The resurrection of the dead: a brief feasibility study

• What is the “Common Task?”

Nikolai Fyodorovich Fyodorov wanted to resurrect every person who had ever died via technological means, a goal he called the Common Task.

It is a curious fact that very few people have thought about the possibility of the Common Task. Granted, in its strong form there is no reason to think that it is possible, except perhaps a naive expectation that technological maturity looks like godhood. Still, the common task has never been ruled out. There is probably less reason to think it is impossible than there is to think travelling faster than the speed of light is impossible.

Further, it is arguably one of the two most practically relevant ultimate questions in physics. The other one being how much energy can we amass. [a variety of more particular questions, e.g. can we escape the heat death of the universe” are in practical terms basically a special case of the “how much energy can we amass and use over our existence as a species” question. In effect, how many lives do we have the capacity to create and sustain- what is our cosmic endowment.] The two questions are complementary- the size of the future we have to expand into, and whether we can correct the wrongs of the past.

Now granted, I will concede once more, it’s not very likely that we’ll be able to accomplish Fyodorov’s common task, but again it’s not very likely at all that we’ll be able to travel faster than light. This has never stopped even respectable physicists speculating about FTL travel- including in well received journal articles. There is, however as far as I can tell, no review paper or even Arxiv(1) preprint comprehensively reviewing the question of whether we might ever be able to reconstruct the past in enough detail to recreate the souls of 3 million years’ worth of homins. Or even, leaving aside the specific question of the common task, whether we will ever be able to view moments from the past with almost total precision, even if just for the sake of curiosity.

If I had to speculate about why this is, I’d point to two factors:

1. There’s not much push from the public for research here, unlike FTL travel. This is understandable, there’s something fairly morbid about contemplating correcting all the wrongs of all time only to conclude that no, we probably never will. Those who cannot live without hope in this domain tend to turn to religion [Fyodorov himself was a lifelong Eastern Orthodox Christian. He didn’t see that as exclusive with his common task. I think he saw resurrecting the dead as a way to implement the salvation of all- Human imposed universalism seemed to be his ideal]. Popular ideas about personhood over time, which see it not so much as either biology or psychological pattern but as a kind of mysterious substance also don’t help here, since they make the Common Task unachievable.

2. Without prejudice or animus, Tipler and a few others have poisoned the well.

Even with these two explanations though, I must admit that I find it odd there is not more writing on the topic.

• Death, the Lyapunov exponent is thy sting: Fundamental physics largely says no to deep past reconstruction

Our best theories suggest that information is never really destroyed, however they give basically no hope for local and practically feasible reconstruction of the past.

I won’t labour the point here, because I’m an amateur and largely instructed by others in this matter. However, from what we can tell, reconstructing the soul of an anonymous peasant who died in 1750 let alone a Paleolithic hunter gatherer circa 30,000 years ago with any fidelity is probably impossible. In the case of many 30,000-year-old hunter gatherers we may never even be able to lock in a target: an approximate sense that a particular person existed and we’re trying to replicate them.

The problem is that while the information technically persists, it is not in a practically recoverable form, and in many cases is not locally available because it has dispersed at the speed of light.

There are caveats on this- game changers we can’t yet rule out. One is the discovery of some unknown form of trace- new physics. The other is that error correction mechanisms of some poorly understood sort might greatly constrain the forms the data generating process can take relative to our evidence, allowing for reconstruction. Other even more exotic caveats- observational time travel for example- exist. It is hard to put a probability on these possibilities.

On one hand, current physics gives us little reason to believe in the achievability of the common task. On the other hand, we can gesture to a sort of hopeful meta-induction. A meta-induction is an inductive inference over patterns of previous inductive inference- their success and failure. The most common example is the pessimistic meta-induction. We observe that past scientific theories, even those that looked strong, were often overturned. We infer from this that we should beware current scientific theories taken as a guide to underlying reality, even if they seem strong. I won’t litigate the pessimistic meta-induction here, I think it has significant flaws as an argument for what it’s worth, but here’s what has some merit in our specific case. Traces that we would never have imagined have become available over time, and people have called an end to new fundamental physics many times before. Nor is the inference that further physics must be subtle necessarily correct- see the atomic bomb. Even if further physics is subtle, it might still provide us with enough information.

I would say there is room to hope, but no room to expect, and I’d put far more weight on the second clause unfortunately. Through thermalisation, the gates of hell will likely prevail against us.

