Monday, October 04, 2004

Too nice ?

Lance asks, and Ernie elaborates on, the following question:
Are theoretical computer scientists too nice ?
Lance uses evidence (lack of questioning at theory seminars) to infer niceness. Jeff goes further and argues that this is true well beyond theory. I'm not sure I agree. I think this issue goes to the heart of another discussion that Lance initiated: the balkanization of theory into subfields.

The assumption is that if a theoretician gives a talk that is attended by other theoreticians, then lively discussion should ensue. However, given the vast scope of theory, how realistic is it for an audience consisting of (say) geometers to be able to discuss details of a speaker's talk on (say) the latest randomized rounding trick for LP-based approximations ? (example chosen completely at random :)).

Obviously one can discuss things like the general motivation etc, but as some commenters pointed out, the motivation is often not the most interesting part of a work. And in theory talks, where one can measure fairly objecively the contribution of a particular work, discussions about contribution are either trivial, or require a much deeper understanding of the sub-area of the talk.

The reason I bring this up is because it is NOT my experience that theory talks are viewed as missives from the Gods, to be watched reverentially and applauded solemnly. At the Math Seminar talks at AT&T, we have heated discussions, and it is very hard for a speaker to get beyond the first ten slides without some serious pounding :). I think the 'message from God' model applies more to conference presentations, where
  1. the problem of sub-area matching comes up: there are probably only a few people who really understand the speaker's work, and they probably figure they can catch the person later on so as not to bore the audience (I have felt that way, and maybe I am wrong to feel so)
  2. Time is really limited: you can't get into any kind of discussion, and it takes a while to grok some of the more technical details.Although Lance uses the Econ Workshop as a model for aggressive questioning, that is really a seminar, and not a typical conference venue
To support this, I can provide as evidence our AT&T one-hour Cookie Talks, that are typically less theoretical and directed at a larger audience of systems, database, programming languages, algorithms folks, and so are less technical. In these talks there are lots of discussions, and these can often get acrimonious (we sometimes have to warn visiting speakers !).

No comments:

Post a Comment

Disqus for The Geomblog