. One of the arguments for the safety of vitrifying the brain is that because the brain has a lot of redundancy (structure), we might be able to reconstruct damage.
I worked for a while in cryptography, where we try to recover original data wholly or partially from encrypted data. Based on those experiences and talking with Peter de Blanc, I looked into this reconstruction problem for a couple of hours at one point. Off the top of my head here’s some tips which might make it easier for me to reconstruct your body, brain and memories if I’m alive:
Record lots of data. Speech is better than writing. Include at least one photo. Videoblogging should record at a ate of least ten to a hundred times faster than writing, and if storage stays as cheap as it now it’ll survive. Freeze a DNA sample (cheap, but riskier) or record one (expensive because you have to scan it, but less likely to be destroyed). This should allow one to reconstruct a physical twin at minimum. I’m going to stick my neck out here and say don’t censor yourself if you’re recording audio. If I want to reconstruct your brain, the first step is probably to reverse-engineer your thoughts, so the more freely you talk the easier to deduce how you arrive at what you say. For example, I would learn more about someone if I watch them solving a cryptogram for ten seconds, than if they just gave me the answer. I have no idea if free association is equally as likely to work as essays. In general, the closer to the source the better in reverse-engineering: I’ve heard an estimate I buy that a minute of high-quality video could replace DNA in a pinch, but would still pay to freeze it, myself.
A friend of mine made a design for a $50 EEG cap, but the bandwidth is low enough that I doubt it’s worth the cash for most people. If you have money to spare it can’t hurt.
I wouldn’t guess I would be able to exactly reconstruct so that you’d notice no difference, which is my personal standard and why I’m not doing any of this myself.
Ciphergoth, I’d be interested in what you have to say about this.
Even if that’s all the better we can do, that’s much better than the nothing that will befall those who would otherwise have been totally lost because they didn’t sign up for cryonics.
I’m curious to know why you make this judgment. I imagine future people choosing between making a new person and making an as-similar-as-a-relative copy of a preserved person. In both cases, one additional person gets to exist. In both cases, that person is not somebody who has ever existed before. In neither case case does a future person get to revive a loved one, because the result will only be somebody similar to that loved one. Reviving the preserved person is better for the preserved person, I guess, but making a new person is better for the new person. Once you’ve lost continuity of identity, you’ve lost any reason why basing new people on recordings is better than making new people the old fashioned way.
Put another way, the nothing that will befall the totally lost feels exactly as bad to me as the nothing that will befall the future unborn whom they displace.
I know that ethical reasoning about potentially-existing people is hard, so I’m not too clear on this, so I’d like to know why you feel the way you do.