Probability weighting as evolutionary second-best