Giving prizes to papers is actually more egalitarian. When you give them to people instead, there’s always a tendency to try to second-guess the results, for example, to overlook someone unexpected who has made a breakthrough on the grounds that they are a one-hit wonder due to luck. The fairest approach is to recognize great work whenever it occurs, and focusing on papers helps psychologically.

The theory community has plenty of other biases regarding giving prizes. (For example, conference best paper awards are much worse than test of time awards, since they promote fads, but TCS has mostly best paper awards.) However, giving a majority of awards to papers rather than people is a good thing.

Why not give prizes to people who made significant contributions? Often major progress is made over the course of several papers. I think that giving prizes to papers misses out on the really big achievements.

What’s an example of a prize-worthy discovery published in several papers, none of which could individually win a prize? In practice, what typically happens is that the last paper wins the prize; earlier papers may end up winning a test of time award as well (and of course some prizes can be given for several related papers). One could imagine a revolution carried out in so many small steps that no step deserved a prize, but I can’t think of any examples offhand.

The bias in favor of later papers may be unfair if the papers have different authors. If you make a breakthrough that is not fully recognized by the world and hence does not win a best paper award, and then I use it to prove great results and do win such an award, then I bet you won’t be happy. This is one reason to decouple awards from conferences and to allow them to be given for several related papers. However, it’s irrelevant to the issue of papers vs. people.

Isn’t it the case that most prizes in other scientific disciplines are not given to a single paper?

Maybe, but plenty of them are given for a single discovery. For example, that’s one crucial difference between the Nobel prize (which is always given for a specific, single discovery) and the Turing award (which is more of a lifetime achievement award in practice).

I’d be happy to switch from single-paper to single-discovery prizes in TCS, especially if it meant de-emphasizing conference best paper awards.

The only way to avoid this is to establish prizes that can be awarded as often as a suitable candidate appears. However, I’d strongly recommend against it, since then the battles over who should get the prize would become enormous. With current prizes, the stakes are lower since everyone knows there’s a lot of chance involved and thus not receiving the prize cannot be considered an insult.

Matteo Frigo, Charles E. Leiserson, Harald Prokop, Sridhar Ramachandran: Cache-Oblivious Algorithms. FOCS 1999: 285-298

- Leighton-Rao on separators
- Goemans-Williamson on Max-Cut SDP
- Bartal on tree embeddings (no journal version)