A chaotic system can be stable and predictable most of the time, then suddenly become dangerously unpredictable. The shape underlying the system decides when small differences stay small and when they blow up, so forecasting skill itself rises and falls with the situation.
Uncertainty is not a reason to wait. The question is whether the estimated risk is large enough to be worth reducing. If an outcome would be catastrophic, even a 40 percent chance of it makes acting overwhelmingly worthwhile, just as people insure against serious disability without being sure it will happen.
Randomness in the brain is useful, not just a nuisance. It lets the mind make connections a fully rigid machine never would, and it lets cheap, simplified models of a complex world stand in for costly exact ones. A machine required never to err cannot also be intelligent.
In chaotic systems, understanding the parts and building upward often fails, so studying overall statistical patterns works better than deriving them from first principles. The book even suggests the universe's largest-scale structure may matter as much for explaining tiny particles as the particles do for explaining the universe.
In a chaotic system, a single best-guess forecast hides how much could go differently. Running the model many times from slightly varied starting points shows the spread of possible outcomes and puts a number on the forecast's own uncertainty. That spread is what lets us anticipate extremes such as violent storms, rather than being blindsided by them.