In any adversarial contest it is only ever entropy that wins. Not only this, but the contest and competition (pick a context, any context) is the carrier wave for a logic and information metric of difference that it itself circularly, recursively invokes. A predictive algorithm asserted incorrectly that Nadal would lose because it knew no better but the presence of inaccuracy becomes the rationale for an endless refinement of forecasting approximations.
We might note it is only ever technology that ultimately benefits from conflict, but this remains an analytical steppe too far and is, as a function of political or psychological complexity as much as of ideological immaturity, still broadly unacknowledged. Similarly, algorithmic forecasts are retrospective systems of belief masquerading as future certainties that approximate to truth often enough to self-validate as credible, useful, valuable.
In and as all historical data and assumptive model features, inductive probability refines itself as an abstraction that gains more from its failures than an impossible closure or axiomatic certainty (as projective hubris or commercial momentum) might ever provide. In this way, as though always looking behind us, we manufacture perspectival expectations of accuracy and yet fall, poorly-prepared and backwards, into one future among many. The bias here is that recurring failures of applied machine intelligence are key methods of its own self-propagation.