Maximum predictive power and the superposition principle

Recently, there has been a discussion on the origin of the quantum probability rules (Deutsch quant-ph/9906015, Polley quant-ph/9906124, Barnum et al. quant-ph/9907024, Finkelstein quant-ph/9907004). This contribution, which is a slightly reformulated version of a paper published in Int.J.Theor.Phys. 33, 171 (1994), points out the follwoing: To an experimenter the world is a persistent stream of discrete data. All that is certain is that with each observation he/she knows more than before, simply because he/she can now answer the question "Which of the possible outcomes have you just registered?", while this was not possible before the observation. One can ask whether this relentless increase of information entails a specific structure. In particular, how must different observations be related in order to ensure that predictions become ever more accurate, the more past observations serve as input? This leads to the quantum rule for adding the complex square roots of probabilities, and not to adding the probabilities themselves, as classical probability would have it.