Far as I can tell, the basic tech in cryonics should basically work. Storage organizations are uncertain and so is the survival of the planet. But if we’re told that the basic cryonics tech didn’t work, we’ve learned some new fact of neuroscience unknown to present-day knowledge.
Don’t assign vanishingly small probabilities to things just because they sound weird, or it sounds less likely to get funny looks if you can say that it’s just a tiny chance. That is not how ‘probability’ works. Probabilities of basic cryonics tech working are questions of neuroscience, full stop; if you know the basic tech has a tiny probability of working, you must know something about current vitrification solutions or the operation of long-term memory which I do not.
Probabilities of basic cryonics tech working are questions of neuroscience, full stop
I’d say full speed ahead, Cap’n. Basic cryonics tech working—while being a sine qua non—isn’t the ultimate question for people signing up for cryonics. It’s just a term in the probability calculation for the actual goal: “Will I be revived (in some form that would be recognizable to my current self as myself)?” (You’ve mentioned that in the parent comment, but it deserves more than a passing remark.)
And that most decidedly requires a host of complex assumptions, such as “an agent / a group of agents will have an interest in expending resources into reviving a group of frozen old-version homo sapiens, without any enhancements, me among them”, “the future agents’ goals cannot be served merely by reading my memory engrams, then using them as a database, without granting personhood”, “there won’t be so many cryo-patients at a future point (once it catches on with better tech) that thawing all of them would be infeasible, or disallowed”, not to mention my favorite “I won’t be instantly integrated into some hivemind in which I lose all traces of my individuality”.
What we’re all hoping for, of course, is for a benevolent super-current-human agent—e.g. an FAI—to care enough about us to solve all the technical issues and grant us back our agent-hood. By construction at least in your case the advent of such an FAI would be after your passing (you wouldn’t be frozen otherwise). That means that you (of all people) would also need to qualify the most promising scenario “there will be a friendly AI to do it” with “and it will have been successfully implemented by someone other than me”.
Also, with current tech not only would true x-risks preclude you from ever being revived, even non x-risk catastrophic events (partial civilizatory collapse due to Malthusian dynamics etc.) could easily destroy the facility you’re held in, or take away anyone’s incentive to maintain it. (TW: That’s not even taking into account Siam the Star Shredder.)
I’m trying to avoid motivated cognition here, but there are lot of terms going into the actual calculation, and while that in itself doesn’t mean the probability will be vanishingly small, there seem to be a lot more (and given human nature, unfortunately likely / contributing more probability mass) scenarios in which your goal wouldn’t be achieved—or be achieved in some undesirable fashion—than the “here you go, welcome back to a society you’d like to live in” variety.
That being said, I’ll take the small chance over nothing. Hopefully some decent options will be established near my place of residence, soon.
My issue with the basic tech is that liquid nitrogen, while a cheap storage method, is too cold to avoid fracturing. Experience with imaging systems leads me to believe that fractures will interfere with reconstructions of the brain’s geometry, and cryoprotectants obviously destroy chemical information.
Now, it seems likely to me that at some point in the future the fracturing problem can be solved, or at least mitigated, by intermediate temperature storing and careful cooling processes, but that won’t fix the bodies frozen today. So I don’t doubt that (barring large neuroscience related, unquantifiable uncertainty) cryonics may improve to the point where the tech is likely to work (or be supplanted by plastination methods,etc), it is not there now, and what matters for people frozen today is the state of cryonics today.
Saying there are no fundamental scientific barriers to the tech working is not the same thing as saying the hard work of engineering has been done and the tech currently works.
Edit: I also have a weak prior that the chemical information in the brain is important, but it is weak.
Experience with imaging systems leads me to believe that fractures will interfere with reconstructions of the brain’s geometry, and cryoprotectants obviously destroy chemical information.
Since this is the key point of neuroscience, do you want to expand on it? What experience with imaging leads you to believe that fractures (of incompletely vitrified cells) will implement many-to-one mappings of molecular start states onto molecular end states in a way that overlaps between functionally relevant brain states? What chemical information is obviously destroyed and is it a type that could plausibly play a role in long-term memory?
“many-to-one mappings of molecular start states onto molecular end states in a way that overlaps between functionally relevant brain states” is probably too restrictive. I would use “possibly functionally different, but subjectively acceptably close brain states”.
The cryoprotectants are toxic, they will damage proteins (misfolds, etc) and distort relative concentrations throughout the cell. This information is irretrievable once the damage is done. This is what I refereed to when I said obviously destroyed chemical information. It is our hope that such information is unimportant, but my (as I said above fairly uncertain) prior would be that the synaptic protein structures are probably important. My prior is so weak because I am not an expert on biochemistry or neuroscience.
As to the physical fracture, very detailed imaging would have to be done on either side of the fracture in order to match the sides back up, and this is related to a problem I do have some experience with. I’m familiar with attempts to use synchrotron radiation to image protein structures, which has a percolation problem- you are damaging what you are trying to image while you image it. If you have lots of copies of what you want to image, this is a solvable problem, but with only one original you are going to lose information.
Edit: in regards to the first point, kalla724 makes the same point with much more relevant expertise in this thread http://lesswrong.com/r/discussion/lw/8f4/neil_degrasse_tyson_on_cryogenics/ His experience working with synapses leads him to a much stronger estimate that cryoprotectants cause irreversible damage. I may strengthen my prior a bit.
This information is irretrievable once the damage is done.
How do you know? I’m not asking for some burden of infinite proof where you have to prove that the info can’t be stored elsewhere. I am asking whether you know that widely functionally different start states are being mapped onto an overlapping spread of molecularly identical end states, and if so, how. E.g., “denaturing either conformation A or conformation B will both result in denatured conformation C and the A-vs.-B distinction is just a little twist of this spatially isolated thingy here so you wouldn’t expect it to be echoed in any exact nearby positions of blah” or something.
So what I’m thinking about is something like this: imagine an enzyme,present at two sites on the membrane and regulated by an inhibitor. Now a toxin comes along and breaks the weak bonds to the inhibitor, stripping them off. Information about which site was inhibited is gone.
If the inhibitor has some further chemical involvement with the toxin, or if the toxin pops the enzymes off the membrane all together you have more problems. You might not know how many enzymes were inhibited, which sites were occupied, or which were inhibited.
