[radu_floricica]

The question is VERY fair, and I don't want to undermine it. I just want to be the first to say that given the huge potential impact, "potential" is bad enough.

When dealing with probabilities there are two measures that need to be considered independently: probability and effect. Traditional statistics multiplies them into "expected utility" and works just with the aggregated value. But that's tricky, because effect is not a continuous function. Imagine a pebble falling on your head plotted based on its weight - there'll be sharp discontinuities at "hurts", "commotion" and "death".

At society's level there are also similar discontinuities. Something that may cause tens of million of cases of cancer over time should be in a separate category from a medicine that may suddenly kill a few hundred people, even though the latter has a bigger emotional impact.