I think you misunderstood. powrtoch proposed having a set of questions in which each individual question is itself very complicated. For example, being a non-practicing Catholic with exactly three children is only one small facet to a single question. By or'ing a bunch of really specific questions together you can come very close to getting exactly 50% of the population to answer yes to a single question.

That's kinda cheating though isn't it? Like chaining a dozen statements on one line with semicolons and going, "look I can write that program in one line!"

That depends what constraints you choose to define on the problem - if they need to be knowable, memorable,... then yes probably. Anyway, sz4kerto has a good comment about using a Karnaugh map which might help see how this works - it's a lot cleverer than just chaining things randomly - but does break a lot of hypothetical arbritary restrictions.

The example was one question with a bunch of OR operators that when combined, would equal exactly 50%.