Hacker News Reader

Comment by victork2 | Hacker News Reader

Sorry to say that but we cannot trust the weather forecast in simulation for more than 5 days and we trust a forecast for mankind 30 years from now?

I know people like these kind of studies, perhaps it conveys the feeling that "after me there won't be anything" and that we are the last generation but it's simply to complex to model. People have been saying the same things for generation and so far, so good (with some exceptions of course).

I am not saying that it's not going to happen, just that you have to be very careful with prediction such as these, take them as "work of fiction". Also don't forget a possible bias. Apocalyptic vision have always fascinated populations and men around the world but how many have happened?

View on HN

You can't tell whether it's going to rain five days from now very reliably, but you can predict within a degree the average monthly temperature in any city in the world, for any month of the year, five years from now. Of course there are unforeseeable events, but that doesn't mean everything is unforeseeable.

In this case, we've had 40 years to test this forecast, and it looks like they've done a pretty reasonable job forecasting the key variables they were forecasting, 40 years in the future. That seems like a good validation of their model, and we should expect it to do nearly as good a job for the next ten years at least, if not 20.

So I don't think they should be taken as a "work of fiction", even if they might be wrong. The model should be taken seriously, and we should attempt to improve it or propose better models, even if those models have more uncertainty in their predictions.

There's a big difference between predicting an apocalypse based on a psychotic episode and predicting an apocalypse based on a dynamical model of the key variables in a world system.

So the extremely mild winter (in the continental US) that everyone's been saying was so shocking was actually forecast years ago? Or are you saying that this capability has only been developed in the last year or so?

I'm not talking about anything new. Your predictions will be wrong occasionally.

The point of the article is that this particular simulation did in fact accurately forecast mankind's future 30 years into the future, from 1970 to 2000. If you had made your argument against the model in 1970, you would have been wrong. The implication now is that since the model accurately predicted the last 30 years, we should also trust it to predict the next 30 years. You might still argue with the conclusions, but you need to be more specific in your criticisms than "these things can't be predicted", because according to the article, apparently they can be. At least for the last 30 years.

Weather patterns are extremely complex and yet fall within a certain set of bounds. A swing of ten degrees is perceived as a dramatic change. This is because the numbers involved are relatively small.

Resources and populations have much wider swings and the forces that perturb them need time to have an effect. The numbers involved are millions of times larger than temperatures. There's also much higher inertia than the weather.

It's impossible to predict the precise number of births tomorrow, or the amount of coal extracted to the ton, but you can come up with a very accurate estimate for the course of a year or a decade with the right data.

We can't predict the weather five days out, but we can predict the climate ten years out.

False analogy.

"weather forecasting" involves fine-resolution (in both space and time) predictions. This is known to be difficult to do reliably ("butterfly effect").

This is more akin to climate analysis: low-resolution in space (continent-scale instead of ZIP-scale) of and time (decades instead of 24 hours). Much more amenable to statistical modeling.

YMMV.

Reply on news.ycombinator.com