Some predictions are based on statistical models. Others require models of alternative worlds, because the world changes, and so does statistical expectation. Causal models can be used to represent alternative worlds. Of course, neither form of prediction can account for unknown unknowns, so they're both inherently limited. Moreover, humans are limited because individually we're ignorant; we depend on others to complete our understanding. These are facts that individuals tend to forget. Accurate prediction in a context with any complexity always requires the integration of advice from multiple experts. The trick is to find a way to represent and reason with a causal model that reflects a community's expertise.

April 11, 2020