you can never control all of the parameters yourself, true, but think of it this way:
the cockpit of an airplane has hundreds of switches and nobs. suppose we want to ask the question "is there some way the pilot could toggle switches or flip knobs that would open the doors in flight?" you're never going to be able to guarantee that the doors won't open, but you can guarantee that the circuit alone won't cause them to open.
or, consider that many games are np-complete. if you are playing such a game against another player, whoever gets the better solution will win.
in addition, the original author's claim that the 95% solution is "vastly easier and entirely sufficient" _may_be true for the numerical optimization problems given in the example, but there are most certainly np-complete problems which cannot be approximated to any degree of accuracy.
for example: given a graph, the max-clique problem asks 'what is the largest fully-connected subgraph?' this problem arises very naturally in social networks; it's equivalent to asking 'what is the largest group of people who are all friends with each other?' This problem cannot be approximated within ANY bound of the optimal solution unless P = NP. The same problem arises in computational biology and chemistry as well.
That's a very good point, but my concern is that the provided examples dumb down issues to a level where this is not easily understood. I mean, who gets closer to grasping this nuance bu supposing a woman building towers of rocks on the beach?
Actually, I think the cockpit-switches-example would be a valuable example.
the cockpit of an airplane has hundreds of switches and nobs. suppose we want to ask the question "is there some way the pilot could toggle switches or flip knobs that would open the doors in flight?" you're never going to be able to guarantee that the doors won't open, but you can guarantee that the circuit alone won't cause them to open.
or, consider that many games are np-complete. if you are playing such a game against another player, whoever gets the better solution will win.
in addition, the original author's claim that the 95% solution is "vastly easier and entirely sufficient" _may_be true for the numerical optimization problems given in the example, but there are most certainly np-complete problems which cannot be approximated to any degree of accuracy.
for example: given a graph, the max-clique problem asks 'what is the largest fully-connected subgraph?' this problem arises very naturally in social networks; it's equivalent to asking 'what is the largest group of people who are all friends with each other?' This problem cannot be approximated within ANY bound of the optimal solution unless P = NP. The same problem arises in computational biology and chemistry as well.