The rules
Two ways to choose what to work on
A follower looks at the approaches near theirs that somebody has already tried, and moves to
whichever paid best. A maverick refuses to work anywhere anybody has worked before, and goes
somewhere nobody has been.
Weisberg and Muldoon (2009) found that a community of followers explores very little, and that adding a
modest share of mavericks transforms how much of the field gets covered. That reads as a formal argument
for funding people who go their own way, and it was received that way.
Alexander, Himmelreich and Thompson (2015) replied that the comparison was rigged by accident: the
follower rule performs worse than moving at random, so almost anything looks good beside it. The third
strategy on this page is wanderers, who simply move somewhere adjacent for no reason, and they are
the whole argument.
The argument
Why anyone cares
This is the page with the most direct practical claim on the site. If it is right, funding bodies should
deliberately support work that nobody else is doing, and the argument for that is formal rather than
sentimental.
Why a division of labour is not obvious
Kitcher (1990) posed the problem: a community where every researcher independently pursues the approach
most likely to pay will pile onto the same one, and that is collectively worse than a spread even though
each individual choice was correct. Individual and community rationality come apart. What is wanted is
some mechanism that produces a spread without anybody having to intend it.
Weisberg and Muldoon's contribution was to model the landscape explicitly and let the spread come from a
difference in behaviour rather than from a central planner. Their mavericks are not more insightful than
anyone else. They are simply constitutionally unwilling to work where others have worked, and that alone
is enough to cover the field.
The reply, which is about controls
Alexander, Himmelreich and Thompson noticed that the follower rule in the original model performs worse
than random movement. Once you notice that, the comparison stops establishing what it appeared to: the
mavericks were not beating a reasonable scientist, they were beating a badly specified one.
Most of what a maverick contributes here is not being a follower, and moving at random does that too.
The comparison above bears this out. A quarter mavericks roughly triples what a follower community covers,
and a community of wanderers achieves nearly the same. Whatever is doing the work, it is not the maverick's
distinctive refusal; it is any behaviour at all that stops researchers clumping.
What neither side quite says
Change what you measure and the ranking inverts. On a larger field, followers reach the single best
approach almost every time and wanderers mostly do not, because hill-climbing is what exploits a slope
once you are standing on one, and a maverick who never revisits anything can never climb.
So there is no strategy here that is simply better. Followers are good at finishing and bad at looking;
mavericks and wanderers are the reverse. A community wanting both needs both, which is a more modest
conclusion than the original paper and a more useful one than the reply. It also relocates the question:
not whether mavericks are valuable, but what mixture is, and that depends on facts about the field that
nobody inside it can see.
What this model is not
The landscape here is fixed, smooth, and has a single best answer that exists before anybody looks. Real
research changes what is worth doing as it proceeds, has no defined maximum, and the value of an approach
depends on what else has been established. Researchers here have no funding, no careers, no collaborators
and no ability to change strategy. Nobody has an idea.