• Recent past reconstruction is probably possible

Our evidence sources for the near past are far richer. For a typical person alive today or recently dead we have, in roughly descending order of importance:

1. The memories of their peers

2. A vast social media trace

3. Their DNA

4. An institutional history- jobs, places lived, financial transactions etc.

Combine this with a fantastic model of human psychology and a fantastic model of what our clues mean processed through a Matrioshka brain. Youyou, Kosinski, and Stillwel (2015) suggests that a few hundred Facebook likes can predict someone’s big five personality traits better than their spouses. Yes, the big five is a vastly blunter and more open target, but our information is likewise so much richer. Think about the power of modelling working with unimaginably more data, based on unimaginably more information, looking for unimaginably more subtle cues of word choice etc, all cross checked against countless sources, combined with memories, genes etc in a cross-reference. The self, I am inclined to think, is very leaky, to the right eyes.

I’m at the extremely, perhaps delusionally, optimistic end of positions on how much you can reconstruct of a person given rich data and a great model. Let me be upfront that the next bit is what we might politely call “a bold conjecture”. When I was four, I decided to have an argument with my brother over possession of a toy bus because I thought it would be interesting. As far as I can remember I’ve never vocalised that to anyone. It’s not at all clear to me that this sort of thing would be beyond reconstruction if the self is as leaky as I think it is. It’s very plausible that given a corpus of production by a person you could reconstruct the salient thoughts- their darkest secrets and greatest hopes. Not just in what they talk about, but in subtle linguistic clues no human would ever notice. But I want to conjecture beyond that. Perhaps even a long dormant argument about a toy bus is observable in the play of billions of recorded and observable behaviours. All this is somewhat beside the point because people don’t really care about saving the bus-arguments. I think you could probably get 99% of most people weighted by what they value- more on weighting by value below. We certainly wouldn’t need to recover bus-arguments to get to 99%, although it’s not out of the question.

Would we ever know how well we’d modelled the soul? It’s likely we could. You can check the accuracy of reconstructions through several converging mechanisms. One is held out data- that is to say, having the model predict the structure of a person with partial data, holding out some other data, and then seeing if the model’s predictions are compatible with the held-out data. Another is inverse modelling. First make up a person and simulate them living a life- such a simulation could be based on perfectly known facts about 20th century culture, human biology etc- it wouldn’t need to be the same model of traces to persons we’re testing. Then see what traces would result from the simulated life and have the model try to predict the person from the traces, see how well it does. It goes without saying you shouldn’t discard any persons created by such modelling processes.

• What even is the goal

Obviously, any attempt to complete Fyodorov’s common task, in whole or part, depends on accepting something like the psychological-pattern theory of identity, or something more mystical like a soul theory of identity plus a claim that souls will be “attracted” by the creation of something with the relevant characteristics. If persons are identical with organisms the project is a non-starter. Thus, we’ll take the psychological pattern view for granted.

Now it’s important to be clear that what we require to reconstruct someone is not their exact psychological state. Why is this? If you fall asleep and wake up in the morning after dreamless sleep, you haven’t died. We could plausibly take this a good deal further before losing much of value at all. What counts as a person is a wide basin of states.

Parfit’s work gives us two useful lights here. The first is that the continuation of personal identity and the persistence of what you value can come apart. Parfit is strongly associated with the psychological view, but according to Edmonds biography at some points in his career he accepted animalism about identity. He just didn’t think identity was what mattered. Psychological continuity is what we value, whether this is “survival” is a far more nebulous matter What we should care about is practices of valuing of various kinds. Maybe we should even equate practices of value as defining survival, a point often made by my old academic supervisor Kristie Miller. The bifurcation between what matters and identity also helps us put aside some standard tricks- what if two copies of a person is made- which is the real one? We don’t really care. It’s survival for our purposes.

The second is that on a Parfitian view there is likely a spectrum of value in preservation. This spectrum is very unlikely to be a straight line; there is probably a point of criticality. It may not even be monotonic in all circumstances- we actively value certain kinds of openness to transformation. Someone frozen psychologically at 25 forever doesn’t have the worst deal in the world, but they have lost something. However, things may not be quite so simple. Claude Fable made a good point about this to me - freezing is itself a form of psychological modification. The vector of transformation - the tendency to evolve in certain ways - is stripped out. A person is partially defined by all the ways different possible combinations of circumstances would make them become another person.

And it’s not just a spectrum weighted by information alone, it’s a spectrum weighted by importance to us. Most of all people don’t want to lose their moral capacities and their closest relationships- although improvements are often (but not always) fine. After this, probably personality and cognitive traits come next in the order of priority. After this, people do not want to lose certain very core memories and tastes. After this, less important memories and tastes and so on.

