Resurrection of the dead via multiverse-wide acausual cooperation

TL;DR: Mea­sure de­cline in ran­dom mind cre­ation may be pre­vented if we take into ac­count very large num­ber of ran­dom mids cre­ated in other uni­verses.

Sum­mary: P.Al­mond sug­gested the idea of the re­s­ur­rec­tion of the dead via a quan­tum ran­dom gen­er­a­tor which cre­ates a ran­dom mind, but such an ap­proach has sev­eral prob­lems: non-hu­man be­ings in our world, non-nec­es­sary suffer­ing of non-perfect copies, and mea­sure de­cline.

Here I sug­gest three patches, which pre­vent most of the un­de­sired effects:

1. Hu­man mind ma­trix to pre­vent pure ran­dom minds ap­pear­ing.

2. Digi­tal im­mor­tal­ity data to cre­ate a per­son which satis­fies all known ex­ter­nal ex­pec­ta­tions, and the use of ran­dom­ness only to fill un­known in­for­ma­tion.

3. Mul­ti­verse-wide co­op­er­a­tion for the “cross-re­s­ur­rec­tion” of the dead be­tween mul­ti­ple wor­lds via quan­tum ran­dom minds, so the to­tal mea­sure of all re­s­ur­rected peo­ple will not de­cline.

1. Introduction

Al­mond in “Many-Wor­lds As­sisted Mind Upload­ing: A Thought Ex­per­i­ment” sug­gested the fol­low­ing idea about the re­s­ur­rec­tion of the dead by the use of a quan­tum ran­dom gen­er­a­tor, which would cre­ate a ran­dom mind within a com­puter (Al­mond, 2006):

[A tech­ni­cian who lost some­one’s brain scan file] writes a com­puter pro­gram which takes in­put from a phys­i­cal sys­tem. The phys­i­cal sys­tem, known as a quan­tum event gen­er­a­tor, gen­er­ates “1”s and “0“s ran­domly as a re­sult of quan­tum events. The pro­gram will use the phys­i­cal sys­tem to tell it what se­quence of “1”s and “0”s will be used to try to recre­ate the lost scan file. The pro­gram starts with an empty scan file which will be filled with “1”s and “0″s.

If the many-wor­lds in­ter­pre­ta­tion of quan­tum me­chan­ics is cor­rect, all pos­si­ble minds will ap­pear in sep­a­rate timelines start­ing from the mo­ment of ran­dom mind cre­ation, which would mean the re­s­ur­rec­tion of ev­ery­one from his own point of view. How­ever, this ap­proach will a) not help an out­side ob­server, who wants to re­s­ur­rect a rel­a­tive, for in­stance, as the ob­server would see only a ran­dom mind, and b) the quan­tum “mea­sure” of ex­is­tence of each mind will be in­finitely small.

2. Prob­lems of Al­mond’s approach

To illus­trate the prob­lems with quan­tum mind up­load­ing, I will ex­plore a sim­plified thought ex­per­i­ment where only names will be re­stored us­ing quan­tum mind up­load­ing. First, here is what Al­mond sug­gested:

Thought ex­per­i­ment “Not-patched quan­tum mind up­load­ing”:

Bob had a friend John Smith. John has died and Bob wants to re­s­ur­rect him. Bob re­mem­bers only first let­ter of John’s name: S.

Bob and John are in­ter­ested only in the unique­ness of name preser­va­tion, and no other iden­tity con­sid­er­a­tions are im­por­tant. Bob wants to ob­serve his friend to be al­ive, and for his friend to be named “John S….” (I would call it im­mor­tal­ity from the point of view of the ex­ter­nal ob­server). John wants his own im­mor­tal­ity, and will be satis­fied only if “John Smith” is cre­ated.

Bob cre­ates ran­dom quan­tum mind A us­ing a quan­tum gen­er­a­tor to choose each new let­ter in the names.

It turns out that A is “jYY2№@11”. Only less than 10-30 share of all such copies in the mul­ti­verse are named John Smith. Both Bob and John are un­happy.

This thought ex­per­i­ment leaves both John and Bob un­satis­fied, and we see three rea­sons for that be­low:

2.1. Prob­lem 1: Mea­sure decline

Prob­lem 1 is a prob­lem for John.

Mea­sure could be defined as a share of an ob­server of a given type be­tween all pos­si­ble ob­servers. If the typ­i­cal size of the simu­lated mind is, say, 10^15 bites, the chances that a ran­domly gen­er­ated mind will be ex­actly the needed per­son is 2^(10^15). In other words, a quan­tum mind gen­er­a­tor re­sults in a mea­sure de­cline of 2^-(10^15) which is an ex­tremely large num­ber. Even in our thought ex­per­i­ment 1 mea­sure de­cline is 1030 times.

