AI Can Now Pick Where to Build a Library — Without Anyone Cheating

Imagine your city has to build one new fire station, and every neighborhood gets to say where they'd like it. Left alone, everyone has an incentive to exaggerate — claim to live farther out than they do, so the station lands closer to them. Mathematicians call a rule "strategyproof" when lying simply doesn't help; the honest answer is always your best move. That's the setting this new paper studies.

The new twist is adding an AI prediction. Instead of relying only on what people report, the rule blends in a computer's guess at the ideal spot, weighted by how much you trust it. Think of it like a GPS that suggests a route, then adjusts based on where drivers actually are. If the prediction is accurate, the final location gets noticeably better. If the prediction is wildly off, the system still falls back to a reasonable answer rather than something absurd.

The paper works out the exact math for this trade-off across many shapes and sizes of problem — from a flat map to high-dimensional spaces — and shows the results are as tight as possible, meaning no cleverer version of this rule can do better. The confidence dial matters: trust the AI too much and a bad guess hurts more; trust it too little and you throw away a genuinely helpful shortcut.

Why should a non-mathematician care? Because the same question shows up everywhere: where to put a warehouse so delivery routes are shortest, where to site a cell tower, how to divide a shared budget, how an app picks what to show you. As AI predictions get baked into real decisions, the interesting question is no longer "can the AI guess?" but "what happens when it guesses wrong, and can people game it?" This paper is one careful answer for a very common kind of choice.