For those of you living under rocks and too lazy to click on the links, authors who submit papers to FOCS this year are required to submit two pages describing their paper with the same level of detail as "a brief conversation or presentation". Umesh Vazirani posted his original proposal. His third paragraph lays out the main argument without any pesky technical details.
To understand the motivation for the abstract and better picture its contents, think about how often the 20 minute presentation at STOC/FOCS provides a better insight into the research than the paper.
Translation: STOC/FOCS authors can't write their way out of a wet paper bag with a flashlight, a map, and a Hanzo sword, but they can spin a good line when their backs are against the wall.
While preparing the talk, the authors can step back and try to explain something interesting about their work - either the core of their proof, or a special case of their theorem, or the new conceptual framework that they introduce.
Translation: Nobody really knows what they're doing until they have to give a talk about it.
The one week period after the mad rush to the STOC/FOCS deadline would provide a chance to reflect, and additionally there would be an incentive for the authors (just as in the conference presentations), to simplify.
Translation: STOC/FOCS program committees don't have the cojones to reject badly-written papers.
This is really the crux of the argument. How do we encourage STOC/FOCS authors to actually write well? The current system tells everyone what writing well means (in particular, give the reader a take-home message that entices them to read the technical details), and then given them an extra week to do it after all the technical lemmas are proved. Or to put in another way, ask for the technical stuff a week in advance of the human-readable message.
STOC and FOCS have a long-standing reputation for valuing technical depth at the expense of accessibility. To put it more bluntly, STOC and FOCS tradition encourages authors to deliberately obfuscate their results, lest they appear too simple. I understand that asking for brief summaries is an effort to reverse this trend, but I think it's exactly backwards. Asking for a separate human-readable summary, which will never be published, reinforces the idea that the actual submission doesn't need to be human-readable!
The point of writing conference papers—the point of writing anything—is to communicate ideas. If a paper doesn't communicate its underlying ideas effectively, it should be rejected, regardless of so-called technical merit. I do not mean that papers should only be accepted if they're easy to read. Many important results are deeply technical, and therefore necessarily require serious and sustained effort from the reader to understand. But the reader needs a reason to climb that mountain. The author should at least provide a few stunning photographs of the summit, if not a detailed trail map and a few experienced sherpas.
This places an extra burden on people in specialized subfields like, for example, oh, I don't know, let's say computational geometry. The STOC and FOCS program committees are unlikely to have many experts in Voronoi diagrams, relative ε-approximations, orthogonal graph drawing, pants decompositions, or cache-oblivious point-location; on the other hand, the program committees are (usually) good samples of the expertise in the STOC/FOCS community. If you can't convince three program committee members that they should care about your new result on probabilistically checkable fixed-genus quantum reset* embeddings, how do you expect to convince the conference attendees, or your later readers?
Oh, right, I forgot. Everyone gives great talks, and they write fantastically after they've let their brains cool down a bit. Uh-huh. Right. Go on, pull the other one.
I can think of one simple change to the FOCS submission process that would greatly improve the eventual outcome—require every brief description to include a pointer to the full technical paper (as a technical report, an ArXiv preprint, or just a PDF file on the author's web page), and then publish only the brief descriptions.
There is an important point in Umesh's last paragraph:
Old timers in the field might remember that program committee members in the big "theory schools" could expect to hear talks about a significant fraction of submitted papers well before the program committee meeting. With the much larger and geographically diverse theory community today, the two page abstract could restore some of these features.
Translation: This used to be a nice neighborhood. When we first moved here, doctors still made house calls, you could buy a hand-made phosphate down at the drug store for five cents, and people knew where they belonged. But now it's all gone to hell. Kids these days, with their inter-tubes and their tweets and their grogs... who understands 'em? Hey! Hey, you!! Get the hell off my lawn!!
*The dual of a coreset, of course!