Memory biases in risky decisions from experience

Abstract: When faced with risky decisions, people often choose differently pending whether they learn about the odds and outcomes through explicit description or through repeated experience. In the latter case, when deciding based on past experiences, people must rely on their memories to guide choice. These memories, however, need not be veridical. In this talk, I discuss two ways in which memories of past outcomes can be distorted, affecting choice in risky decisions from experiences. First, people tend to remember the most extreme outcomes encountered (both best and worst), making them occasionally more risk seeking for gains than losses. Second, subtle reminders of past wins can induce people to gamble more. These biases are not uniquely human, and we have observed similar patterns of risky choice in pigeons as well. I interpret these results through a reinforcement-learning model, derived from studies on animal learning, that learns similarly from both real and replayed experiences.