+Moshe Vardi points out that in the computer science conference selection process, after picking off the top 10% (or so) of submissions as easy accepts and the bottom 25% as easy rejects, the rest of the selections are largely random and arbitrary. But the rest of the selections are also where most of the work happens. Why are we putting so much effort into a part of the decision process where that effort is so unrewarded by improved results?

As Moshe argues, accepting that the remaining submissions are not differentiable into good and bad, and just accepting them all, would allow us to accommodate the growth in our field by significantly increasing the fraction of papers accepted per conference, and growing the size of the conferences. Instead, we've been keeping the number of accepted papers per conference more or less constant and greatly increasing the number of different conferences.

To put it more bluntly: Computer scientists are supposed to be the experts on how to make computational processes scale. Why are we not applying this expertise to our conference publication process?