I can accept the fact that on a Roulette wheel (as long as there are no defects or imbalances in the wheel or ball) that the odds are the same each spin and previous spin outcomes have no influence over the current spin. However, if I see black come up 32 times in a row I am betting on red for the next spin.
In a Bayesian sense this would be called updating your prior. You assume the wheel is truly random. After many observations that assumption seems not to hold so you adjust your prior probability that any given spin will land on black to be higher.