Many au­thors claim that large mea­sure de­cline should be treated as death or as an in­finitely small chance of sur­vival. Such dis­cus­sions ap­peared in the con­text of so-called quan­tum im­mor­tal­ity, that is, the coun­ter­fac­tual pos­si­bil­ity to sur­vive death via ex­ist­ing in quan­tum mul­ti­verse timelines where a per­son will not die.

Even if the mea­sure de­cline is not bad per se, it leads to a world where very small prob­a­bil­ity out­comes will dom­i­nate pos­si­ble fu­tures of an ob­server, and such par­a­sitic out­comes may be full of suffer­ing. For ex­am­ple, the quan­tum im­mor­tal­ity im­prob­a­ble sur­vival land­scape may be dom­i­nated by peo­ple who are very old and dy­ing but can’t die (it could be patched by sign­ing up for cry­on­ics).

If we use some ex­pected util­ity calcu­la­tions, and mea­sure de­cline re­sults in de­clin­ing util­ity of any use­ful out­come as­so­ci­ated with it, we could just ig­nore my copies with in­finitely small mea­sures.

2.2. Prob­lem 2: Non-hu­man and not wel­comed minds

Prob­lem 2 is mostly for Bob.

Another prob­lem is that most ran­dom minds will be non-hu­man, and will not be adapted to our world, so they will suffer or cause suffer­ing to peo­ple liv­ing here. In our thought ex­per­i­ment “jYY2№@11” is an ex­am­ple of a non-hu­man ran­dom mind.

Such ran­dom minds are also ex­tremely bad for any out­side ob­server, like Bob, as he will be very un­likely to meet any­one re­sem­bling his friend John Smith.

2.3. Prob­lem 3: Da­m­aged minds

Prob­lem 3 is a prob­lem for both Bob and for John.

Most ran­domly-cre­ated minds will be not minds at all, but some garbage code, or at “best case,” dam­aged minds. For ex­am­ple, if Bob wants to re­s­ur­rect John Smith, there will be much more copies where his name (as well as his other prop­er­ties) is a par­ody of the name Smith, for ex­am­ple Smthi, Smiht, Misth, Smitt, etc. For n bits long name, there are n in­di­vi­d­ual names which have 1 bit differ­ence.

Thus, for any real per­son, there will be much larger set of his-her dam­aged copies, which im­plies suffer­ing for such a per­son as the most prob­a­ble out­come of the quan­tum ran­dom re­s­ur­rec­tion and s-risks for all peo­ple.

3. Patches

For­tu­nately, quan­tum ran­dom mind up­load­ing could be patched, so it will provide much more satis­fac­tion for John and Bob.

Patch 1. The use of the hu­man mind’s uni­ver­sal model as a start­ing point

The goal of this patch is to es­cape minds of “aliens” or of non-work­able gib­ber­ish code, and thus pre­vent suffer­ing of most cre­ated minds. For ex­am­ple, for a hu­man mind model, his-her pos­si­ble name will be gen­er­ated not as ran­dom sym­bols but from the pre­set of typ­i­cal hu­man names.

Such a hu­man mind model may look like an un­trained neu­ral net­work which has the gen­eral ar­chi­tec­ture of a hu­man mind, with some other con­straints, so any ran­dom set of pa­ram­e­ters will cre­ate a more-or-less nor­mal hu­man mind. We as­sume that some fu­ture as­sis­tant AI will be able to find an ap­pro­pri­ate model.

In that case, Bob uses a ran­dom mind gen­er­a­tor for pa­ram­e­ters of the uni­ver­sal hu­man mind model. He gets “Maria Stu­art”. This will in­crease the share of the wor­lds where real John Smith is re­s­ur­rected to 10-10. Both John and Bob are a lit­tle bit more satis­fied, as Bob gets a hu­man friend, and John in­creases his mea­sure.

Ob­vi­ously, some minds may not want to be re­s­ur­rected, but this could an im­por­tant pa­ram­e­ter in the model, and mod­els, where “re­s­ur­rec­tion prefer­ence = false” will be ig­nored.

Patch 2. The use of the digi­tal im­mor­tal­ity data to cre­ate only minds which com­ply with our expectations

The prob­lem of Bob’s satis­fac­tion could be over­come by the use of Bob’s ex­pec­ta­tions as pri­ors, if there are no other cur­rent of fu­ture sources of data about John.

In that case, Bob could use his mem­o­ries about John S. to cre­ate a model of John S. He re­mem­bers that John was ei­ther John Smith or John Simp­son. He uses a ran­dom quan­tum coin to choose be­tween Smith or Simp­son, and gets “John Simp­son”.

In an­other branch of the quan­tum mul­ti­verse, where the coin fails tails, John Smith ap­pears, but his mea­sure de­clines to 0.5. Both John and Bob are partly satis­fied. Bob got some­one who looks like his friend, but Bob knows that it is not ex­actly his friend, and that his friend has now smaller mea­sure of ex­is­tence.

