Pascal's wager's application to real-world cost-benefit calculations is pretty much always a fallacy:

http://lesswrong.com/lw/z0/the_pascals_wager_fallacy_fallacy...

Large payoff and small chance (with a wide confidence interval) has nothing to do with "infinite" payoff and "infinitesimal" chance. There are valid and interesting decision theoretic problems and paradoxes that arise with unbounded payoffs, but those aren't real-world in the way that outlier risk scenarios are real-world in banking and investment.

Cryonics is absolutely real-world in that sense, and we can and should hang some expected value tag there.

I have a fairly low confidence that cryonics will in fact work, but I wholly stand by my point that if it turns out to work, then that should be cause for a huge "we fucked up" moment coupled with a copious amount of soul-searching. But this shouldn't be put into a reference frame centered on "guilt" or "responsibility", although such things would be inevitably involved. Rather, it's a case of established models turning out to be wrong, and what matters is that everyone updates and learns from it as quickly as possible. Ignorance can be condoned but not willful and destructive ignorance in the face of ample evidence.