AI Matchmaking Just Got Faster — And a Bit Easier to Game
The same math picks your dates, jobs, and shopping suggestions.
Every time an app decides which job to show you, which product to suggest, or which person to match you with, it is solving a "matching problem." Traditionally that means comparing everyone's full ranked wish list against everyone else's — a slow, messy job that gets out of hand fast. Irene Aldridge's new paper, posted on the research site arXiv, offers a shortcut: describe people and things by their traits, then use one mathematical squeeze (she calls it a spectral projection) to collapse the whole tangle into a single sorted line. Sorting a line is something computers do blindingly fast, even with millions of entries.
The payoff is speed and fairness. Her method hits the best possible outcome on a measure economists call Nash Social Welfare, which balances how happy everyone is overall with how evenly that happiness is spread around — a fancy way of saying it avoids giving one person everything. She tested it in an AI shopping assistant that buys things on your behalf, and the method correctly predicted one case where the assistant would succeed and one where it would fail. A 100-run stress test backed up the results.
Now the catch, and it is a real one. The system is not strategy-proof, which means it can be gamed. The paper openly includes an example of someone lying about their preferences and walking away with a better deal. In practice, that would mean a job candidate inflating skills, a seller fudging product specs, or a shopper hiding their true budget. Any company using this would need to add safeguards — or accept that clever users will work the system.
What should you actually do with this? Nothing yet. It is a 20-page academic preprint, not a product, and it has not been peer-reviewed. But the ideas behind it tend to seep into real recommendation engines within a few years. If matching gets cheaper and faster, expect smarter shopping assistants, tighter job suggestions, and — eventually — apps that know what you want slightly better than you do.
- A new matching method compares people and products by traits instead of full ranked lists, making it far faster to compute.
- It scores well on a fairness measure called Nash Social Welfare and was tested on an AI shopping assistant.
- The catch: it is not strategy-proof — the paper shows how someone can lie about their preferences and get a better outcome.
Why It Matters
Better matching means smarter shopping, job, and dating apps — but also new ways for people to game the system.