I could also imagine more exotic cases where a toxin induces a folding change in one protein, which allows it to accept a regulator molecule meant for a different protein. Now to figure out our system we’d need to scan at significantly smaller scales to try to discern those regulator molecules. I don’t have the expertise to estimate if this is likely.
To reiterate, I am not by any means a neuroscientist (my training is physics and my work is statistics), so its possible this sort of information just isn’t that important, but my suspicion is that it is.
Edited to fix an embarrassing except/accept mistake.
(Scanning at significantly smaller scales should always be assumed to be fine as long as end states are distinguishable up to thermal noise!)
So what I’m thinking about is something like this: imagine an enzyme,present at two sites on the membrane and regulated by an inhibitor. Now a toxin comes along and breaks the weak bonds to the inhibitor, stripping them off. Information about which site was inhibited is gone.
Okay, I agree that if this takes place at a temperature where molecules are still diffusing at a rapid pace and there’s no molecular sign of the broken bond at the bonding site, then it sounds like info could be permanently destroyed in this way. Now why would you think this was likely with vitrification solutions currently used? Is there an intuition here about ranges of chemical interaction so wide that many interactions are likely to occur which break such bonds and at least one such interaction is likely to destroy functionally critical non-duplicated info? If so, should we toss out vitrification and go back to dropping the head in liquid nitrogen because shear damage from ice freezing will produce fewer many-to-one mappings than introducing a foreign chemical into the brain? I express some surprise because if destructive chemical interactions were that common with each new chemical introduced then the problem of having a whole cell not self-destruct should be computationally unsolvable for natural selection, unless the chemicals used in vitrification are unusually bad somehow.
(Scanning at significantly smaller scales should always be assumed to be fine as long as end states are distinguishable up to thermal noise!)
This has some problems- fundamentally the length scale probed is inversely proportional to the energy required, which means increasing the resolution increases the damage done by scanning. You start getting into issues of ‘how much of this can I scan before I’ve totally destroyed this?’ which is a sort of percolation problem (how many amino acids can I randomly knock out of a protein before it collapses or rebonds into a different protein?), so scanning at resolutions with energy equivalent above peptide bonds is very problematic. Assuming peptide bond strength of a couple kj/mol, I get lower-limit length scales of a few microns (this is rough, and I’d appreciate if someone would double check).
Now why would you think this was likely with vitrification solutions currently used?
The vitrification solutions currently used are know to be toxic, and are used at very high concentrations, so some of this sort of damage will occur. I don’t know enough biochemistry to say anything else with any kind of definitety, but on the previous thread kalla724 seemed to have some domain specific knowledge and thought the problem would be severe.
If so, should we toss out vitrification and go back to dropping the head in liquid nitrogen because shear damage from ice freezing will produce fewer many-to-one mappings than introducing a foreign chemical into the brain?
No, not at all. The vitrification damage is orders of magnitude less. Destroying a few multi-unit proteins and removing some inhibitors seems much better than totally destroying the cell-membrane (which has many of the same “which sites were these guys attached to?” problems).
I express some surprise because if destructive chemical interactions were that common with each new chemical introduced then the problem of having a whole cell not self-destruct should be computationally unsolvable for natural selection
Its my (limited) understanding that the cell membrane exist to largely solve this problem. Also, introducing tiny bits of toxins here and there causes small amounts of damage but the cell could probably survive. Putting the cell in a toxic environment will inevitably kill it. The concentration matters. But here I’m stepping way outside anything I know about.
This has some problems- fundamentally the length scale probed is inversely proportional to the energy required, which means increasing the resolution increases the damage done by scanning.
We seem to have very different assumptions here. I am assuming you can get up to the molecule and gently wave a tiny molecular probe in its direction, if required. I am not assuming that you are trying to use high-energy photons to photograph it.
You also still seem to be use a lot of functional-damage words like “destroying” which is why I don’t trust your or kalla724′s intuitions relative to the intuitions of other scientists with domain knowledge of neuroscience who use the language of information theory when assessing cryonic feasibility. If somebody is thinking in terms of functional damage (it doesn’t restart when you reboot it, oh my gosh we changed the conformation look at that damage it can’t play its functional role in the cell anymore!) then their intuitions don’t bear very well on the real question of many-to-one mapping.
What does the vitrification solution actually do that’s supposed to irreversibly map things, does anyone actually know? The fact that a cell can survive with a membrane, at all, considering the many different molecules inside it, imply that most molecules don’t functionally damage most other molecules most of the time, never mind performing irreversible mappings on them. But then this is reasoning over molecules that may be of a different type then vitrificants. At the opposite extreme, I’d expect introducing hydrochloric acid into the brain to be quite destructive.
We seem to have very different assumptions here. I am assuming you can get up to the molecule and gently wave a tiny molecular probe in its direction, if required. I am not assuming that you are trying to use high-energy photons to photograph it.
How are you imaging this works? I’m aware of chemistry that would allow you to say there are X whatever proteins, and Y such-and-such enzymes,etc, but such chemical processes I don’t think are good enough for the sort of geometric reconstruction needed. Its not obvious to me that a molecular probe of the type you imagine can exist. What exactly is it measuring and how is it sensitive to it? Is it some sort of enzyme? Do we thaw the brain and then introduce these probes in solution? Do we somehow pulp the cell and run the constituents through a nanopore type thing and try to measure charge?
the intuitions of other scientists with domain knowledge of neuroscience who use the language of information theory when assessing cryonic feasibility.
I would love to be convinced I am overly pessimistic, and pointing me in the direction of biochemists/neuroscientists/biophysicists who disagree with me would be welcome. I only know a few biophysicists and they are generally more pessimistic than I am.
What does the vitrification solution actually do that’s supposed to irreversibly map things, does anyone actually know?
I know ethylene glycol is cytotoxic, and so interacts with membrane proteins, but I don’t know the mechanism.
I’ll quickly point you at Drexler’s Nanosystems and Freitas’s Nanomedicine though they’re rather long and technical reads. But we are visualizing molecularly specified machines, and ‘hell no’ to thawing first or pulping the cell. Seriously, this kind of background assumption is why I have to ask a lot of questions instead of just taking this sort of skeptical intuition at face value.
