Researchers have long theorized about the processes through which family background and childhood experiences shape life outcomes. However, statistical models that use data on family background and childhood experiences to predict life outcomes often have poor predictive performance. In this talk, we present results from three interrelated studies of the predictability of life outcomes: a scientific mass collaboration involving hundreds of participants, a high-throughput study using hundreds of machine learning pipelines to predict hundreds of life outcomes, and a qualitative study involving in-depth interviews with 40 families. Collectively these studies help to assess and understand the limits of predictability of life outcomes, which has implications for social science theory and for algorithmic decision-making in high-stakes settings.