On Targeting Proxies

I've been thinking about how we, as biased humans, often set absurd targets. We might think we're heading in the right direction, where the reality is completely opposite. Analyzing how this happens, I found they usually have something in common: these absurd targets are always "proxies" created to measure something else. They don't measure what people think they do, and only in the best cases can provide useful information. Many times, however, they are completely misleading.

In many situations we don't need proxies. For example, you can measure the amount of people in a group just by counting them. Or if you want to sort the people in that group by height, using a ruler you can get their heights.

There are more complex situations where measuring is harder or just impossible, and still proxies are not required. For example, when measuring Earth's mass, we can't directly measure it, but we can do it indirectly. We can measure the acceleration of a known mass on the Earth's surface, with that we can calculate the gravitation force, and then, using physic's laws, calculate Earth's mass.

Methods –e.g. counting or measuring indirectly– or tools –e.g. a ruler– could have flaws and errors, but they give you the information you were looking for, often with some error tolerance. Proxies, however, are a completely different story.

Proxies are created for situations when we just don't know any method or tool to measure what we want. For example, how intelligent is this person? How do they compare with other people?

There's no real solution for that. How do we measure intelligence, or happiness? We don't. Here is where we use proxies. They are something we can measure, and then use to guess what we were trying to quantify in the first place.

Classic proxies are using someone's net worth to infer happiness, or someone's score on a test to assess how much they've learnt about a topic. It's obvious –at least for you, I hope– that having more money in the bank doesn't make you happier. You might need some money to be happy, but you'll probably agree that you also need some other things besides money.

I personally feel related with the second example. I wasn't a very good student, at least before university. My scores weren't great, and some teachers thought I wasn't very intelligent, or at least not hardworking enough due to my scores. Many years later, I know my way of learning doesn't match the traditional style of teaching, and I need to understand why a topic is useful before I care about learning it. But once I'm motivated about something, I learn very fast and deeply.

I was lucky my parents thought I was intelligent enough and didn't force me to study more to get better scores. Imagine a different situation where I was pressured to get better scores. I could've studied more, but developing hate towards learning. Or maybe I didn't study anyways but became a master of cheating. At the end of the day, getting better scores means I've learnt more, doesn't it?

I mean, you want to measure how much someone has learnt about a topic, you set up a system where they take tests, the better the score, the more they've learnt. However, that whole system is causing people –at least some– to focus on improving their scores at the expense of learning less. Targeting the proxy is leaving some people further from their original goal. Isn't that completely backwards?

The same thing happens with the "money is happiness" proxy. Some people looking for happiness are ending up way more miserable than they would be if they didn't focus on money. All this while they think they're focusing on becoming happier.

This kind of situation happens when we set a proxy, and the proxy becomes the target. Then we lose sight of the actual goal we had, and just aim for the proxy. I don't mean proxies are bad. I actually think they are necessary. But we should be more aware of proxies –they are everywhere!– and keep in mind the information they give us could be completely unrelated with what we want to measure. Just blindly trusting proxies is the recipe for disaster.

As Goodhart's law says:

When a measure becomes a target, it ceases to be a good measure

Don't let proxies become targets.