From a Bayesian perspective, the only way to play and not lose is to not update your priors, at all, forever. (I suppose that "not playing" winds up much the same - you don't have any input on which to falsely update your priors.)

But people who absolutely will not update their priors is not likely to work out well in the long run...