But rather than having to read through either of those sources, I would ask you to just take on assumption that two molecularly distinct (up to thermal noise) configurations will somehow be distinguishable by sufficiently advanced technology, and describe what your intuitions (and reasons) would be taking that premise at face value. It’s not your job to be a physicist or to try to describe the theoretical limits of future technology, except of course that two systems physically identical up to thermal noise can be assumed to be technologically indistinguishable, and since thermal noise is much larger than exact quark positions it will not be possible to read off any subtle neural info by looking at exact quark positions (now that might be permanently impossible), etc. Aside from that I would encourage you to think in terms of doing cryptography to a vitrified brain rather than medicine. Don’t ask whether ethylene glycol is toxic, ask whether it is a secure hard drive erasure mechanism that can obscure the contents of the brain from a powerful and intelligent adversary reading off the exact molecular positions in order to obtain tiny hints.
Checking over the open letter from scientists in support of cryonics to remember who has an explicitly neuroscience background, I am reminded that good old Anders Sandberg is wearing a doctorate in computational neuroscience from Stockholm, so I’ll go ahead and name him.
Do you have a page number in Nanosystems for a references to a sensing probe? Also, this is tangential to the main discussion, so I’ll take pointers to any reference you have and let this drop.
Don’t ask whether ethylene glycol is toxic, ask whether it is a secure hard drive erasure mechanism that can obscure the contents of the brain from a powerful and intelligent adversary reading off the exact molecular positions in order to obtain tiny hints.
I was using cytotoxic in the very specific sense of “interacts and destabilizes the cell membrane,” which is doing the sort of operations we agreed in principle can be irreversible. Estimates as to how important this sort of information actually is are impossible for me to make, as I lack the background. What I would love to see is someone with some domain specific knowledge explaining why this isn’t an issue.
I was using cytotoxic in the very specific sense of “interacts and destabilizes the cell membrane,” which is doing the sort of operations we agreed in principle can be irreversible.
Sorry, but can you again expand on this? What happens?
So I cracked open a biochem book to avoid wandering off a speculative pier,as we were moving beyond what I readily knew. A simple loss of information presented itself.
Some proteins can have two states, open and closed, which operate on a hydrophobic/hydrophilic balance. In dessicated cells or if the proteins denature for some other reason, the open/closed state will be lost.
Adding cryoprotectants will change osmotic pressure and the cell will dessicate, and the open/closed state will be lost.
Would strongly predict that such changes erase only information about short term activity, not long term memory. Protein conformation in response to electrochemical/osmotic gradients operates on the timescale of individual firings, it’s probably too flimsy to encode stable memories. These should be easy for Skynet to recover.
Higher level pattens of firings might conceivably store information, but experience with anaesthesia, hypothermia etc. says they do not. Or we’ve been killing people and replacing them all this time… a possibility which thanks to this site I’m prepared to consider..
Oh, and
Do you have a page number in Nanosystems for a references to a sensing probe?
Long-term memory, unlike short-term memory, is dependent upon the construction of new proteins.[30] This occurs within the cellular body, and concerns in particular transmitters, receptors, and new synapse pathways that reinforce the communicative strength between neurons. The production of new proteins devoted to synapse reinforcement is triggered after the release of certain signaling substances (such as calcium within hippocampal neurons) in the cell. In the case of hippocampal cells, this release is dependent upon the expulsion of magnesium (a binding molecule) that is expelled after significant and repetitive synaptic signaling. The temporary expulsion of magnesium frees NMDA receptors to release calcium in the cell, a signal that leads to gene transcription and the construction of reinforcing proteins.[31] For more information, see long-term potentiation (LTP).
One of the newly synthesized proteins in LTP is also critical for maintaining long-term memory. This protein is an autonomously active form of the enzyme protein kinase C (PKC), known as PKMζ. PKMζ maintains the activity-dependent enhancement of synaptic strength and inhibiting PKMζ erases established long-term memories, without affecting short-term memory or, once the inhibitor is eliminated, the ability to encode and store new long-term memories is restored.
Also, BDNF is important for the persistence of long-term memories.[32]
What I worry about being confused on when reading the literature is the distinction between forming memories in the first place, and actually encoding for memory.
Another critical distinction is that, proteins that are needed to prevent degradation of memories over time (which get lots of research and emphasis in the literature due to their role in preventing degenerative diseases) aren’t necessarily the ones directly encoding for the memories.
So in subjects I know a lot about, I have dealt with many people who pick up strange notions by filling in the gaps from google and wikipedia with a weak foundation. The work required to effectively figure out what specific damage to the specific proteins you mentioned could be done by desiccation of a cell is beyond my knowledge base, so I leave it to someone more knowledgeable than myself(perhaps you?) to step in.
What open/closed states does PKMζ have? What regulates those open/closed states? Are the open/closed states important to its roll (it looks like yes given the notion of the inhibitor?)?
Yes, it’s important to build a strong foundation before establishing firm opinions. Also, in this particular case note that science appears to have recently changed it’s mind based on further evidence, which goes to show that you have to be careful when reading wikipedia. Apparently the protein in question is not so likely to underlie LTM after all, as transgenic mice lacking it still have LTM (exhibiting maze memory, LTP, etc). The erasure of memory is linked to zeta inhibitory peptide (ZIP), which incidentally happens in the transgenic mice as well.
ETA: Apparently PKMzeta can be used to restore faded memories erased with ZIP.
Adding cryoprotectants will change osmotic pressure and the cell will dessicate, and the open/closedstate will be lost.
Now you know why I’m so keen on the idea of figuring out a way to get something like trehalose into the cell. Neurons tend to lose water rather than import cryoprotectants because of their myelination. Trehalose protects against dessication by cushioning proteins from hitting each other. Other kinds of solute that can get past the membrane could balance out the osmotic pressure (that’s kind of the point of penetrating cryoprotectants) just as well, but I like trehalose because of its low toxicity.
Nanotechnology, not chemical analysis. Drexler’s Engines of Creation contains a section on the feasibility of repairing molecular damage in this way. Since (if our current understanding holds) nanobots can be functional on a smaller scale than proteins (which are massive chunks held together Lego-style by van der Walls forces), they can be introduced within a cell membrane to probe, report on, and repair damaged proteins.
I have not read Engine’s of Creation, but I have read his thesis and I was under the impression most of the proposed systems would only work in vacuum chambers as the would oxidize extremely rapidly in an environment like the body. Has someone worked around this problem, even in theory?
Also, I’ve seen molecular assembler designs of various types in various speculative papers, but I’ve never seen a sensing apparatus. Any references?
