There doesn't need to be a special mechanism for power to corrupt; normal reinforcement learning should work perfectly well. When you're corrupt, you take actions to benefit yourself instead of those you're supposed to be benefiting. And if those actions do indeed benefit yourself, well, then, that's obviously the kind of thing that reinforcement learning is designed to teach you to do. You take the bribe, or set up a harem, or whatever, because being corrupt means that you are doing things that feel good to you (and are therefore reinforcing) instead of things that benefit the rest of the group.
It's easy to say you won't give into temptation when you've never been tempted before, but it's a lot harder to say that and also be right.
There's another mechanism which is a bit more like paperclipping: rulers come up with random ideas, which they think are doing good because their yes-men say so. (Example]. So you have two mechanisms, one which can go anywhere, and one which converges onto a narrow set of features, such as having multiple sexual partners. In view of the second mechanism, it becomes clear what a great piece of social technology the idea of an Official Opposition is.
Today's post, Why Does Power Corrupt? was originally published on 14 October 2008. A summary (taken from the LW wiki):
Discuss the post here (rather than in the comments to the original post).
This post is part of the Rerunning the Sequences series, where we'll be going through Eliezer Yudkowsky's old posts in order so that people who are interested can (re-)read and discuss them. The previous post was The Ritual, and you can use the sequence_reruns tag or rss feed to follow the rest of the series.
Sequence reruns are a community-driven effort. You can participate by re-reading the sequence post, discussing it here, posting the next day's sequence reruns post, or summarizing forthcoming articles on the wiki. Go here for more details, or to have meta discussions about the Rerunning the Sequences series.