In Monte Carlo, there is something called "stratified" or "biased" sampling. It's used to overcome the problem that if you are trying to sample a highly nonunif

In Monte Carlo, there is something called "stratified" or "biased" sampling. It's used to overcome the problem that if you are trying to sample a highly nonuniform probability distribution, a "fair" sampling based on the frequency distribution won't tell you about the tails unless you use a lot of samples. Suppose that we are trying to sample a distribution that is, say, 90% male and 10% female. If we do that with a fair sampling, and take 5 samples, chances are that we will end up with no females in the sample. Someone observing the sampled distribution would draw the conclusion that there are no women in the population from which the samples are drawn. The solution is to not sample fairly, because to do so results in oversampling the common cases and undersampling the rare cases. Instead, you construct your sample such that it samples the rare cases with an artificially higher probability. You of course needs to take this into account when interpreting the samples, but that's fine. So for the topic at hand: if you want to expose someone to the widest possible range of viewpoints, sampling fairly is an inefficient thing to do. Instead, you need to split the samples into several strata, be they gender, cultural background, wealth, anything you can think of, and then make sure that all parts of those distributions are represented in your sample. Otherwise, with a limited number of samples, you'll never encounter the rare instances. In other words, you need to "carefully structure inequality in", as you say.