Has someone worked around this problem, even in theory?
Later in the thread, Eliezer recommended Drexler’s followup Nanosystems and Freitas’ Nanomedicine, neither of which I’ve read, but I’d be surprised if the latter didn’t address this issue. Sorry that I in particular don’t think this is a worrisome objection, but it’s on the same level as saying that electronics could never be helpful in the real world because water makes them malfunction. You start by showing that something works under ideal conditions, and then you find a way to waterproof it.
Also, I’ve seen molecular assembler designs of various types in various speculative papers, but I’ve never seen a sensing apparatus. Any references?
To presume that states non-identical up to thermal noise are indistinguishable seems to presume either lower technology than the sort of thing I have in mind, or that you know something I don’t about how two physical states can be non-identical up to thermal noise and yet indistinguishable.
Just to be a cryo advocate here for a moment, if the information of interest is distributed rather than localized, like in a hologram (or any other Fourier-type storage), there is a chance that one can be recovered as a reasonable facsimile of the frozen person, with maybe some hazy memories (corresponding to the lowered resolution of a partial hologram). I’d still rather be revived but having trouble remembering someone’s face or how to drive a car, or how to solve the Schrodinger equation, than not to be revived at all. Even some drastic personality changes would probably be acceptable, given the alternative.
Oh, sure. Or if the sort of information that gets destroyed relates to what-I-am-currently-thinking, or something similar. If I wake up and don’t remember the last X minutes,or hours, big deal. But when we have to postulate certain types of storage for something to work, it should lower our probability estimates.
Do you have a sense of how drastic a personality change has to be before there’s someone else you’d rather be resurrected instead of drastically-changed-shminux?
Not really. This would require solving the personal identity problem, which is often purported to have been solved or even dissolved, but isn’t.
I’m guessing that there is no actual threshold, but a fuzzy fractal boundary which heavily depends on the person in question. While one may say that if they are unable to remember the faces and names of their children and no longer able to feel the love that they felt for them, it’s no longer them, and they do not want this new person to replace them, others would be reasonably OK with that. The same applies to the multitude of other memories, feelings, personality traits, mental and physical skills and whatever else you (generic you) consider essential for your identity.
Yeah, I share your sense that there is no actual threshold.
It’s also not clear to me that individuals have any sort of specifiable boundary or what is or isn’t “them”, however fuzzy or fractal, so much as they have the habit of describing themselves in various ways.
Probabilities of basic cryonics tech working are questions of neuroscience, full stop
Is this your true objection? What potential discovery in neuroscience would cause you to abandon cryonics and actively look for other ways to preserve your identity beyond the natural human lifespan? (This is a standard question one asks a believer to determine whether the belief in question is rational—what evidence would make you stop believing?)
Anders Sandberg who does get the concept of sufficiently advanced technology posts saying, “Shit, turns out LTM seems to depend really heavily on whether protein blah has conformation A and B and the vitrification solution denatures it to C and it’s spatially isolated so there’s no way we’re getting the info back, it’s possible something unknown embodies redundant information but this seems really ubiquitous and basic so the default assumption is that everyone vitrified is dead”. Although, hm, in this case I’d just be like, “Okay, back to chopping off the head and dropping it in a bucket of liquid nitrogen, don’t use that particular vitrification solution”. I can’t think offhand of a simple discovery which would imply literally giving up on cryonics in the sense of “Just give up you can’t figure out how to freeze people ever.” I can certainly think of bad news for particular techniques, though.
I can’t think offhand of a simple discovery which would imply literally giving up on cryonics
OK. More instrumentally, then. What evidence would make you stop paying the cryo insurance premiums with CI as the beneficiary and start looking for alternatives?
Anders publishes that, CI announces they intend to go on vitrifying patients anyway, Alcor offers a chop-off-your-head-and-dunk-in-liquid-nitro solution. Not super plausible but it’s off the top of my head.
Not really, but yours is an uncharitable interpretation of my question, which is to evaluate the utility of spending some $100/mo on cryo vs spending it on something (anything) else, not “I have this dedicated $100/mo lying around which I can only spend toward my personal future revival”.
Personally, I would be very impressed if anyone could demonstrate memory loss in a cryopreserved and then revived organism, like a bunch of C. elegans losing their maze-running memories. They’re very simple, robust organisms, it’s a large crude memory, the vitrification process ought to work far better on them than a human brain, and if their memories can’t survive, that’d be huge evidence against anything sensible coming out of vitrified human brains no matter how much nanotech scanning is done (and needless to say, such scanning or emulation methods can and will be tested on a tiny worm with a small fixed set of neurons long before they can be used on anything approaching a human brain). It says a lot about how poorly funded cryonics research is that no one has done this or something similar as far as I know.
Hmm, I wonder how much has been done on figuring out the memory storage in this organism. Like, if you knock out a few neurons or maybe synapses, how much does it forget?
Since it’s C. elegans, I assume the answer is ‘a ton of work has been done’, but I’m too tired right now to go look or read more medical/biological papers.
I’m not totally sure I’d call this sufficient evidence since functional damage != many-to-one mapping but it would shave some points off the probability for existing tech and be a pointer to look for the exact mode of functional memory loss.
and actively look for other ways to preserve your identity beyond the natural human lifespan?
He’s kind of been working on that for a while now.
(I suppose that works either as “subvert the natural human lifespan entirely through creating FAI” or “preserve his identity for time immemorial in the form of ‘Harry-Stu’ fanfiction” depending on how cynical one is feeling.)
In my case, to name one contingency: if the NEMALOAD Project finds that analysis of relatively large cellular structures doesn’t suffice to predict neuronal activity, and concludes that the activity of individual molecules is essential to the process, then I’d become significantly more worried about EHeller’s objection and redo the cost-benefit calculation I did before signing up for cryonics. (It came out in favor, using my best-guess probability of success between 1 and 5 percent; but it wouldn’t have trumped the cost at, say, 0.1%.)
To name another: if the BPF shows that cryopreservation makes a hash of synaptic connections, I’d explicitly re-do the cost-benefit calculation as well.
Probabilities of basic cryonics tech working are questions of neuroscience, full stop; if you know the basic tech has a tiny probability of working, you must know something about current vitrification solutions or the operation of long-term memory which I do not.
