Tuesday, October 05, 2004

Although I disagreed with Lance and Jeff on whether theory folks are too nice to ask hard questions, let us assume for argument's sake that there is some merit to this argument (there definitely is some just in comparison with other non-CS fields). What might be the cause of this restraint ?

A distinct possibility is the fairly objective nature of our discipline itself. Suppose that tomorrow I present a paper describing an algorithm that can do range searching on collections of fetid schizoid tapeworms in logarithmic time. Now, what is clearly true is that my algorithm takes logarithmic time; about this there can be no objection. However, it is likely that some folks will find the area of fetid schizoid tapeworms somewhat uninteresting; they might be partial towards energetic mobile ringworms instead. There might still be others that claim that most schizoid tapeworms are merely malodorous, for which logarithmic time range searching queries are trivial.

However, they cannot dispute the objective validity of my algorithm, and their dislike of my choice of object is not in itself objectively defensible.

This, to me, is one possible reason why theoreticians don't argue so much. It is not that we don't have strong opinions; if you talk to people one-on-one they will often criticize results based on non-technical reasons. It is that maybe people are less comfortable with non-formal reasoning processes.

Aha, but then you will say that "pure" mathematicians should not be the cantankerous bunch that they often can be. Indeed. However, theoretical computer science is most close in spirit to combinatorics, a field defined not by a foundational structure of axioms and theorems, but by a myriad of problems all connected together in countless different ways. It is hard to make objective judgements about the value of problems; most problems are quite interesting to think about in the abstract, at least to someone. It is a lot easier to evaluate the quality of a theoretical edifice, and in fact one of the knocks on combinatorics is precisely the "apparent" lack of such an edifice.


  1. I agree with Lance and Jeff that theory CS folks
    are 99% non-confrontational and it is quite noticeable
    in conferences and seminars.

    People often have reservations about a line of work
    but many times unexpected connections are found
    that are not apparent in the beginning. I used to
    wonder why metrical task systems were so interesting
    but then Bartal's important work
    on tree embeddings was directly motivated by them and
    this work has had a big influence on approximation algorithms and finite metric space theory. Same with
    property testing - I was initially skeptical but then
    the math turned out to be quite interesting and useful
    in other areas. These experiences make me cautious
    in making judgements about the usefulness of a line of
    work - better to wait and see as long as the
    technical part is interesting and innovative.


  2. Suresh wrote "However, they cannot dispute the objective validity of my algorithm, and their dislike of my choice of object is not in itself objectively defensible."

    True. Indeed, speaking of objects, I was troubled when I saw your paper advertised on Rajeev's site as "Object Accessibility for Java is Decidable" at STOC. I remember thinking, "What kind of nut writes about Java at a STOC conference?" But after I read it, I thought "that's nice."

  3. ...And one reason for not being agressive in talks in theory is because the audience recognize the fact that they probably not understanding whats going on. Consider for example the ARV result from the last STOC - a very nice result, but not something you can understand from the talk (which was nice, BTW).


Disqus for The Geomblog