[Previous] OpenAI Fires Then Rehires CEO | Home | [Next] Critical Fallibilism Introduction

Government Policy Proposals and Local Optima

Suppose you can influence government policy. You can make a suggestion that will actually be followed. It could be on any big, complex topic, like economic policy, COVID policy, military policy, etc. What will happen next?

Your policy will probably be stopped early or late. It will probably have changes made. There will probably be other policies implemented, at the same time, which conflict with it.

Most people who influence policy only do it once. Only a small group of elites get to influence many policies. This post isn't about those elites.

Most people who try to influence government policy never have a single success. But some are luckier and get to influence one thing, once.

If you do get listened to more than once, it might be for one thing now, and something else several years later.

If you only get to affect one thing, what kind of suggestion should you make?

Something that works well largely independently of what other policies the government implements. Something that's robust to changes so it'll still work OK even if other people change some parts of it. Something that's useful even if it's done for too short or long of a time period.

Also, you'll be judged by the outcome for the one idea you had that was used, even though a bunch of other factors were outside of your control, and the rest of your plan wasn't followed.

If you suggest a five-part plan, there's a major risk: people will listen to one part (if you're lucky), ignore the other four parts, and then blame you when it doesn't work out well. And if you tell them "don't do that part unless you do all the rest", first of all they may do it anyway and still blame you, but otherwise you're just not going to influence policy at all. If you make huge demands and want to control lots of things, and say it's all or nothing, you can expect to get nothing.

In other words, when suggesting government policy, there's a large incentive to propose local optima. You want proposals that work well in isolation and provide some sort of benefit regardless of what else is going on.

It's possible to choose local optima that you think will also contribute to global (overall) benefit, or not. Some people make a good faith effort to do that, and others don't. The people who don't have an advantage at finding local optima that they can get listened to about because they have a wider selection available and can optimize for other factors besides big picture benefit. The system, under this simplified model, does not incentivize caring about overall benefit. People might already care about that for other reasons.

When other people edit your policy, or do other policies simultaneously, they will usually try to avoid ruining the local benefits of the policy. They may fail, but they tend to have some awareness of the main, immediate point of the policy (otherwise they wouldn't listen to it at all). But the overall benefit is more likely to be ruined by changes or other policies.

This was a thought I had about why government policy, and political advocacy, suck so much.

Also, similar issues apply to giving advice to people on online forums. They will often listen to one thing, change it, and also do several things that conflict with it. Compared to with government policy, there's way more chance they listen to a few things instead of only one. But it's unlikely they'll actually listen to a plan as a whole and avoid breaking parts of it. And when they do a small portion of your advice, but mostly don't listen to you, they'll probably blame you for bad outcomes because they listened to you some. These issues even apply to merely writing blog posts that discuss abstract concepts: people may interpret you as saying some things are good or bad, or otherwise making some suggestions, and then listen to one or a few parts, change the stuff they listen to in ways you disagree with, screw up everything else, and then blame you when it doesn't work out. One way bloggers and other types of authors may try to deal with this is by saying fewer things, avoiding complexity, and basically just repeating a few simple talking points (repeating a limited number of simple talking points may remind you of political advocacy).

Elliot Temple on December 13, 2023


Want to discuss this? Join my forum.

(Due to multi-year, sustained harassment from David Deutsch and his fans, commenting here requires an account. Accounts are not publicly available. Discussion info.)