We propose a simple theoretical model of supervised learning that is potentially useful to interpret a number of empirical phenomena. The model captures a basic tradeoff between sheltering the child from the consequences of his mistakes, and allowing him to learn from experience. We characterize...