A Reversed Turing Test of Human Random Number Generation

Abstract

Based upon distinctions between true randomness, probabilistic randomness, and stochastic randomness, a reversed Turing Test was performed to compare human-generated to computer-generated random numbers. Tests of three hypotheses showed that humans more often fail to behave randomly when assessed by a “probabilistic interrogator” (based on distribution-free nonparametric tests) than by a “stochastic interrogator” (based on parametric testing), that computer-generated numbers displayed both probabilistic and stochastic randomness, and that human failure to pass the reversed Turing Test may be attributed to a nonrandom response pattern embedded in the group data corresponding to the highly automatized human counting skill. In addition to supporting the ubiquitous observation that humans do not behave randomly, these findings suggest that humans, unlike computers, may spontaneously interpose relative amounts of order that preclude successful random generation requiring relative amounts of disorder.