There is empirical literature on this, based on third person vignettes in which a participant assesses the persistence of identity. The research tends to vindicate 1. that moral identity is core 2. That improvements are better received than diminutions and 3. causal centrality of a psychological feature to a person’s actions intimately is intimately linked to that psychological feature’s importance. Caregiver studies find Alzheimer’s, even though it involves greater memory loss, is typically found by relatives to be less person destroying than Frontotemporal dementia. This is because FTD targets character more rapidly, strongly and directly than Alzheimer’s. How important the persistence of cognitive abilities is to the persistence of perceived personal identity is less tested. This would be an excellent gap in the literature to fill. I would urge experimenters to beware social desirability bias though. Saying of someone that they have lost a lot of who they were due to a massive drop in intellectual capacity feels like a very mean thing to say. Thus, any study should watch out for participants downplaying the importance of cognitive abilities for this reason. I would suggest looking at revealed preferences, and particularly revealed preferences in relation to self rather than others.

The fortunate news here is that the traits that are more stable and causally central are also those that are easier to reconstruct. This also applies within categories like memory. Longer lasting and more identity-narrative shaping memories are probably easier to find or uncover. This pattern is greatly helpful for obvious reasons. I may be an optimist about my toy-bus case, but ultimately answering whether I have a specific, never stated memory like that is difficult for the same reason it doesn’t matter much: it had little impact on my life. Fortunate indeed. There are exceptions. Perhaps a person might treasure privately a memory but take steps to stop it ever leaking into their behaviour. However, in the main there is a strong correlation between what matters and what can be recovered.

My great aunt (2) died in approximately the year 2004 at the age of 100. Will we ever be able to recreate her in some form she would value? Could we achieve 99% recovery of that which matters? I think it’s a genuinely open question. She wasn’t on social media much (at all), so we’ll have her genome, the memories of those who knew her, and bits of records. How much information would we need to recreate her? Vastly less than that needed to reconstruct the whole connectome (10^15 or so bits), vastly less than that necessary to reconstruct literally all knowledge available to long-term memory (long term memory is often given as 10^9 bits. However this is one of those figures that ‘goes around’ without clear justification that we should perhaps be sceptical of). I want to guess between 10^7 bits(3) with good compression and a great model of human psychology could get us well over 99%. However, I have no strong defence for this figure, except that it allows for a few hundred core memories (including key relationships), a narrative of self, a highly accurate compression of the genome (if thought necessary) and a large number of personality and cognitive dimensions (including moral dimensions).

• The ethics of resurrection

Some people might not want to be resurrected into a utopia. There are things you could do to ameliorate the culture shock of course. Access to simulations that mimic the favourite bits of the recreated person’s lives and cultural contexts for example. How sufficient that would be is unclear.

Here are some difficult questions for which I don’t have the answers:

1. If your best guess is someone will initially reject having been recreated, can you still ethically recreate them if you think they’ll eventually approve of the choice?

2. What if you can recreate Bob in detail, but not Alice, and Alice and Bob were life partners?

3. If recreating someone requires running through a simulation of their life, with all the suffering that entails, do you still have the right to do it?

4. Do you have the right to forbid a newly created being from destroying itself because you think it will eventually be happy?

5. Do you recreate people with the parts of themselves that will cause them suffering? If so, presumably the compromise is to let them slowly alter these bits of themselves after coming back into being.

6. A point looming over all of this is that most of those we could potentially resurrect do not believe the psychological continuity theory of what valuable survival means. Teasing out the implications of this could be its own essay.

I wonder. Is there a good sense, a la Walter Benjamin or Fyodorov himself, in which we can redeem the past and not merely cancel out its evils? Chisholm draws a distinction between defeating and merely outweighing evil. Roughly, evil is outweighed if it becomes overall worthwhile, if it becomes a necessitating condition for an overall better state of affairs. Evil is defeated if it is made no longer evil at all, merely apparently evil- when seen up close- and not as part of the larger moral order of things.

One thought that comes to mind here is Parfit’s non-identity problem. The problem is that you seemingly cannot be harmed by a condition that was necessary for you to come into existence. Harm implies a comparison between how you are in one situation and how you are in another. The people living in utopia wouldn’t be the people they are without their prior experiences, good or bad. Hence, they cannot complain without rejecting their own existence. If you can’t complain about X, maybe it wasn’t really an evil for you then!

I would like something more substantial than this. Something more ethically and aesthetically whole. Appealing to the non-identity problem as an argument generally feels like an attempt to win a legal case through a technicality or standing concern. That’s definitely true here. Perhaps we might make non-identity as a defeater more palatable as follows: the evil might be defeated (in Chisholm's sense) by something like:

The resurrected feeling a positive love of all apparent evils that happened to them because they gave them existence.

