I have two questions on Metaculus that compare how good elements of a pair of cryonics techniques are: preservation by Alcor vs preservation by CI, and preservation using fixatives vs preservation without fixatives. They are forecasts of the value (% of people preserved with technique A who are revived by 2200)/(% of people preserved with technique B who are revived by 2200), which barring weird things happening with identity is the likelihood ratio of someone waking up if you learn that they’ve been preserved with one technique vs the other.
Interpreting these predictions in a way that’s directly useful requires some extra work—you need some model for turning the ratio P(revival|technique A)/P(revival|technique B) into plain P(revival|technique X), which is the thing you care about when deciding how much to pay for a cryopreservation.
One toy model is to assume that one technique works (P(revival) = x), but the other technique may be flawed (P(revival) < x). If r < 1, it’s the technique in the numerator that’s flawed, and if r > 1, it’s the technique in the denominator that’s flawed. This is what I guess is behind the trimodality in the Metaculus community median: there are peaks at the high end, the low end, and at exactly 1, perhaps corresponding to one working, the other working, and both working.
For the current community medians (as of 2021-04-18), using that model, using the Ergo library, normalizing the working technique to 100%, I find:
I have two questions on Metaculus that compare how good elements of a pair of cryonics techniques are: preservation by Alcor vs preservation by CI, and preservation using fixatives vs preservation without fixatives. They are forecasts of the value (% of people preserved with technique A who are revived by 2200)/(% of people preserved with technique B who are revived by 2200), which barring weird things happening with identity is the likelihood ratio of someone waking up if you learn that they’ve been preserved with one technique vs the other.
Interpreting these predictions in a way that’s directly useful requires some extra work—you need some model for turning the ratio P(revival|technique A)/P(revival|technique B) into plain P(revival|technique X), which is the thing you care about when deciding how much to pay for a cryopreservation.
One toy model is to assume that one technique works (P(revival) = x), but the other technique may be flawed (P(revival) < x). If r < 1, it’s the technique in the numerator that’s flawed, and if r > 1, it’s the technique in the denominator that’s flawed. This is what I guess is behind the trimodality in the Metaculus community median: there are peaks at the high end, the low end, and at exactly 1, perhaps corresponding to one working, the other working, and both working.
For the current community medians (as of 2021-04-18), using that model, using the Ergo library, normalizing the working technique to 100%, I find:
Alcor vs CI:
EV(Preserved with Alcor) = 69%
EV(Preserved with Cryonics Institue) = 76%
Fixatives vs non-Fixatives
EV(Preserved using Fixatives) = 83%
EV(Preserved without using Fixatives) = 34%
(here’s the Colab notebook)