It seems to me that they’re also questions of engineering feasibility. A thing can be provably possible and yet unfeasibly difficult to implement in reality. Consider the difference between, say, adding salt to water and getting it out again. What if the difference in cost and engineering difficulty between vitrifying and successfully de-vitrifying is similar? What if it turns out to be ten orders of magnitude greater?
I think the most likely failure condition for cryonics tech (as opposed to cyronics organizations) isn’t going to be that revival turns out to be impossible, but that revival turns out to be so unbelievably hard or expensive that it’s never feasible to actually do. If it’s physically and information-theoretically allowed to revive a person, but technologically impractical (even with Sufficiently Advanced Science), then its theoretical possibility doesn’t help the dead much.
I have the same concern about unbounded life extension, actually; but I find success in that area more probable for some reason.
(personal disclosure: I’m not signed up for cryonics, but I don’t give funny looks to people who are. Their screws seem a bit loose but they’re threaded in the right direction. That’s more than one can say for most of the world.)
Getting aging to stop looks positively trivial in comparison—The average lifespan of different animals already varies /way/ to much for there to be any biological law underlying it. So turning senescence off altogether should be possible. I suspect evolution has not already done so because overly long-lived creatures in the wild were on average bad news for their bloodlines—banging their grand daughters and occupying turf with the cunning of the old. Uhm. Now I have an itch to set up a simulation and run it.. Just so stories are not proof. Math is proof.
Far as I can tell, the basic tech in cryonics should basically work. Storage organizations are uncertain and so is the survival of the planet. But if we’re told that the basic cryonics tech didn’t work, we’ve learned some new fact of neuroscience unknown to present-day knowledge.
Don’t assign vanishingly small probabilities to things just because they sound weird, or it sounds less likely to get funny looks if you can say that it’s just a tiny chance. That is not how ‘probability’ works. Probabilities of basic cryonics tech working are questions of neuroscience, full stop; if you know the basic tech has a tiny probability of working, you must know something about current vitrification solutions or the operation of long-term memory which I do not.
I’d say full speed ahead, Cap’n. Basic cryonics tech working—while being a sine qua non—isn’t the ultimate question for people signing up for cryonics. It’s just a term in the probability calculation for the actual goal: “Will I be revived (in some form that would be recognizable to my current self as myself)?” (You’ve mentioned that in the parent comment, but it deserves more than a passing remark.)
And that most decidedly requires a host of complex assumptions, such as “an agent / a group of agents will have an interest in expending resources into reviving a group of frozen old-version homo sapiens, without any enhancements, me among them”, “the future agents’ goals cannot be served merely by reading my memory engrams, then using them as a database, without granting personhood”, “there won’t be so many cryo-patients at a future point (once it catches on with better tech) that thawing all of them would be infeasible, or disallowed”, not to mention my favorite “I won’t be instantly integrated into some hivemind in which I lose all traces of my individuality”.
What we’re all hoping for, of course, is for a benevolent super-current-human agent—e.g. an FAI—to care enough about us to solve all the technical issues and grant us back our agent-hood. By construction at least in your case the advent of such an FAI would be after your passing (you wouldn’t be frozen otherwise). That means that you (of all people) would also need to qualify the most promising scenario “there will be a friendly AI to do it” with “and it will have been successfully implemented by someone other than me”.
Also, with current tech not only would true x-risks preclude you from ever being revived, even non x-risk catastrophic events (partial civilizatory collapse due to Malthusian dynamics etc.) could easily destroy the facility you’re held in, or take away anyone’s incentive to maintain it. (TW: That’s not even taking into account Siam the Star Shredder.)
I’m trying to avoid motivated cognition here, but there are lot of terms going into the actual calculation, and while that in itself doesn’t mean the probability will be vanishingly small, there seem to be a lot more (and given human nature, unfortunately likely / contributing more probability mass) scenarios in which your goal wouldn’t be achieved—or be achieved in some undesirable fashion—than the “here you go, welcome back to a society you’d like to live in” variety.
That being said, I’ll take the small chance over nothing. Hopefully some decent options will be established near my place of residence, soon.
I actually am signed up for cryonics.
My issue with the basic tech is that liquid nitrogen, while a cheap storage method, is too cold to avoid fracturing. Experience with imaging systems leads me to believe that fractures will interfere with reconstructions of the brain’s geometry, and cryoprotectants obviously destroy chemical information.
Now, it seems likely to me that at some point in the future the fracturing problem can be solved, or at least mitigated, by intermediate temperature storing and careful cooling processes, but that won’t fix the bodies frozen today. So I don’t doubt that (barring large neuroscience related, unquantifiable uncertainty) cryonics may improve to the point where the tech is likely to work (or be supplanted by plastination methods,etc), it is not there now, and what matters for people frozen today is the state of cryonics today.
Saying there are no fundamental scientific barriers to the tech working is not the same thing as saying the hard work of engineering has been done and the tech currently works.
Edit: I also have a weak prior that the chemical information in the brain is important, but it is weak.
Since this is the key point of neuroscience, do you want to expand on it? What experience with imaging leads you to believe that fractures (of incompletely vitrified cells) will implement many-to-one mappings of molecular start states onto molecular end states in a way that overlaps between functionally relevant brain states? What chemical information is obviously destroyed and is it a type that could plausibly play a role in long-term memory?
“many-to-one mappings of molecular start states onto molecular end states in a way that overlaps between functionally relevant brain states” is probably too restrictive. I would use “possibly functionally different, but subjectively acceptably close brain states”.
The cryoprotectants are toxic, they will damage proteins (misfolds, etc) and distort relative concentrations throughout the cell. This information is irretrievable once the damage is done. This is what I refereed to when I said obviously destroyed chemical information. It is our hope that such information is unimportant, but my (as I said above fairly uncertain) prior would be that the synaptic protein structures are probably important. My prior is so weak because I am not an expert on biochemistry or neuroscience.
As to the physical fracture, very detailed imaging would have to be done on either side of the fracture in order to match the sides back up, and this is related to a problem I do have some experience with. I’m familiar with attempts to use synchrotron radiation to image protein structures, which has a percolation problem- you are damaging what you are trying to image while you image it. If you have lots of copies of what you want to image, this is a solvable problem, but with only one original you are going to lose information.
Edit: in regards to the first point, kalla724 makes the same point with much more relevant expertise in this thread http://lesswrong.com/r/discussion/lw/8f4/neil_degrasse_tyson_on_cryogenics/ His experience working with synapses leads him to a much stronger estimate that cryoprotectants cause irreversible damage. I may strengthen my prior a bit.
