COMPUTING SCIENCE
Machine Politics
Brian Hayes
Beyond Human Control
When computer-aided redistricting was first talked about 30 years
ago, it was supposed to end political gerrymandering. So far, it has
mainly had the opposite effect, providing a better tool for the
manipulation and coordination of political data. As computers and
software systems grow more powerful, gerrymandered districts will
doubtless become both more effective and more subtly hidden.
The algorithmic approach offers an alternative, but it would require
a major shift in attitude and expectations—a meta- political
revolution. No longer would redistricting be an opportunity to seize
political advantage; it would have to be seen as a neutral or
arbitrary event, beyond human control, above politics, subject to
luck, much like the random choice of which candidate's name is
listed first on a ballot. Redistricting would also be lost as an
instrument for achieving social goals, such as creating a more
racially balanced Congress.
If algorithmic redistricting is not to become another form of
high-tech gerrymandering, the mandated algorithm would have to be
spelled out in exacting detail, leaving no discretion to the
programmer or the operator of the computer. The specification of the
algorithm would also have to be openly published, presumably as part
of a statute, so that anyone could write a program to check on the
result of the official computation. And the specification would have
to be so explicit and detailed that every correct implementation of
the algorithm would give identical results on all legal inputs.
Legislators have little experience writing algorithmic
specifications. Even for experts, creating a correct and
ambiguity-free definition of a practical redistricting algorithm
would be a daunting challenge. Of necessity, the algorithm would
have to be a simple one. Elaborate weighing and balancing of
multiple criteria, or complicated hints and heuristics, would leave
too much room for mistakes and mischief. Also, the algorithm would
have to be deterministic, so that it would always yield the same
result on the same input data. And the required inputs themselves
would have to be simple and trusted, perhaps limited to what the
Census Bureau supplies.
After a few weeks of experimenting with redistricting algorithms, am
I prepared to turn the nation's political map over to computers? I'm
unsure. There is no doubt that people can draw much better districts
than any simple program can. Unfortunately, people can also draw
much worse districts. Thus the question of whether we would be
better off with people or machines drawing district lines depends on
one's assessment of the intentions and character of the people or
the machines who would do the drawing.
» Post Comment