What economics knows and doesn't know about alignment

The economy is a collective intelligence, and in some respects it is vastly smarter than any human being, able to coordinate incredibly complex production and allocation activities over billions of people, thousands of miles, and tens of years. The question of how to align the economy’s interests with our own is therefore one of the most important questions of modern society.

In one sense, economics knows a tremendous amount about alignment. It’s a solved problem, at least in theory: You use prices. Conflict between agents is conflict over resources, and prices resolve that conflict by getting agents to adopt plans that are mutually consistent with each other’s. The result is alignment between agents—although the question of what they’re aligned to may be another example of a binding problem, which I’ll write about later.

I don’t perceive that economics knows very much about alignment across scales, however. Prices align humans with humans, and bioelectricity aligns cells with cells, but what aligns humans with cells? If a new technology was invented that could replace cells with artificial cells that work better from a human perspective, I think most humans would cast aside the trillions of cells that currently compose them in favor of the new, better alternative.

It’s also important to note that economic alignment doesn’t mean agreement or shared values, at least not in the usual sense. Prices coordinate people, which is alignment in one sense, but it doesn’t mean people end up agreeing with each other about fundamental values. The reason for this is that fundamental values aren’t a thing and don’t actually influence behavior in any way. Instead, people end up agreeing on values as measurements, those measurements being the quantities that bring people’s plans into balance. But if you mess up the prices, or get rid of them, then these previously “aligned” agents will start disagreeing with each other and forming conflicting plans pretty much instantly.

I also don’t know that economics knows much about aligning the economy with people, just like we don’t know much about aligning people with the cells that compose them. We don’t usually regard the economy as having separate principles of motion from what are implied by the aggregate behavior of the people who compose it, but we do think of people as having preferences and abilities that are not found in any of their cells. Nor do we know how to get the economy to prioritize our well-being any more than we know how to get humans to prioritize their cells.

My best guess as to what the economy prioritizes is being a good model of relative scarcities. However, “good” does not mean accurate but rather useful for some purpose. I don’t know what that purpose is, other than that it’s probably incorrect to think of it the way we also incorrectly think of our own purposes: as stored, prescribed goals that are instantiated distinctly from perceptions and actions and acting as a bridge between the two.

One thing that economics does know is how to talk to alien systems about what their plans are and how best to manipulate certain parameters to align those systems toward an outcome. The solution is finance, where instead of trying to come up with a complex model of how some system behaves, you just ask the system what it’s going to do and what it needs you to do for certain outcomes to be achieved.

I think it’s plausible that finance can be used to communicate down a scale: to ask cells what we need to do to get them to achieve certain outcomes, for example. Maybe finance can also be used going up a scale: to ask the economy what it needs us to do to get certain outcomes from it. Perhaps this method can be used to align AI as well, though I don’t have any specific ideas about how to do so.