Digi­tal im­mor­tal­ity, or in­di­rect mind up­load­ing, is the col­lect­ing in­for­ma­tion about a per­son while he is al­ive with hope that fu­ture ad­vanced AI may be able to re­s­ur­rect the per­son, by cre­at­ing an ad­vanced model of the per­son­al­ity based on all available in­for­ma­tion. Such a model will, by defi­ni­tion, satisfy Bob and all other rel­a­tives, as all available in­for­ma­tion has already been taken into ac­count, in­clud­ing all rel­a­tives’ ex­pec­ta­tions. How­ever, large chunks of in­for­ma­tion will never be known, and thus have to be re­placed with some ran­dom data. Even if quan­tum ran­dom­ness is used to fill the gaps, John will have an in­finitely small share of all pos­si­ble wor­lds, and in most other wor­lds he will be re­placed by some­one else.

Patch 3. The use of mul­ti­verse-wide co­op­er­a­tion for the cross-resurrection

The next step is that Bob con­sid­ers that not only his uni­verse ex­ists, but all pos­si­ble other uni­verses ex­ist in the Mul­ti­verse.

Bob con­cludes that be­cause all pos­si­ble ob­servers ex­ist in the Mul­ti­verse, his John Simp­son cre­ated via a quan­tum ran­dom gen­er­a­tor is a re­s­ur­rec­tion of some John Simp­son from an­other uni­verse, while John Smith who lived in our uni­verse, will be re­s­ur­rected in some other uni­verse where an­other copy of Bob will do the same ex­per­i­ment.

In other words, Bob and Bob’s copies in other uni­verses co­op­er­ate to re­s­ur­rect the ex­act John Smith.

As the sec­ond uni­verse is ex­actly the same as ours ex­cept for John’s name, there is an­other ex­act copy of Bob in it, and this Воb’s copy is also want­ing to re­s­ur­rect his friend John S., so he uses an­other quan­tum ran­dom mind gen­er­a­tor. Now the fol­low­ing hap­pens:

So, the to­tal mea­sure of John Smith has not de­clined, if Bob takes into ac­count that other copies of Bob in other uni­verses will run the same ex­per­i­ment. By de­cid­ing to start the ran­dom mind gen­er­a­tor (and to not turn off the re­sult­ing mind), Bob joins a large group of other minds, who think similarly, but who are lo­cated in causally dis­con­nected parts of the Mul­ti­verse. Every­one ex­pects that some other ran­dom gen­er­a­tor recre­ates an ex­act copy of their loved one.

In a real case of large miss­ing data, like gi­gabytes, this re­quires a si­mul­ta­neous run of an ex­tremely large num­ber of quan­tum ran­dom mind gen­er­a­tors, like 10^(10^9), which is only pos­si­ble via mul­ti­verse-wide co­op­er­a­tion. The mea­sure will not de­cline in such a case too, as for ev­ery dead per­son there will be one ran­dom per­son, and given the large num­bers, any per­son will be ran­domly recre­ated, at least in ap­prox­i­mately one world. (Some may go deeper and take into ac­count stan­dard de­vi­a­tion, but be­cause we use quan­tum gen­er­a­tors in the many wor­lds in­ter­pre­ta­tion, each uni­verse cre­ates ex­actly its share of John, and there will be no fluc­tu­a­tions, which would re­sult in non-ex­is­tence of some Johns and two copies of an­other.)

Any of Воb’s copies can join such a mul­ti­verse-wide co­op­er­a­tion by cre­at­ing just one quan­tum ran­dom mind (and treat­ing the re­sult­ing mind well).

4. Re­main­ing problems

Mul­ti­verse. What if the mul­ti­verse doesn’t ac­tu­ally ex­ist? In that case, Bob and John get partly satis­fy­ing re­sults, as Bob gets John’s copy, but John’s copy is not perfect from John’s point of view. If the quan­tum mul­ti­verse is not real, but some other form of the mul­ti­verse ex­ists, like the one based on in­fla­tional cos­mol­ogy, the re­s­ur­rec­tion method will still work.

Defec­tion. Bob may not cre­ate any ran­dom mind gen­er­a­tors at all but still ex­pect that some­one else will recre­ate his friend. In gen­eral, the rate of defec­tions may be known and com­pen­sated by in­creas­ing the num­ber of ran­dom minds by those who have more re­sources.

There are sev­eral other pos­si­ble generic prob­lems of mul­ti­verse-wide co­op­er­a­tion, in­clud­ing in­finite ethics, the pos­si­bil­ity of acausal black­mail, a method to mea­sure similar­ity be­tween agents, and prob­lems with agents that have other val­ues as de­scribed in EA post’s com­ment.

Conclusion

I hope that this post may in­crease one’s hope in the fu­ture per­sonal re­s­ur­rec­tion by su­per­in­tel­li­gent AI.