We conducted an online experiment with 243 medical professionals, comparing the number of irrational risky choices between two experimental treatments. In one treatment, we labeled the source of information as AI estimates; in the other treatment, the identical information was labeled as scientific estimates. We defined irrational choices as deviations from normative decision theories represented by cognitive biases, e.g., risky choice framing effects. We further expected individual characteristics of decision makers, i.e., AI anxiety, AI knowledge, and intention to use AI, to influence rationality further. We found a significant reduction in irrational choices when the information was labeled as AI estimates compared to scientific estimates. Contrary to expectations, increasing AI anxiety was associated with fewer irrational choices. However, knowledge about AI and the intention to use it did not influence rationality. Our findings provide implications for designing and implementing AI information systems in medical decision-making processes.