Being Fair When You Decide Instantly Costs Almost Nothing
New math shows fairness costs only about four percent of total value.
Every day, computer systems hand out things that can't be split: a slot in search results, a concert ticket, a delivery job, a loan approval, a table at a restaurant. People show up one at a time, and the system must decide immediately, with no idea who comes next. Mathematicians call this a "prophet inequality" problem — your real-time choices measured against a fortune-teller who already saw everyone.
For decades the goal was simple: grab the most total value. Classic results say you can reliably capture about 74.5% of the fortune-teller's haul. But that approach ignores who gets what — in theory, one lucky person could take everything. This new paper asks what happens if you aim for fairness instead, judging outcomes by how evenly benefits are spread: the smallest share anyone receives, or the geometric average across everyone.
The answer is reassuring. The authors prove that even the strictest fairness goal — making sure the worst-off person does as well as possible — still captures at least 70.6% of what the fortune-teller would get. Compare that with 74.5% for pure value-grabbing, and fairness costs only about four percentage points. They also show the best fair rule is elegant: it gives every person the same expected payoff, rather than favoring whoever looks most valuable in the moment.
There is a catch. Their proof assumes many more items than people. When items and people are equal in number, the fairness guarantee collapses toward zero. In plain terms: fairness is nearly free when you have plenty to hand out, but genuinely expensive when supply is scarce and everyone wants the same thing.
- Researchers studied one-at-a-time decisions, like handing out tickets or job offers when you can't see who's coming next.
- Being fair costs about 4% of total value: 70.6% versus 74.5% of what an all-knowing planner would get.
- The guarantee only holds when there are many more items than people — scarce supply makes fairness expensive.
Why It Matters
If you've ever been picked or passed over by an algorithm, this says fairer rules barely cost anything.