To deeply explore the variations in individual risk-taking behavior within group decisions and their cognitive foundations, and subsequently optimize group decision-making effectiveness, this study employs the Balloon Analogue Risk Task (BART). It compares individuals' sequential risk decisions across different decision-making modes (individual decision-making, joint decision-making, and alternating decision-making) and under varying social distance group conditions. Furthermore, a Bayesian model of reinforcement learning is utilized to construct a model of the decision-making process. The research reveals that, when using traditional behavioral indicators (average number of balloon pumps), compared to individual decision-making, individuals exhibit more risky behavior when decision-making involves two people. However, Bayesian model fitting reveals that this is not due to an increase in individual risk preference but rather an enhancement in their prior belief that the balloon will not burst. Moreover, this increase in prior belief is more pronounced in groups with close social distances.