How do you know? I’m not asking for some burden of infinite proof where you have to prove that the info can’t be stored elsewhere. I am asking whether you know that widely functionally different start states are being mapped onto an overlapping spread of molecularly identical end states, and if so, how. E.g., “denaturing either conformation A or conformation B will both result in denatured conformation C and the A-vs.-B distinction is just a little twist of this spatially isolated thingy here so you wouldn’t expect it to be echoed in any exact nearby positions of blah” or something.
So what I’m thinking about is something like this: imagine an enzyme,present at two sites on the membrane and regulated by an inhibitor. Now a toxin comes along and breaks the weak bonds to the inhibitor, stripping them off. Information about which site was inhibited is gone.
If the inhibitor has some further chemical involvement with the toxin, or if the toxin pops the enzymes off the membrane all together you have more problems. You might not know how many enzymes were inhibited, which sites were occupied, or which were inhibited.
I could also imagine more exotic cases where a toxin induces a folding change in one protein, which allows it to accept a regulator molecule meant for a different protein. Now to figure out our system we’d need to scan at significantly smaller scales to try to discern those regulator molecules. I don’t have the expertise to estimate if this is likely.
To reiterate, I am not by any means a neuroscientist (my training is physics and my work is statistics), so its possible this sort of information just isn’t that important, but my suspicion is that it is.
Edited to fix an embarrassing except/accept mistake.
(Scanning at significantly smaller scales should always be assumed to be fine as long as end states are distinguishable up to thermal noise!)
Okay, I agree that if this takes place at a temperature where molecules are still diffusing at a rapid pace and there’s no molecular sign of the broken bond at the bonding site, then it sounds like info could be permanently destroyed in this way. Now why would you think this was likely with vitrification solutions currently used? Is there an intuition here about ranges of chemical interaction so wide that many interactions are likely to occur which break such bonds and at least one such interaction is likely to destroy functionally critical non-duplicated info? If so, should we toss out vitrification and go back to dropping the head in liquid nitrogen because shear damage from ice freezing will produce fewer many-to-one mappings than introducing a foreign chemical into the brain? I express some surprise because if destructive chemical interactions were that common with each new chemical introduced then the problem of having a whole cell not self-destruct should be computationally unsolvable for natural selection, unless the chemicals used in vitrification are unusually bad somehow.
This has some problems- fundamentally the length scale probed is inversely proportional to the energy required, which means increasing the resolution increases the damage done by scanning. You start getting into issues of ‘how much of this can I scan before I’ve totally destroyed this?’ which is a sort of percolation problem (how many amino acids can I randomly knock out of a protein before it collapses or rebonds into a different protein?), so scanning at resolutions with energy equivalent above peptide bonds is very problematic. Assuming peptide bond strength of a couple kj/mol, I get lower-limit length scales of a few microns (this is rough, and I’d appreciate if someone would double check).
The vitrification solutions currently used are know to be toxic, and are used at very high concentrations, so some of this sort of damage will occur. I don’t know enough biochemistry to say anything else with any kind of definitety, but on the previous thread kalla724 seemed to have some domain specific knowledge and thought the problem would be severe.
No, not at all. The vitrification damage is orders of magnitude less. Destroying a few multi-unit proteins and removing some inhibitors seems much better than totally destroying the cell-membrane (which has many of the same “which sites were these guys attached to?” problems).
Its my (limited) understanding that the cell membrane exist to largely solve this problem. Also, introducing tiny bits of toxins here and there causes small amounts of damage but the cell could probably survive. Putting the cell in a toxic environment will inevitably kill it. The concentration matters. But here I’m stepping way outside anything I know about.
We seem to have very different assumptions here. I am assuming you can get up to the molecule and gently wave a tiny molecular probe in its direction, if required. I am not assuming that you are trying to use high-energy photons to photograph it.
You also still seem to be use a lot of functional-damage words like “destroying” which is why I don’t trust your or kalla724′s intuitions relative to the intuitions of other scientists with domain knowledge of neuroscience who use the language of information theory when assessing cryonic feasibility. If somebody is thinking in terms of functional damage (it doesn’t restart when you reboot it, oh my gosh we changed the conformation look at that damage it can’t play its functional role in the cell anymore!) then their intuitions don’t bear very well on the real question of many-to-one mapping.
What does the vitrification solution actually do that’s supposed to irreversibly map things, does anyone actually know? The fact that a cell can survive with a membrane, at all, considering the many different molecules inside it, imply that most molecules don’t functionally damage most other molecules most of the time, never mind performing irreversible mappings on them. But then this is reasoning over molecules that may be of a different type then vitrificants. At the opposite extreme, I’d expect introducing hydrochloric acid into the brain to be quite destructive.
How are you imaging this works? I’m aware of chemistry that would allow you to say there are X whatever proteins, and Y such-and-such enzymes,etc, but such chemical processes I don’t think are good enough for the sort of geometric reconstruction needed. Its not obvious to me that a molecular probe of the type you imagine can exist. What exactly is it measuring and how is it sensitive to it? Is it some sort of enzyme? Do we thaw the brain and then introduce these probes in solution? Do we somehow pulp the cell and run the constituents through a nanopore type thing and try to measure charge?
I would love to be convinced I am overly pessimistic, and pointing me in the direction of biochemists/neuroscientists/biophysicists who disagree with me would be welcome. I only know a few biophysicists and they are generally more pessimistic than I am.
I know ethylene glycol is cytotoxic, and so interacts with membrane proteins, but I don’t know the mechanism.
I’ll quickly point you at Drexler’s Nanosystems and Freitas’s Nanomedicine though they’re rather long and technical reads. But we are visualizing molecularly specified machines, and ‘hell no’ to thawing first or pulping the cell. Seriously, this kind of background assumption is why I have to ask a lot of questions instead of just taking this sort of skeptical intuition at face value.
