Theory is not a product
Thesis: As an unintended consequence of otherwise desirable changes
in the math research enterprise over the last generation, theoretical
mathematical research is becoming viewed as if it were a product,
and this is not necessarily desirable.
(xxx This version is too unfocused!)
The need for speed?
We are all aware of shifts in the style of conducting research in mathematics and the mathematical sciences
over the last generation.
One external force has been increased emphasis on interdisciplinary research,
prompted by funding agencies and general societal pressure for "relevant"
rather than "ivory tower" research.
Most of the shifts have been enabled by new technology making collaboration and communication easier
and faster -- see e.g. Terence Tao's nice 2010 talk
The Future Impact of Internet-Based Technologies on Academia.
Though I often pose as a technophobe,
I am actually enthusiastic about exploring the possibilities opened up by technology,
and was a relatively early adopter of email, LaTeX, arXiv,
and even claim
(without much evidence except the archaic look) to have been about the 13,000th person with their own web page.
However, the discussions I have seen implicitly assume certain things
(speed, in particular) are
intrinsically desirable.
I will argue on this page that this assumption is based on a partly misleading analogy between
"theory" and "product".
What do I mean by those words?
Well, on a grand scale, developing a 5G phone network or a vaccine for HIV
is developing a product. Counterparts in
the history of the mathematical sciences, on a similar grand scale, might be
Game Theory or Relativity or even the basic setup of math probabability --
"intellectual ideas encapsulated as mathematics".
You and I may not work on such a grand scale, but on any scale we should be seeking either a product or an intellectual idea
(= "theory"), else what exactly are we doing?
Now I'm quite happy to agree that, for product development, speed is usually
desirable, whether for intrinsic human welfare or the demands of a competitive marketplace.
But is this self-evident for theory?
After all, the proposition
if something is worth doing, then it's worth doing faster
hardly stands scrutiny as a universal generalization;
otherwise no-one would spend 3 hours at a baseball game or a concert, or 90 minutes on a good dinner.
And eating food reminds us that the proposition
if doing something is worthwhile, then doing more of it is more worthwhile
is equally questionable.
To be as concrete as I can, let us stipulate that over the last generation,
largely as a consequence of technological changes,
the time to go from initially thinking about a math project to publication of a write-up
(nowadays typically via posting a paper on arXiv) has gone down, and that the
volume of mathematical research (number of published pages, per person per year)
has gone up.
Is this self-evidently a good thing,
as most discussants seem to implicitly assume?
Or an indicator of a kind of over-caffeinated culture, as this fridge magnet
(on the file cabinet behind my monitor, as I type) suggests?
Three spectra
To avoid getting bogged down in philosophical/aesthetic debate about the purpose of
theoretical math research, let me describe three spectra -- where between the two extremes
should a research project be?
The product development model vs the traditional mathematics model
Here's what I envisage as typical features of a product development project.
- A group of people (not just one or two) are involved.
- Each spends a substantial proportion of their time on the project.
- Different people often bring different skills.
- Weekly group meetings, partly to address big picture/strategy/new ideas ......
- ..... and partly small-scale management -- knowing and coordinating what
each person is doing.
- Considerable effort put into fine-tuning engineering details.
- There is an overall plan and timetable and budget (not necessarily adhered to!) for the project.
And most of these features are equally appropriate for a research lab in the experimental sciences.
Now I doubt that anyone explicitly proposes the above as a future model for research in theoretical mathematics.
And the decline of the traditional mathematics model -- one individual working alone, not communicating his
(usually, his) work until a
polished paper appears in a print journal 2 years after the key ideas were worked through
-- is deservedly as unregretted as any imperial decline.
But this all begs the question -- between these two extremes, what do we want as a replacement "standard model" for
conducting theoretical mathematics research?
Terence's Tao's article presents a rosy picture of possibilities, which I am all in favor of exploring,
but doesn't really suggest any answer more explicit than "technology-enabled collaboration".
Incremental vs transformative
The ESPRC (the U.K. analogue of NSF) asks proposal reviewers to comment on where a research proposal falls on the
"incremental to transformative" spectrum, explictly prefering the latter.
This raises several issues.
Even in an ideal world, could major breakthroughs happen in a vacuum rather than an atmosphere of more routine progress?
Can reviewers like you and I really identify proposals that will lead to transformative research?
But setting aside such issues and assuming we want institutional structures that promote "transformative research",
what model is best for theoretical math?
The ESPRC wants timetables for what is to be achieved in the first year, second year etc.
This is surely a core aspect of the "product development" model.
Does anybody think one can really plan a transformative idea?
Concentration vs dispersion
Amongst a given number of mathematicians working in a given field, would it be better to have more researchers
working within each of fewer topics,
or fewer researchers working within each of more topics?
