Sure, not every thought about murder leads to murder, and not every red light you run leads to an accident. But people that run more red lights have more accidents, and people who think about bad things more (let's posit) will do more bad things.
"Oh, you only think about murder 3 times a week? Discount!"
Still ready to let your thoughts about Suzy get archived on some hard drive?
(Imagine e.g. someone in Florida being rated for hurricane insurance on the mistaken assumption they lived in Kansas, or a cab driver rated as if the car was just used for a daily commute.)
That insight in no way contradicts the principle that a completely predictable event is not insurable, or that insurance allows distribution of the costs of an event across a class.
But I still think it's fair to call someone a freeloader if they deliberately get mixed in with a low risk class while knowing they're higher risk.
Edit: And you're right, that in the case of perfect RAA, the concept of insurance breaks down because it means you know people either will or won't experience the insured event.
Consider the extreme case: let's say technology allows an external observer to directly read your thoughts as they happen.[2] Hey, people that think worse things are riskier, so what's the problem if they get higher rates? If you don't want higher rates, don't think about cutting through traffic and running red lights!
I just hope no one breaches that database and learns what you think of Suzy...
[1] as a safe driver long resentful of the statistics of my demographic
[2] Because of the "control/choice" issue, assume there's a difference between deliberate and unchosen thoughts and the device only takes the former.