Uncertainty is at the heart of statistics. How to contend with uncertainty, both technically but also philosophically, separate the paradigms comminating social science research. The causal paradigm see uncertainty as something that can be managed, distributed, and reapportioned. The predict paradigm is open to certainty – or at a minimum, agnostic – about it. In the former, the medicalized handling of uncertainty sometimes gives it manipulators a false sense of having dealt satisfactorily with uncertainty. But, as H.L. Mencken has accurately quipped, ‘Penetrating so many secrets, we cease to believe in the unknowable. But there it sits nevertheless, calmly licking its chops.’ For the later, as we have seen, uncertainty troubles the interpretability of results.

 
Even for AI, uncertainty is the new devil in the details. Prediction – even with LLMs – requires us to make guesses about future outcomes and to do so, forces us to make assumptions. Most commonly, the ‘undisturbed’ assumption that the future will look like the past. However, new models – the world model – for examples asks whether we can predict even if we aren’t totally sure how the future will be.

The next large jump in our abilities – academically, industrially, philosophically – will be in our handling of uncertainty. The sooner we understand the best ways to work with uncertainty, the future will look quite different. And we’ll already know.