“Of course, you can opt out on all of this stuff if you don’t want to, but if you want dynamic insurance, for example, that’s paid for by how safe a driver you are and you want coaching and we can say, “Hey, listen, you’re a 45 percent driver. If you leave another five meters between you and the car in front and you do this and you do that, your insurance premium will come down.” That’s a dynamic insurance premium. You don’t pay it once a year.”
The problem with an algoritm deciding how “safe” you are is that, in addition to it almost always being opaque, is that it is currently necessarily immature. My Tesla, for example, is extremely over-cautious, braking suddenly for a car crosssing way in front of me. That’s the kind of driving that’s a cause of more rear-end collisions, for example. The point is that the kind of driving it rewards isn’t meaningfully “safer”, just over-cautious and annoying to other, more mature drivers.
Another way to explain this is to observe how the Tesla autopilot works- in general, it drives like a 14 year-old driver: tentative at times, over-correcting at others, becoming confused at unexpected situations like lanes lines suddenly going away, slowing excessively for corners (and worse, crossing the yellow do-not-pass lines on some turns!, thereby actively trying to kill you) and in general, not smooth at all. This isn’t a criticism of Tesla per se, only to note that the current state of the art of Level 2 autonomy is at the level of a relatively poor driver- yet we are expected to somehow ape that behavior in order to qualify for lower rates.
In my opinion, variable premium insurance products should have to make their algorithms public, and also show, with actual data, that the behavior they are trying to incent is actually demonstrably, meaningfully, safer-- and that the additional premium is proportionally connected to that behavior. Otherwise, it becomes just another scheme to separate people from their money under the guise of “safety”.
“Of course, you can opt out on all of this stuff if you don’t want to, but if you want dynamic insurance, for example, that’s paid for by how safe a driver you are and you want coaching and we can say, “Hey, listen, you’re a 45 percent driver. If you leave another five meters between you and the car in front and you do this and you do that, your insurance premium will come down.” That’s a dynamic insurance premium. You don’t pay it once a year.”
The problem with an algoritm deciding how “safe” you are is that, in addition to it almost always being opaque, is that it is currently necessarily immature. My Tesla, for example, is extremely over-cautious, braking suddenly for a car crosssing way in front of me. That’s the kind of driving that’s a cause of more rear-end collisions, for example. The point is that the kind of driving it rewards isn’t meaningfully “safer”, just over-cautious and annoying to other, more mature drivers.
Another way to explain this is to observe how the Tesla autopilot works- in general, it drives like a 14 year-old driver: tentative at times, over-correcting at others, becoming confused at unexpected situations like lanes lines suddenly going away, slowing excessively for corners (and worse, crossing the yellow do-not-pass lines on some turns!, thereby actively trying to kill you) and in general, not smooth at all. This isn’t a criticism of Tesla per se, only to note that the current state of the art of Level 2 autonomy is at the level of a relatively poor driver- yet we are expected to somehow ape that behavior in order to qualify for lower rates.
In my opinion, variable premium insurance products should have to make their algorithms public, and also show, with actual data, that the behavior they are trying to incent is actually demonstrably, meaningfully, safer-- and that the additional premium is proportionally connected to that behavior. Otherwise, it becomes just another scheme to separate people from their money under the guise of “safety”.