Is there such thing? For example, flipping a coin does not generate a random result, because it is possible to calculate the outcome based on the force applied to it's shape and weight... taking account for wind resistence and so on.
It appears that it is possible to build a device that produces uniformly predictable coin tosses (The Not So Random Coin Toss
But the existence of a non-random coin flipping machine does not disprove the randomness of coin-flipping in general, just as a card sharp's ability to deal from the bottom of a deck does not disprove the randomness of card-dealing in general. Indeed, the sole difference between the results obtained by the coin-flipping machine and those obtained by traditional coin-flipping is precisely the randomness
of the results.
Predicting the results of a traditional coin flip by relying upon such informatin as shape, weight, wind speed, force, velocity, and so forth is simply not possible: one would not only need to calculate all of those factors before the actual flip, but one would also need to calculate all of the relevant factors during the course of the flip
, up to the point where it landed either heads or tails. Yet all of those
factors would, in turn, have to be accurately predicted in order to say that the results were predicted; for if we said that those factors would only need to be measured
(rather than predicted), we would necessarily have to admit that those factors would have to be measured at all stages of the flip-event up to the point when the coin landed
. And if the measurement coincides exactly
with the event, then there is simply no point in talking about prediction at all, since the measurement and the "prediction" are the same thing. In other words, we could no longer reasonably talk about "predicting" the results of the coin-flip if what we're really doing is "describing" the coin-flip.
We can, then, only accurately retrodict
the results of a coin flip by using mathematical calculations: we cannot predict
those results (unless, of course, we are using a predictably reliable coin-flipping device). Thus, randomness depends upon whether one is viewing an event ex ante
or ex post
. Viewed as a completed event, it can be accurately retrodicted; viewed as a future event, it cannot be accurately predicted.