It has been proposed that one reason for the success of Homo sapiens is our advanced learning abilities. Theoretical models suggest that complex cultural adaptations can arise from an optimal mix of (1) individual learning that is of sufficient accuracy plus (2) social learning that is of sufficiently high fidelity and is payoff-biased. Here I review the findings of a series of experimental studies of human learning, designed to simulate the kind of technology-based tasks that our ancestors would have faced. Results of these studies support the predictions of the models, and show that contemporary humans’ learning strategies are broadly adaptive. Performance typically improves through effective individual learning and payoff-biased social learning. The latter crucially allows participants to escape low-fitness locally optimal artifact designs and jump to higher-fitness designs, assuming a realistic multimodal adaptive landscape underlying artifact fitness. On the other hand, people also exhibited predictable flaws in their learning, such as the copying of neutral traits exhibited by successful models along with their functional traits (i.e., cultural hitchhiking), and an unwillingness to share information with others under certain circumstances.