Probably a fuzzy question. Randomness.

I'm going to prod around again (as I did in earlier questions here on Primes ) this time in another area of vast personal ignorance : randomness. Sorry . . . , well, here we go : 

If we accept, as general principals (spelling deliberate of that word) of randomness, the two "examples" of 1) nuclear decay, and 2) a national lottery ( a bit of a weaker example I suppose on which someone might correct me later), then, I'm presuming that a formula couldn't really be developed to predict outcomes for either of these, except to say that in the case of 1), a decay event will happen, but we don't know when, and in the case of 2), a selection event will happen, but we do know when.

Is it not the case that the randomness of these two scenarios effectively means that no algorithm (including AI algorithms) stands more than a probabalistic chance of predicting a correct outcome ? In the case of the lottery though, its dependency on exact timing is far less of a factor than any determination applicable to the nuclear event? Again in the case of the lottery, the certainty of an outcome can be more or less guaranteed to be a singular outcome, whereas the nuclear decay outcome might contain more than one event, or no event at all. However,  how long should a lottery number selection process be allowed to last if it is to be considered random at all ? 10 seconds . . . ? . . . three minutes . . . one year . . .?

I suppose my question is better put like this : the maths used to guide a spacecraft to Mars is, afaiaa, based on Newtonian principles, (although I imagine that deep down there could be Quantum effects at work). But let's say the trajectory can be accurately computed in advance almost without fail. Then doubtless AI could be applied to optimise and verify such calculations, in which case I imagine the ML involved would be looking to improve the heuristic patterns it develops and follows in order to reach ever more accurate determinations ? So would it be right to see this application as a search for patterns in underlying data (weight, speed, distance, thrust, momentum, acceleration, inertia, etc etc ), and that there is a difference between AI for certain applications (Space Travel being maybe one), and the prediction of nuclear decay, earthquakes, and lotteries etc. ? Does AI become less useful, the more random the data is that it has to deal with ? How does a designer know whether the data his application is tasked with analysing is going to turn out to hold no pattern (and by pattern, I include conditions that can be said to be givens in the Newtonian sense), that and so cannot ultimately reveal anything more informing than what simple probability can tell us about something ?