Warren Powell provided interesting comments the the latest NY Times article “When it comes to math, AI is dumb” that states: “Early computers followed rules. AI follows probabilities. But in mathematics, there is no probable answer, only the right one”.

Probabilists will argue (correctly) that we have very elegant mathematics for describing probabilities. After all, the world is stochastic, and we need to model this uncertainty. Industries like airlines and hotels have mastered the art of modeling an uncertainty as part of their revenue management problems. @optimaldynamics models the uncertainty of shipper demands when planning fleets.

It is true that when making decisions, we ultimately have to pick a single decision despite having to consider its impact on an uncertain future. And in theory there is a single optimal decision, not just in terms of maximizing performance now, but which maximizes performance into an uncertain future. However, finding what we think (and hope) is the best decision, we absolutely have to model the uncertainties that affect how well this decision performs. Link