Compared to the common “first past the post” voting, yes it is a good thing. This is because the common voting system encourages a “two extremist parties” system where everyone is effectively voting “greater horrible evil vs lesser horrible evil”.

https://youtu.be/s7tWHJfhiyo

I don’t think the problem with having only two viable parties is that they will both tend to be extreme (and that doesn’t strike me as empirically true at the US federal level). Rather, I think the problem is that parties have aggregate power, which means that individual policy preferences get bundled together in ways that may not match an individual voter’s preference. What if I prefer one policy that’s only part of the party X platform, and another that is only part of the party Y platform? Or, another way to think about this party aggregation: what if I prefer the candidate from party X for my local congressional district, but I can’t stand the thought of the likely Speaker of the House of party X?

Here's the problem though; IRV doesn't really do the thing you want it to. It can actually make an extreme candidate win even when voter opinion is centered around a moderate candidate. It's very obvious in this visualization (from the site linked a few comments ago): http://zesty.ca/voting/sim/54,47_77,64_13,10_irv.png

> It's very obvious in this visualization (from the site linked a few comments ago): http://zesty.ca/voting/sim/54,47_77,64_13,10_irv.png

Just a friendly note: this visualization means nothing without context, and is far from "obvious". Unfortunately, I'd place this in the category of "things that are obvious to people who already believe it". Is there a better way to go about explaining your idea?

Sorry, I should have mentioned again the above link, which provides context: http://zesty.ca/voting/sim

To paraphrase, this is a simulation which places candidates and voters on a two-parameter preference space. This is useful because two dimensions are easy to visualize, but the same intuitions apply to higher-dimensional spaces.

Each point is colored to answer "which candidate wins when the popular opinion is normally distributed about this point?". That means the fractured visualization shows IRV's non-monotonicity, and how even when popular opinion is close to a centrist candidate (here in red) it might be that the voting system results in the election of an extreme candidate (green).

What is obvious about that image?