There can be no right answer, but I surmise that changes over the last generation are pushing toward
the "concentration of effort" end of the spectrum.
Here's one example.
Traditional annual meetings covering a broad field
are now complemented by ad hoc meetings and research institute semesters on
``active topics".
Maybe the intention is "here's a new and active topic -- let's
publicize it and attract more people to work on it".
But in practice it takes a comparatively long time to organize a meeting,
while the speed-up of communication means that interesting
new techniques or problem topics are quickly seized upon by enterprising researchers.
So by the time the actual meeting is held, the effect may be to attract still
more people into an already-crowded
topic -- the opposite of the original intention!
Collaborative papers and LaTeX
At the level of individual papers there has been a clear
shift toward multiple-author papers.
Potential advantages are obvious:
- ``Two heads are better than one" for solving a given problem,
or devising variant problems, even if the two people have similar background knowledge ...
- ... and even more so if authors have different background knowledge ....
- ... in which case collaboration promotes diffusion of knowledge between
different fields, both via others reading the paper and via the
social networking aspect.
There seems no intrinsic downside to collaboration.
Obviously the arrival of
LaTeX and email made collaboration-at-a-distance much easier
(previously one had to mail manuscripts or typescripts back and forth).
Though as an aside, for those who view math as more a creative art than a science, it should be
puzzling that other creative arts
(e.g. writing novels) have not seen such increasing technology-enabled collaboration.
LaTeX has made the writing part of writing a paper much easier,
and surely it has changed the way one thinks about writing a paper.
Pre-LaTeX, the single author would
hand-write the mathematical arguments to convince himself he had obtained some
substantial result; then make a conscious decision to ``write a paper",
which involved starting over with blank paper and deciding how to
present the material to other people.
Post-LaTeX, it is tempting to write fragments as one thinks about a problem,
and have a document grow by accretion.
This has two negative consequences.
Post-LaTeX papers tend to be longer -- pre-LaTeX one had incentive to
choose only the main results to write up, but post-LaTeX it's
easier to include everything.
And the self-imposed threshold of "novelty and significance" for the content of
a paper tends to be lowered, because one is reluctant to
discard LaTeX-ed material, and because arXiv provides a venue
to post without any threshold constraint.
Here's the main point in this section.
I suspect that the combination of collaboration and LaTeX has
magnified the negative consequences above, mainly for social reasons.
For instance: pruning to the main results might delete one co-author's contributions;
different co-authors have different views on what the important material is;
one junior co-author might want to expand publication list.
All these effects pull toward more and longer papers.
Statistical aside.
One could get aggregate time-series data on quantities like average paper
length and number of authors, to quantify the changes we are discussing, but
it's hard to imagine serious statistical analysis of the causes and effects
of these changes,
since we're concerned with the "intellectual content" of different papers
at different times.
Conclusion?
What I have discussed comprises highly selective aspects of the whole
mathematical research enterprise,
but recall I seek to focus on "unintended consequences".
It would be interesting to do a "case study" comparing a few significant new
mathematical ideas originating around 1975 with a few others originating around
2005; how did each
topic develop over the subsequent 5 years?
Here's what I would expect one would find, for 2005 in comparison with 1975.
-
The "exploration process" -- discovering what (in retrospect)
are the main consequences of the basic idea and
investigating what (in retrospect) turn out to be sidelines
-- happens more quickly in time,
and involves more papers and authors.
There's nothing bad about this, but it's not clear to me that there's much
positive good either.
On the other hand, I expect one would also find
- increased speed has been accompanied by a lowering of the threshold
of novelty/significance of individual papers
and the latter seems to me both intrinsically undesirable and
having undesirable consequences. For instance, early-career people
seeing peers publishing 7 papers a year may be discouraged from
spending a year on a single hard problem.
Similarly, the "increased concentration on fewer topics"
mentioned earlier (which contributes to the first point above)
is neither good nor bad in itself but may have some undesirable consequences,
for instance a polarizing effect on young people's careers.
A few in each topic may stand out as having made substantial contributions while the others are
perceived to have made little marginal contribution; the latter would have achieved more
(in substance and in perception)
in a more dispersed research environment.
Ultimately one cannot avoid rather philosophical considerations involving the point of
theoretical math research, so let me conclude with two.
-
I am dubious whether any institutional arrangements can predictably increase the rate of introduction of
transformative ideas.
- Liking speed reflects a "waste of time" view of slowness.
A nine month wait between acceptance to a traditional print journal and availability to readers is indeed a
"waste of time" -- rapid electronic publication is great.
But this can morph into the notion that the time you spent on thinking about mathematics that didn't
result in a paragraph in a published paper was a similar "waste of time".
This notion is mistaken.
Some part -- not near 100% as in sports, nor near 0% as in product development, but some part --
of the point of math research is in the process itself, not the result.