a complex systems model is almost never as good as its authors think it is

(And that includes models I've built)

Corollary: a complex systems model is *never* as good as press officers and journos think it is.

io9 | Alasdair Wilkins | A computer that can predict hurricanes up to ten years in advance

I'll believe it when I see it in 2020.


PS As far as I can tell from the abstract of the paper this is based on, the system doesn't predict hurricanes, it predicts hurricane frequency across a season.

PPS At first glance the authors seem to have attempted to keep their training and validation data separate, which is better than a lot of climate models I have seen, but I can't tell what assumptions are built in that may have been drawn from the test period data.