The 2nd law doesn't argue against the first meaning, that microscopic events look the same in both time directions. That's true. The reason that doesn't generalize to macroscopic events (i.e. the 2nd law arguing against the second meaning) is illustrated by diffusion. Suppose you have a sufficiently hot interface between two pure metal slabs- there's lots of vacancies and the atoms are dancing around in the crystal structure. At the interface, you might have a vacancy surrounded by three atoms of metal A and two atoms of metal B. Each atom is roughly equally likely to hop to that vacancy, but it's 60-40 that an atom of metal A will move there instead of an atom of metal B. The result is that a sharp interface becomes a diffuse interface until eventually you have one slab of both A and B, with local distributions of A and B in thermal equilibrium with each other. Once this has happened, you won't be able to tell which side of the slab was originally A and which side was B.
What makes cryonics work is that it's very cold, which means that diffusion happens on massive timescales. The main question I have about information-theoretic death is how long someone's brain has to be dead at room temperature for information to be permanently lost. The long-term damage of even a few minutes of anoxic deprivation before standard revival is massive. What information you need from a frozen dead cell to make a functional live cell isn't well known- if it's just the knowledge that neuron A is connected to neuron B, we're in good shape. If it's what the local ion distributions were in the cellular soup and you let them diffuse for thirty minutes, it might be impossible.
Well, there's also acoustic fracture events, which Alcor cops to in their FAQ but sort of downplays the significance of. Even though vitrification prevents ice crystal formation, fractures occur at just a few degrees below the glass transition. Feeling lucky about the odds of checking and correcting the damage to 10^15 unmapped connections?
If an American signs up for cryonics and pays their ~$300/year, what are their odds of being revived? Talking to people at LessWrong meetups I've heard estimates of 1 in 2. My friend George Dahl, whose opinion I respect a lot, guesses "less than 1 in 10^6". Niether has given me reasons, those numbers are opaque. My estimate of these odds pretty much determines whether I should sign up. I could afford $300/year, and I would if I thought the odds were 1:2, but not if they were 1:10^6. [1]
In order to see how likely this is to work, we should look at the process. I would sign up with a cryonics company and for life insurance. I'd go on living, enjoying my life and the people around me, paying my annual fees, until some point when I died. After death they would drain my blood, replace it with something that doesn't rupture cell walls when it freezes, freeze me in liquid nitrogen, and leave me there for a long time. At some point, probably after the development of nanotechnology, people would revive me, probably as a computer program.
There's a lot of steps there, and it's easy to see ways they could go wrong. [3] Let's consider some cases and try to get probabilities [4]:
Update: the probabilities below are out of date, and only useful for understanding the comments. I've made a spreadsheet listing both my updated probabilities and those for as many other people as I can find: https://docs.google.com/spreadsheet/...
Combined Probability Of Failure: 99.82%
Odds of success: 1 in 567.
If you can think of other ways cryonics might fail, moving probability mass from "other" to something more quantifiable, that would be helpful. If you think my numbers are off for something, please let me know what a better number would be and why. This is not final.
Am I going about this right? Do people here who think it's rational to sign up for cryonics take a "the payoff is really high, so the small probability doesn't matter" view? Am I overly pessimistic about its chances of success?
Note: I originally posted this on my blog, and the version there has a silly javascript calculator for playing with the probabilities.
[1] To figure out what odds I would accept, I think the right approach is to treat this as if I were considering signing up for something certain and see how much I would pay, then see what odds bring this below $300/year. Even at 1:2 odds this is less effective than Village Reach at averting death [2], so this needs to come out of my 'money spent on me' budget. I think $10,000/year is about the most I'd be willing to spend. It's a lot, but not dying would be pretty nice. This means I'd need odds of 1:33 to sign up.
[2] Counter argument: you should care about quality adjusted life years and not deaths averted. Someone revived maybe should expect to have millenia of life at very high quality. This seems less likely to me than just the claim "will be revived". A lot less likely.
[3] In order to deal with independence issues, all my probability guesses are conditional on everything above them not happening. Each of these things must go right, so this works. For example, society collapsing and my cryonics organization going out of business are very much not independent. So the probability assigned to the latter is the chance that society won't collapse, but my organization goes out of business anyway. This means I can just multiply up the subelements to get probabilities for sections, and then multiply up sections to get an overall probability.
[4] This has a lot in common with the Warren formula, which was inspired by the Drake equation. Robin Hanson also has a breakdown. I also found a breakdown on LessWrong that seems really optimistic.
EDIT 2011-09-26: jsalvatier suggested an online spreadsheet, which is very sensible. Created
EDIT 2011-09-27: I've updated my probabilities some, and made the updates on the spreadsheet.