If you had a perfectly accurate crystal ball and the number of accidents you would have in your lifetime is zero then your account wouldn't need to have any money in it.
It's true that perfectly accurate crystal balls would render insurance irrelevant. It doesn't strike me as that interesting of a revelation, but it is true.
I guess it's a response to people saying "They need to charge bad drivers more that they do, in proportion to how badly they drive." If insurance could accurately do that, it wouldn't be insurance--it would just be an individualized pre-paid accident savings account.
More to the point, it erodes the value of the system as insurance.
Suppose the insurance thinks that Alice has a 90% probability of an accident and Carol has a 15% probability, so they want to charge Alice six times the premiums of Carol. Then in practice Carol is the one who has the accident and not Alice, because it's not perfect.
But the pool Alice is in is much smaller than the other one, so if they were merged, the combined group would only be paying 10% more than Carol does, which would be serving the purpose of insurance -- spreading risk. Whereas if you separate them, Alice is screwed -- even though she isn't even going to have an accident -- because now she has to pay >$7000/year in insurance rather than ~$1300 when you combine these imperfect predictions with smaller risk pools.
