In practice it's a cargo cult ritual that leads to the conclusion that we should give within epsilon of zero fucks about anything that happens N years from now for sufficiently large N (which you get to choose as the author of an analysis).
That's partly because there's no mathematical difference in exponential discounting for the role played by 'pure rate of time preference' and the 'good' discount rate. It's always just 3.5% for no particular reason other than that's what in the back of the Treasury handbook written before 2008 during the Great Moderation.
Oh, and is 3.5% a reasonable number? Sounds kinda like inflation/mortgage rates, right? But if you look at CPI $100 dollars in 1817 is, equivalent to roughly $2k dollars in 2017. But would I rather spend this 'same' amount of money in 1817 or 2017? It's a no-brainer as long as I prefer penicillin and smart phones to legal opiods and vintage mustache waxing. This is really hard stuff to model, which is a clue that it doesn't get done.
If Ord and Wiblin could pick one easy fight to change about government policy making and promote longtermism it would be savaging the use of discounting in long term policy decisions rather than beating around the bush.