I think it’s similar to weather forecasts that say things like there is an x% chance there will be a thunderstorm at time y at location z. Do you think such weather forecasts are meaningless?
5 Likes
Maybe not meaningless but wrong more often than not 
2 Likes
How can a prediction like an x% chance there will be a thunderstorm at time y at location z be wrong? (unless x is exactly 0% or 100%)
You can only evaluate the quality of your model by collecting statistical data to compare a large number of its predictions with what actually happened.
8 Likes
Exactly.
And thankfully for weather you have many events of “rain” so it is possible to evaluate a prediction model.
When it’s a single non-replicable event, it’s in principle impossible to determine whether a prediction is correct or wrong. However it doesn’t mean it’s worthless. The predictions for this asteroid relies on known physics and we have little reason to doubt it’s accuracy, even if we won’t be able to prove whether it was correct afterward.
6 Likes
The probability of collision is calculated given that our data are incomplete and lack precision. It says that if we pick random initial conditions in our model of the universe, that match our current state of knowledge, then in 0.3% (or whatever) of simulations, the asteroid will hit the earth. But every day we gather new data and our knowledge of the universe improves, so our model, and hence the probability, evolve.
6 Likes
Weather forecasts is the first example of nonsense probabilities given in that book. That day will never happen again so who knows if the probability was right or wrong
1 Like
Well, the trajectories of asteroids and planets are routinely predicted and those trajectories can also be routinely measured (and have been for many centuries already) to allow comparison with the predictions. So the accuracy of the model is well known.
3 Likes
I don’t buy that. It seems to me that taking that position means rejecting statistics (and most scientific knowledge) as a valid means to make any kind of prediction.
5 Likes
That’s a strange position. A given event never happens twice in exactly the same way, by definition. If we apply this reasoning, then the entirety of forecasting becomes meaningless.
2 Likes
Let’s see it like this: I’m rolling 12 six-sided dice, and my goal is to get exactly 4 ones. Since, on average, only 2 dice are expected to be ones out of 12 (since the probability is 1/6 to get a 1 per die), it’s hard to do that, so the chance is only 8%.
Now you roll the first die, and it turns out to be a 5. This makes it even harder to get exactly 4 ones, so your probability decreases to 7%.
Next you roll the second die, it is a 2. Again, it gets harder, probability drops to 5%.
But then you roll a 1! Now you only have to roll exactly 3 more 1’s in the remaining 9 rolls, which increases the probability to 13%.
Next roll, you get another 1! The probability increases to 26%, with 8 rolls remaining.
The next roll is a 6. Your probability has dropped again to 23%.
And next you roll a 3, the probability becomes 20%.
And a 1, the probability becomes 40%.
Etc.
The moral is, the information you gain can change the probabilities in either direction.
9 Likes
I suspect that perhaps @teapoweredrobot does not reject those probability calculations, but he thinks it’s useless/unsatisfactory because you can’t actually predict the outcome of the next die roll (which is correct, but irrelevant IMO).
2 Likes
You can’t predict the outcome of one roll, but you can reasonably predict what to expect from many rolls.
It’s a scaled down version, but if I’d roll 6 million dice, I’d expect roughly a million to have a 1 facing upwards, with 2 million facing a 1 being absurdly unlikely.
4 Likes
2 Likes
The difference with actual events, is that prediction of outcomes is not telling you whether it happens or not. It is an estimate of how many of the modelled circumstances will coincide with the scenario you’re trying to avoid / reach. But, this assumes we know our models are correct.
Suppose in the above dice example, I didn’t tell you that the last three dice are heavily weighted, and have almost no chance of being a 1. That significantly changes the probabilities, but you have no way of knowing that until you get to roll those dice several times and calculate their weight distribution.
With this asteroid, we do know a lot, like, we have a decent estimate of everything in the solar system that could influence it. But we also don’t know a lot, since a 1% error of how much the asteroid weighs can significantly change its path. Moreover, gravitational computations can be relatively stable, such as with the most heavy objects orbiting in our solar system (the planets, their moons, the sun). But with objects of equal weight or for objects with an insignificant amount of weight, gravitational computations are chaotic: you can only predict them for a short time into the future, after which miniscule perturbations will have enormous effects.
1 Like
Also, the asteroid is tiny and far away, making it much harder to accurately predict this astronomical event than (say) a solar eclipse or landing a man on the moon.
probability is not part of the event, it exist in your mind
when you have poor vision and something green is moving towards you, you can be 1% sure that it is crocodile. When it is more near and you see blurry tail, you will be 10% sure that is crocodile
when you found glasses, you will be 100% sure it is crocodile
3 Likes
Nah, too late!
2 Likes
I think that’s a good way to think about probabilistic forecasts, namely as a probability conditioned on the partial information that is available.
But additionally I’d like to point out that the current physics model is not deterministic. So even if we had all information, the predicted future is uncertain.
3 Likes
So, exactly like rolling dice 
2 Likes
In theory, assuming that the laws of physics are actually deterministic, then the result of dice rolls can be calculated given complete information of the universe.
4 Likes