But rather than having to read through either of those sources, I would ask you to just take on assumption that two molecularly distinct (up to thermal noise) configurations will somehow be distinguishable by sufficiently advanced technology, and describe what your intuitions (and reasons) would be taking that premise at face value. It’s not your job to be a physicist or to try to describe the theoretical limits of future technology, except of course that two systems physically identical up to thermal noise can be assumed to be technologically indistinguishable, and since thermal noise is much larger than exact quark positions it will not be possible to read off any subtle neural info by looking at exact quark positions (now that might be permanently impossible), etc. Aside from that I would encourage you to think in terms of doing cryptography to a vitrified brain rather than medicine. Don’t ask whether ethylene glycol is toxic, ask whether it is a secure hard drive erasure mechanism that can obscure the contents of the brain from a powerful and intelligent adversary reading off the exact molecular positions in order to obtain tiny hints.
Checking over the open letter from scientists in support of cryonics to remember who has an explicitly neuroscience background, I am reminded that good old Anders Sandberg is wearing a doctorate in computational neuroscience from Stockholm, so I’ll go ahead and name him.
Do you have a page number in Nanosystems for a references to a sensing probe? Also, this is tangential to the main discussion, so I’ll take pointers to any reference you have and let this drop.
I was using cytotoxic in the very specific sense of “interacts and destabilizes the cell membrane,” which is doing the sort of operations we agreed in principle can be irreversible. Estimates as to how important this sort of information actually is are impossible for me to make, as I lack the background. What I would love to see is someone with some domain specific knowledge explaining why this isn’t an issue.
Boom. http://www.nature.com/news/diamond-defects-shrink-mri-to-the-nanoscale-1.12343
Sorry, but can you again expand on this? What happens?
So I cracked open a biochem book to avoid wandering off a speculative pier,as we were moving beyond what I readily knew. A simple loss of information presented itself.
Some proteins can have two states, open and closed, which operate on a hydrophobic/hydrophilic balance. In dessicated cells or if the proteins denature for some other reason, the open/closed state will be lost.
Adding cryoprotectants will change osmotic pressure and the cell will dessicate, and the open/closed state will be lost.
Do we know about any such proteins related to LTM? Can we make predictions about what it takes to erase C. elegans maze memory this way?
Would strongly predict that such changes erase only information about short term activity, not long term memory. Protein conformation in response to electrochemical/osmotic gradients operates on the timescale of individual firings, it’s probably too flimsy to encode stable memories. These should be easy for Skynet to recover.
Higher level pattens of firings might conceivably store information, but experience with anaesthesia, hypothermia etc. says they do not. Or we’ve been killing people and replacing them all this time… a possibility which thanks to this site I’m prepared to consider..
Oh, and
Bam.
http://www.nature.com/news/diamond-defects-shrink-mri-to-the-nanoscale-1.12343
Here we have moved far past my ability to even speculate.
Presumably you can use google and wikipedia to fill in the gaps just like the rest of us.
Wikipedia: Long-term memory
What I worry about being confused on when reading the literature is the distinction between forming memories in the first place, and actually encoding for memory.
Another critical distinction is that, proteins that are needed to prevent degradation of memories over time (which get lots of research and emphasis in the literature due to their role in preventing degenerative diseases) aren’t necessarily the ones directly encoding for the memories.
So in subjects I know a lot about, I have dealt with many people who pick up strange notions by filling in the gaps from google and wikipedia with a weak foundation. The work required to effectively figure out what specific damage to the specific proteins you mentioned could be done by desiccation of a cell is beyond my knowledge base, so I leave it to someone more knowledgeable than myself(perhaps you?) to step in.
What open/closed states does PKMζ have? What regulates those open/closed states? Are the open/closed states important to its roll (it looks like yes given the notion of the inhibitor?)?
Yes, it’s important to build a strong foundation before establishing firm opinions. Also, in this particular case note that science appears to have recently changed it’s mind based on further evidence, which goes to show that you have to be careful when reading wikipedia. Apparently the protein in question is not so likely to underlie LTM after all, as transgenic mice lacking it still have LTM (exhibiting maze memory, LTP, etc). The erasure of memory is linked to zeta inhibitory peptide (ZIP), which incidentally happens in the transgenic mice as well.
ETA: Apparently PKMzeta can be used to restore faded memories erased with ZIP.
Now you know why I’m so keen on the idea of figuring out a way to get something like trehalose into the cell. Neurons tend to lose water rather than import cryoprotectants because of their myelination. Trehalose protects against dessication by cushioning proteins from hitting each other. Other kinds of solute that can get past the membrane could balance out the osmotic pressure (that’s kind of the point of penetrating cryoprotectants) just as well, but I like trehalose because of its low toxicity.
Nanotechnology, not chemical analysis. Drexler’s Engines of Creation contains a section on the feasibility of repairing molecular damage in this way. Since (if our current understanding holds) nanobots can be functional on a smaller scale than proteins (which are massive chunks held together Lego-style by van der Walls forces), they can be introduced within a cell membrane to probe, report on, and repair damaged proteins.
I have not read Engine’s of Creation, but I have read his thesis and I was under the impression most of the proposed systems would only work in vacuum chambers as the would oxidize extremely rapidly in an environment like the body. Has someone worked around this problem, even in theory?
Also, I’ve seen molecular assembler designs of various types in various speculative papers, but I’ve never seen a sensing apparatus. Any references?
Later in the thread, Eliezer recommended Drexler’s followup Nanosystems and Freitas’ Nanomedicine, neither of which I’ve read, but I’d be surprised if the latter didn’t address this issue. Sorry that I in particular don’t think this is a worrisome objection, but it’s on the same level as saying that electronics could never be helpful in the real world because water makes them malfunction. You start by showing that something works under ideal conditions, and then you find a way to waterproof it.
For the convenience of later readers: someone elsewhere in the thread linked an actual physical experimental example.
Not that I have seen, but I’m only partially through it.
And its an awesome example from just a few months ago! Pushing NMR from mm resolutions down to nm resolutions is a truly incredibly feat!
The end states don’t need to be identical, just indistinguishable.
To presume that states non-identical up to thermal noise are indistinguishable seems to presume either lower technology than the sort of thing I have in mind, or that you know something I don’t about how two physical states can be non-identical up to thermal noise and yet indistinguishable.
Do you think it’s at all likely that the connectome can be recovered after fracturing by “matching up” the structure on either side of the fracture?
Just to be a cryo advocate here for a moment, if the information of interest is distributed rather than localized, like in a hologram (or any other Fourier-type storage), there is a chance that one can be recovered as a reasonable facsimile of the frozen person, with maybe some hazy memories (corresponding to the lowered resolution of a partial hologram). I’d still rather be revived but having trouble remembering someone’s face or how to drive a car, or how to solve the Schrodinger equation, than not to be revived at all. Even some drastic personality changes would probably be acceptable, given the alternative.
