There's also "purely statistical" problems with the argument the 50 million at 0.50 probability is the better decision: it makes an erroneous ergodicity assumpt

There's also "purely statistical" problems with the argument the 50 million at 0.50 probability is the better decision: it makes an erroneous ergodicity assumption. In other words, it assumes that the expectation of a one-shot decision over people is the same as it is for the decision within-person. Sure, if you were making the choice over and over and over again, it would be better to repeatedly hit the green button. But that's not the situation. The situation is a mixture of outcomes, so for any given individual the expected outcome might be zero. There's also I think an implicit stationarity assumption built in, that if you say you'll pay me X amount over time, that you'll actually do that, that inflation wont eat it into oblivion, etc. It's a classic case of theoretical models not working in reality. This is kind of the point of the essay, but I think it could have been made more rigorously (as people here are pointing out).