3 ms·
I think other people have sort of answered it, but a more general version might be: If you observe an extremely unlikely event, you should update your Bayesian
by brianpan 4y ago
I think other people have sort of answered it, but a more general version might be:
If you observe an extremely unlikely event, you should update your Bayesian prior about how likely that event is.
- taeric 4y agoMakes sense. I was stuck with the idea that "seeing a car on the side of the road" is closer to seeing a tree that had been struck. Actually seeing it get struck would be like seeing the car actively crash. Close in thought to each other, but still different things. Granted, I think your point still fully holds. There is a decay function for the likelihood of repeating events. And if you don't have evidence that what you are looking at is old, much safer to assume it is not and that the circumstances are still primed for it to happen again.
- brianpan 4y agoIt's not the time, it's the unknown/unseen factor that's making it likely or unlikely. You think you are seeing lighting striking a tree. But maybe what you are actually seeing is lighting striking a lighting rod. Lightning will not strike a TREE again because usually it destroys the tree in some way, changing it's height. But for many other things (buildings, mountains, etc), lightning will happily strike it over and over again. Is the car in a ditch for a truly unlikely reason, or there a hidden lightning rod there? Perhaps you shouldn't assume a seemingly unlikely event won't happen again.
- namaria 4y agoRepeating and periodic are not the same. Seeing a singed tree or a spun out car tells you nothing about how frequent these events are. They do tell you this place is likely to see those events, at any rate.