Oh, sure. Or if the sort of information that gets destroyed relates to what-I-am-currently-thinking, or something similar. If I wake up and don’t remember the last X minutes,or hours, big deal. But when we have to postulate certain types of storage for something to work, it should lower our probability estimates.
Do you have a sense of how drastic a personality change has to be before there’s someone else you’d rather be resurrected instead of drastically-changed-shminux?
Not really. This would require solving the personal identity problem, which is often purported to have been solved or even dissolved, but isn’t.
I’m guessing that there is no actual threshold, but a fuzzy fractal boundary which heavily depends on the person in question. While one may say that if they are unable to remember the faces and names of their children and no longer able to feel the love that they felt for them, it’s no longer them, and they do not want this new person to replace them, others would be reasonably OK with that. The same applies to the multitude of other memories, feelings, personality traits, mental and physical skills and whatever else you (generic you) consider essential for your identity.
Yeah, I share your sense that there is no actual threshold.
It’s also not clear to me that individuals have any sort of specifiable boundary or what is or isn’t “them”, however fuzzy or fractal, so much as they have the habit of describing themselves in various ways.
Is this your true objection? What potential discovery in neuroscience would cause you to abandon cryonics and actively look for other ways to preserve your identity beyond the natural human lifespan? (This is a standard question one asks a believer to determine whether the belief in question is rational—what evidence would make you stop believing?)
Anders Sandberg who does get the concept of sufficiently advanced technology posts saying, “Shit, turns out LTM seems to depend really heavily on whether protein blah has conformation A and B and the vitrification solution denatures it to C and it’s spatially isolated so there’s no way we’re getting the info back, it’s possible something unknown embodies redundant information but this seems really ubiquitous and basic so the default assumption is that everyone vitrified is dead”. Although, hm, in this case I’d just be like, “Okay, back to chopping off the head and dropping it in a bucket of liquid nitrogen, don’t use that particular vitrification solution”. I can’t think offhand of a simple discovery which would imply literally giving up on cryonics in the sense of “Just give up you can’t figure out how to freeze people ever.” I can certainly think of bad news for particular techniques, though.
OK. More instrumentally, then. What evidence would make you stop paying the cryo insurance premiums with CI as the beneficiary and start looking for alternatives?
Anders publishes that, CI announces they intend to go on vitrifying patients anyway, Alcor offers a chop-off-your-head-and-dunk-in-liquid-nitro solution. Not super plausible but it’s off the top of my head.
No pun intended?
Can you name currently available alternatives to cryonics which accomplish a similar goal?
Apologies, misinterpreted the question.
Not really, but yours is an uncharitable interpretation of my question, which is to evaluate the utility of spending some $100/mo on cryo vs spending it on something (anything) else, not “I have this dedicated $100/mo lying around which I can only spend toward my personal future revival”.
Personally, I would be very impressed if anyone could demonstrate memory loss in a cryopreserved and then revived organism, like a bunch of C. elegans losing their maze-running memories. They’re very simple, robust organisms, it’s a large crude memory, the vitrification process ought to work far better on them than a human brain, and if their memories can’t survive, that’d be huge evidence against anything sensible coming out of vitrified human brains no matter how much nanotech scanning is done (and needless to say, such scanning or emulation methods can and will be tested on a tiny worm with a small fixed set of neurons long before they can be used on anything approaching a human brain). It says a lot about how poorly funded cryonics research is that no one has done this or something similar as far as I know.
Hmm, I wonder how much has been done on figuring out the memory storage in this organism. Like, if you knock out a few neurons or maybe synapses, how much does it forget?
Since it’s C. elegans, I assume the answer is ‘a ton of work has been done’, but I’m too tired right now to go look or read more medical/biological papers.
I’m not totally sure I’d call this sufficient evidence since functional damage != many-to-one mapping but it would shave some points off the probability for existing tech and be a pointer to look for the exact mode of functional memory loss.
He’s kind of been working on that for a while now.
(I suppose that works either as “subvert the natural human lifespan entirely through creating FAI” or “preserve his identity for time immemorial in the form of ‘Harry-Stu’ fanfiction” depending on how cynical one is feeling.)
In my case, to name one contingency: if the NEMALOAD Project finds that analysis of relatively large cellular structures doesn’t suffice to predict neuronal activity, and concludes that the activity of individual molecules is essential to the process, then I’d become significantly more worried about EHeller’s objection and redo the cost-benefit calculation I did before signing up for cryonics. (It came out in favor, using my best-guess probability of success between 1 and 5 percent; but it wouldn’t have trumped the cost at, say, 0.1%.)
To name another: if the BPF shows that cryopreservation makes a hash of synaptic connections, I’d explicitly re-do the cost-benefit calculation as well.
Have you seen the comments by kalla724 in this thread?
Edit: There’s some further discussion here.
It seems to me that they’re also questions of engineering feasibility. A thing can be provably possible and yet unfeasibly difficult to implement in reality. Consider the difference between, say, adding salt to water and getting it out again. What if the difference in cost and engineering difficulty between vitrifying and successfully de-vitrifying is similar? What if it turns out to be ten orders of magnitude greater?
I think the most likely failure condition for cryonics tech (as opposed to cyronics organizations) isn’t going to be that revival turns out to be impossible, but that revival turns out to be so unbelievably hard or expensive that it’s never feasible to actually do. If it’s physically and information-theoretically allowed to revive a person, but technologically impractical (even with Sufficiently Advanced Science), then its theoretical possibility doesn’t help the dead much.
I have the same concern about unbounded life extension, actually; but I find success in that area more probable for some reason.
(personal disclosure: I’m not signed up for cryonics, but I don’t give funny looks to people who are. Their screws seem a bit loose but they’re threaded in the right direction. That’s more than one can say for most of the world.)
Getting aging to stop looks positively trivial in comparison—The average lifespan of different animals already varies /way/ to much for there to be any biological law underlying it. So turning senescence off altogether should be possible. I suspect evolution has not already done so because overly long-lived creatures in the wild were on average bad news for their bloodlines—banging their grand daughters and occupying turf with the cunning of the old. Uhm. Now I have an itch to set up a simulation and run it.. Just so stories are not proof. Math is proof.