Think about this. If someone tells you that your design will fail once in 500,000 years, you will surely go home and rest. Your assessment is that the design and the system are good enough. You will go home and feel that you have done a great job. Not bad, in statistics, 1/500,000 is such a great number. If you take that into days, you will get more leverage.
Then consider that in 2003, the Nuclear Safety Commission of Japan said that a fatality arising from radiation exposure will occur once per million years. Interpretation? We are safe and there is no problem – at least we have to live till next 500,000 years before we get to worry.
That is the problem. We spread risk and put it in the future without any reason. That a nation can have a nuclear disaster one per million years does not mean it cannot happen the week after or even the next day. The probability that it could be any day is evenly spread unless that risk or failure is tied to age or fatigue of the system. But when the events could be external like tornadoes, wars, earthquakes, etc, that chance of once per million does not matter. It is probable today as it is in the next million years.
So next time you see those statistics, do not think they will take ages to happen. For Japan, the million years chance happened after just 8 years. They must be nuclear related disaster free to even justify that model of risk.