When you come across an article about slime mould designing
rail networks your attention is grabbed. Especially if you are, very
tangentially, involved with designing railway networks. I work with a team that
creates computational algorithms for routing rail and road systems. I’m just
the translator, but the insight it gives me into the process is very
interesting. The idea is striking and absurd, and whatever slime mould is (it’s
this,
apparently) it has the sort of name that fits well in absurd situations.
You might think that all the slime mould has done is, once
it’s worked out the quickest way to get between the places where the food is,
and how far apart they are and how much food is in each place, it’s quite easy just
to go back over the short paths and everything just becomes a network of
straight lines. But in fact it doesn’t. You can clearly see that the lines are more
or less straight and regular, but it isn’t a question of joining every node to
every other. That would be highly inefficient. The network created uses some existing
nodes as waypoints to get to others, and creates new nodes when it is efficient
to do so. This is what network designers do also.
If a collection of unicellular organisms with no nervous
system, let alone sense organs or a brain, can do this, then, you might say, it
can’t be too difficult, can it? Isambard Kingdom Brunel did it for most of the
country with just a pencil and paper, and he seems to have done a pretty good
job, but then he had a fully functioning brain and a team of similar organisms
to work with, which always puts you ahead of the game.
The team I work with, and other teams that they work with,
are busy creating new algorithms to make network design better. The problems
are of unimaginable complexity, enormously greater than is generally realized.
Huge quantities of data to do with terrain characteristics, land value, current
demand, future demand (itself a function of the final design), and economic
impact, have to be obtained to an acceptable degree of accuracy, and then
analysed to find the most efficient solution. The really hard work, the clever
bit where the computer engineers earn their money, is in designing algorithms that
will enable you to find a good solution before the sun becomes a supernova and
renders the whole exercise academic. The computational cost of these algorithms
is truly astronomical, and getting it right, as opposed to nearly right, can
cut millions off the cost of a project, and add many more millions to the
economy of the area whose transportation it facilitates. This last factor is
probably impossible to quantify, but it is undoubtedly true. Brunel had to play
it by ear, and with hindsight, it could certainly have been done better.
I don’t see our road and rail networks being designed by
slime mould in the future, but many heuristic algorithms are modelled on
analogies with, or even direct observation of, natural processes. Experiments
could easily incorporate topographical and geological features, as I imagine
slime mould avoid steep hills and unsatisfactory environments the way most of
us do, and a lot could be learned from observing them. Even if it only leads to
the refinement of existing algorithms they will have done us a great service.
Next week: I am beaten at chess by a team of flagellate protozoa.
Artificial intelligence may not be what we think it will be.
No comments:
Post a Comment