This paper tests experiential learning as a debiasing tool against gambling and lottery behavior in South Africa. We implement a simple, interactive dice game that simulates worsening winning odds of rolling sixes as more dice are added to the game. The analysis exploits two levels of exogenous variation, first from random assignment into the debiasing game, and second from the number of rolls it takes to obtain the sixes. Treated individuals who needed above-median number of rolls to obtain simultaneous sixes are significantly less likely than the control group to gamble or play the lottery in the following year. The converse is true for individuals who needed below-median number of rolls, suggesting a perverse treatment effect among this group. We also find suggestive evidence that the debiasing affected the sensitivity to varying winning odds. Changes in entertainment utility or risk preferences cannot explain these findings, rather the results are consistent with changes in risk beliefs.