In chapter 3 we’ll take a step back to understand why predicting the future is so hard. Our answer is that its challenges are ultimately not about AI, but rather the nature of social processes; it is inherently hard to predict human behavior, and we’ll see many reasons for this. We’ll review evidence from many efforts to predict the future, from crime to children’s life outcomes. We’ll draw from academic studies as well as the rare cases where commercial products have been subjected to independent scrutiny. We’ll look at prediction of both positive outcomes, such as succeeding at a job or publishing a bestseller, and negative outcomes, such as failing to pay back a loan; all of these turn out to be hard to predict. We’ll also look at less consequential but more easily analyzed prediction tasks such as identifying which social media posts will go viral. And finally, in addition to outcomes about individuals, we’ll look at macrolevel predictions such as the evolution of pandemics. Across all of these domains, strikingly common patterns emerge, which lead us to conclude that the limitations of predictive AI won’t go away in the foreseeable future.