The drunkard and the lamplight

I’m a lame jokester, so I’ll cut straight to the point:

A man comes home late after a night of heavy drinking. He reaches for the keys, but they fall out of his hands. Immediately, he goes to the lamplight on the corner and gets on all fours.

His beer buddy asks: ‘hey, what are you doing? Your keys fell here somewhere!’

He replies: ‘Yes, but this is where I can see!’

I tell this story not so much as a joke, but as a parable mostly applicable to people dealing with science (although astute readers will no doubt find it useful for many other areas of human experience). This is the dark side of the Golden Hammer Syndrome: “If the only tool you have is a hammer, then everything looks like a nail.”

We use models not because they are The Truth, but because they are useful. Repeat that to yourself until it doesn’t make sense.

Don’t believe me? Consider the Newtonian and Lagrangian models of mechanics. Conceptually, Newton tells us that things move according to a few key principles (things stay the way they are unless a force disrupts that state). Conceptually, Lagrange tells us that things move following paths of least action.

Which one is correct?

Well, it turns out both models correctly predict how things move. Moreover, you can deduce one from the other. So again, which one is correct? Do things move the way they move because of disruptions to their inertial state or do they move because they try to follow paths of least resistance?

The short answer is: we don’t know for sure.

Before any undergrads try and attack me, consider: have you seen with your own eyes how particles move? Have you come up with a definitive way of determining why particles and planets move the way they do?

No, those models tell us how things move. They frame the main ideas in a conceptual framework that can be translated to mathematical language. The conceptual framework allows us to start making useful questions and assumptions that can be tested in a systematic way. But they don’t answer the philosophical question of why.

What does that have to do with the drunkard and the lamplight?

Any tool or model can become a lamplight. In this context, it’s something that is used regardless of whether it’s really useful. Models are—supposedly—built on a determined set of assumptions, boundary conditions, and other restrictions. The model makes a correct prediction and away we go using it in the future, running the risk of forgetting the underlying assumptions.

My go-to example these last few years is blockchain technology. Sure, it works for some things and under some circumstances it can be a powerful tool. That doesn’t mean it should be used for everything.

In theory, can you use blockchain to build a body of knowledge like an encyclopedia? Yes, you can. Is it somehow better, friendlier to its users, more secure, more robust…? In theory, you can argue yes to all of those. In practice, however, it’s not.

Can you use blockchain to hold national elections? Well, in theory, a blockchain can have many useful features for such a procedure. In practice, it’s hard to argue that the cost of implementing such a system outweighs its benefits. Paper ballots are used because it’s very hard to beat them in terms of simplicity, trust, accessibility, third-party verification and high cost of scalable attacks.¹

Ironically, this fallacy can be a trap for proponents of new technologies, not just for people clinging to known models. One must look for answers where the answers can be found and not just where you can make some fake “progress.” Sure, you may get results of some kind, but that doesn’t mean you will fundamentally get closer to what you want.

The world at large suffers from this. We should learn to recognize when one is looking under the lamplight because it’s where the light is.

Corollary: There’s a second interpretation:

He uses [statistics] the way a drunkard uses a lamppost: for support rather than for illumination.

1. Before anyone comments: I’m not saying paper ballots are perfect, and I’m not saying blockchain elections will never happen. But the paper ballot system has many good features combined that currently blockchain technology cannot reasonably offer.

   Sure, blockchain systems can be created that address some of those concerns, and maybe there’s one that addresses all of those, but blockchain—or whatever proposed technology to replace paper ballots—also needs to consider the costs of adoption, and these are not just financial in nature. A new voting system needs to be—to mention two specific traits—both simple enough to be trusted by the population at large and complicated enough to be hard to disrupt. As 4chan has demonstrated before, if the cost of increasing an electronic tally by one or one million is the same, the system can be abused and there’s an incentive to do so. It’s hard to stuff a box with a hundred extra ballots without immediately arising suspicions.