<html><head></head><body>Another problem with autonomous vehicles is that at first there will be little safety data. Insurance runs on being able to predict risk well, so in domains where there are few data-points there will be an extra uncertainty premium. If the vehicles actually are very safe there might still be little data after years of using them, keeping insurance rates higher than they "should" be. But I am not worried about that: there will likely be more than enough accidents due to humans and data gathered from near misses.<div><br></div><div>The real problem might be correlations. Human make mistakes independent of each other. But if all cars of the same make or in the same road system act in the same way - especially if it is a distributed problem like a hack in their road network - then there might be correlated losses. That is the kind of things that break insurance. <br><div><br></div><div><br><br>Anders Sandberg,
Future of Humanity Institute
Philosophy Faculty of Oxford University</div></div></body></html>