Roko comments on Cryonics Wants To Be Big

[Roko]

what probability did you get out of your program? (I ask this as my figure for that probability for myself is ~ 15%, when I actually pay up for the contract)

[apophenia]

Here’s my program, with my probability estimates removed to avoid bias. If anyone wants to use it, feel free. It calculates the probablilty a person will be revived after X years. If you want a final probability, plug in some priors, including a distribution over how long you think it will take you to be revived.

#lang scheme (require plot)

(define (cycle-to-chance cycle-length) (/​ 1 cycle-length))

(define business-failure-cycle-length XXX) ;How often will a cryonics agency go out of business?

(define dark-age-cycle-length XXX) ;How often will a technical/​ethical/​economic dark age happen?

(define depression-cycle-length XXX) ;How often will a great depression hit North America?

(define Pnd XXX) ;Probability of a natural distaster

(define Pr XXX) ;Chance a corpse will be revived when cryogenic tech is available

(define Pbd XXX) ;Chance the information needed to revive a corpse is present

(define Pdh XXX) ;Chance humanity is destroyed (considered as a single time-indepenent factor)

(define Pcif (cycle-to-chance business-failure-cycle-length)) ;Chance the cryonics institute fails in a reasonable economy, per year

(define Pcif-cc XXX) ;Chance the corpse unfreezes if the cryonics institute fails in a reasonable economy

(define Pgef (cycle-to-chance depression-cycle-length)) ;Chance of a great depression in north america, per year

(define Pgef-cc XXX) ;Chance the corpse unfreezes in a great depression

(define Pda (cycle-to-chance dark-age-cycle-length)) ;Chance of a dark age, per year

(define Pda-cc XXX) ;Change the corpse unfreezes in a dark age

(define Pvc XXX) ;Probability a corpse is viable for freezing on death

(define Pst XXX) ;Probability a corpse is safely transported and frozen

(define entry-shock-pass-chance (* Pst Pvc (- 1 Pdh)))

(define exit-shock-pass-chance (* Pr Pbd))

(define yearly-pass-chance (min (- 1 ( Pcif Pcif-cc )) (- 1 ( Pgef Pgef-cc)) (- 1 (* Pda Pda-cc))))

(define (success-distribution time) (* entry-shock-pass-chance exit-shock-pass-chance (expt yearly-pass-chance time)))

(define (approx-integrate f start end step-size)

(apply +

    (map (lambda (inc-start)

           (* (/ (+ (f inc-start) 

                    (f (+ inc-start step-size))) 

                 2)

              step-size))

         (for/list ((i (in-range start (- end step-size) step-size))) 

           i))))

Here is my final probability, rot13ed: Gur cebonovyvgl V’yy or erivirq va svsgl, bar uhaqerq lrnef ner svir-cbvag-guerr-creprag, gjb-cbvag-guerr-creprag erfcrpgviryl.

Edit: I’m having a hard time getting this program source to display properly, but it should run fine with some line breaks.

[apophenia]

Okay, so I updated my chance that the correct information is present in the brain—from 75% to 99% after seeing this talk by Brian Wowk, with the team that preserved a rabbit liver. My estimates went down, not up. There was an error in my program, which I’ve corrected above. Even with this correction, I strongly recommend against using it without checking it yourself first.

My probability estimates are now nearly ten times higher, and I will sign up for cryonics. I guess I should either not stake my life on my programming skills, or program better.

[Roko]

What do you think life after revival would be like? What is your median expectation? Your 80% expectation (i.e. the post-revival life that is just better than four fifths of the possibilities)?

[apophenia]

I haven’t the faintest idea. I don’t apply Bayesian rules in everyday life, and I don’t like to guess. But, let me even put that aside. Suppose my personal happiness would be about what it is now, and that I would continue to enjoy life for at least as long as I’ve been alive currently (about 2 decades). That’s already more happiness than I can get a subjective impression of, so I don’t feel like I can come up with a helpful answer.

[Roko]

There seems little point in spending a relatively long time coming up with probabilities if you don’t have a rough model of what the utilities of the outcomes are…

One argument in favour of cryonics is that the utility of a life on the “other side” would be high, because the technology and willingness to revive, combined with your savings having grown via compound interest, seems to imply that you would be rich and in a world where the good life is very cheap. The unfriendly/​friendly AI dichotomy reinforces this point.

The value you are placing on your post-cryo life implied by your cost and probability is $200,000 or so, which is about 41 times smaller than the $5,000,000 EDIT $8,200,000 standard statistical value of a life today. If you just assumed that your post revival life would be worth the same as the standard statistical value of a life, you would be prepared to pay $410/​month for cryo. (unless you discount your future self, of course, in which case if depends on your discount rate)

EDIT: Suppose you discount exponentially at 5% per annum and expect to die in about 50 years, then once you die you’d be reanimated subjectively instantly. You should discount by a factor of 11 for 50 years, so you should pay $410/​11 ~ $38 per month for cryo.

If you make some allowance for increased quality of life, and increased length, this should go up. I am not sure how to make such allowances: discounting a billion year life exponentially at 5% seems silly to me.

[apophenia]

Where did you get $5,000,000, and what exactly does it represent?

I’ll think about this, though. You’re right in saying I’ve spent too much time thinking about the probability of success, and not enough on the value of success. I strongly suspect my $10 comes from some idea of what a reasonable monthly fee should be as an anchor, adjusted for the probability. As such, I should reconsider it.

[Roko]

So you should look up value of a statistical life

Based on an extensive review of the research literature, the U.S. EPA (1997) suggests that a reasonable estimate of the VSL has a mean of $4.8 million with a confidence interval of plus or minus $3.2 million (in 1990 dollars. Note this is 8.2M +- 5.4M in 2010 dollars).

[Larks]

IIRC the figure is the average life insurance pay-out in the US. In the UK, it’s more like £1,000,000

[Roko]

Incidentally, what value would you place on life optimized by a friendly AI? i.e. how much per month would you be prepared to pay for, say, a 10% shot at it?