AI is an allocator
Strip most consequential decisions down to their core and you find an allocation problem. Who gets the capital. Where the housing goes. Which projects get funded, which patients get seen, which neighborhoods get the investment. We dress these up in politics and process, but underneath they are questions of how to distribute scarce resources across enormous numbers of people and constraints.
Humans are not built for that. A skilled decision-maker can hold maybe a handful of variables in working memory at once and reasons largely by heuristic and precedent. That’s not a knock on them; it’s the hardware. But it means our allocation decisions are made on a tiny slice of the relevant information, and the gap between the choices we make and the choices we could make, given everything we actually know, is vast.
This is the thing AI is unusually good at. A model can weigh millions of variables and competing priorities simultaneously and search for outcomes far closer to a real social optimum than any committee can. Over the coming years more and more allocation will be informed by, and eventually handed to, systems that can actually see the whole board. That’s why AI for government is the highest-leverage frontier of the next five to ten years. It’s not the flashiest application. It’s the one that touches the most lives per decision.
The promise isn’t a machine that rules. It’s a machine that lets us finally use what we know.
I’ve chosen to work on this in housing, and not by accident. Housing is the single largest component of the economy and one of the places where bad allocation is felt most immediately: in cost, in supply, in who gets to live where. Small improvements in how housing capital and policy are allocated compound into enormous differences in people’s lives. If AI-assisted allocation is going to prove itself anywhere, this is where it should.
I want to be careful about the obvious objection, because it’s a good one: an allocator optimizing the wrong objective, or one no one can interrogate, is worse than the messy human process it replaces. Legitimacy, transparency, and accountability aren’t friction to be optimized away; they’re part of the objective. The goal isn’t to remove humans from the loop. It’s to give the humans who are accountable a vastly clearer view of the trade-offs they’re actually choosing between. Done right, that’s not less democratic. It’s more honest.