A bit like Nietzsche’s concept of amor fati. Of course, it is likely not everyone will agree, so evil might not be completely defeated in this sense.

Finally, why aim to raise the dead at all? Why not just create lots of happy interesting new people? There are several reasons.

The first only applies if we could raise all the dead. Essentially, existing in a state where some of your loved ones -or even just forerunners- are dead is a state of alienation- a wound. The wound is worse if you feel terrible about it, but it is constitutively harmful even if psychologically you’ve come to terms with it. This was roughly Fyodorov’s most famous reason for aiming at the common task. However, given that we probably won’t be able to resurrect everyone it’s not clear to what degree this argument works. If you recreate your parents but can’t recreate your grandparents, you’re just shifting the wound to your parents. I guess you could try to argue that since there were fewer people in the past, the quantum of alienation is diminishing the further you push it back.

The second reason is a fairness argument. There would probably be demand for the recreation of lost loved ones after the singularity assuming technical feasibility, meaning at least some people will be recreated. It would seem unfair to exclude other people who could be recreated just because no one happened to “request” them.

The third reason is what we might call a duty to rescue – if someone is dead, and you can undo that, you have, holding all else equal, a duty to do so. This is both independently plausible and can be motivated in various ways. For example, at least if you buy into the psychological continuity theory of personal identity, you’d probably appreciate it if other people bought you back if they had the chance. Do unto others and so on.

The fourth reason is because our values as a species have always been somewhat backwards looking as well as forwards looking. We want to correct what went wrong in the past. We also want to honour and enhance the good that came of it. Redemption of what went wrong, and honouring what went right is important to us. The case for resurrection as a chance to undo or at least compensate tragedy is obvious, especially for all lives with significant injustices, e.g. all of them. The case for resurrection as an attempt to honour sacrifices, good deeds and achievements is perhaps less compelling, but still present.

• What are the practical implications now?

If you really wanted to prevent the permanent death of as many people as possible, racing towards the technological maximum would be tempting. More memories, more traces of those who died, and fewer overall dead. However, for existential risk reasons I would not endorse this.

Perhaps the only practical upshot from what I’ve described that I can see is that it makes a case for good record keeping. Projects like the internet archive, scans of books etc. should be preserved. I’m not suggesting pursuing this where it may come into conflict with privacy. Even if it were in some abstract sense “right”., “We need to not destroy privacy compromising records so one day we’ll better be able to raise the dead with greater fidelity” is not politically or morally tenable and shouldn’t be. However more modest steps certainly are. And we should be better record keepers anyway, for all sorts of reasons.

One thing that you should absolutely not do on a personal level or encourage on a social level is use the possibility of recreating the dead as the basis for complacency of any sort about anything at all.

• Philosophy Bear, for some bizarre reason I want to know what you think, what are you aiming at. What do you think society should look like at technological maximum. What should we do when we can turn stars into fuel for our dreams?

I’m glad you asked, fictional interlocutor. Let’s talk about actually ideal political and ethical theory. My ideal is a society in which:

1. Superintelligence pushes transaction, coordination and planning costs to zero, leading to an abundant, democratic society which has abolished the economic and political as we currently understand them as separate spheres of life. [This should not be read as an endorsement of building superintelligent machines now]

2. Immortality and technological self-transformation allow us to embark on a project of self-perfection and experimentation. Seeking to be creatures that live lives of the richest welfare, where welfare is understood in broad terms, not flattened to something like hedonic state or preference satisfaction. This might involve, for example, experimenting with a series of incarnations in simulated worlds. Allowing you to experience all kinds of lives and explore possibilities in living artistically. Important goods here include, but are not limited to, friendship, happiness, virtue, knowledge, play and beauty.

3. We do our best to create all sorts of interesting beings and lives. That includes, as far as possible, raising the dead. Not just at the end of their lives either- but probably at different life slices.

Most consideration of what a world of genuinely high technology would look like- infinite self-modifiability, infinite potential surveillance, near infinite creative power- is that it would lock in the values we already have. Against this backdrop, thinking about what you really want before we get locked in and trying to shape a better world, is critical.

(1)- I’m not putting a fucking capital letter in the middle of a word. Not doing it. Sorry.

(2)- I think she might have been my great-great aunt, but we called her great aunt? Not sure.

(3)- In some moods I could argue as low as 10^5, but I’m trying to be pessimistic here.