What They Did
The authors created “creatures” in a virtual
three-dimensional world and evolved them through 10 trials of 1500 generations
in a variety of experimental conditions. The virtual world consisted of a flat
area interrupted by two valleys, beyond which was a target. The creatures that
got closest to the target were included in the following generation, along with
a few more produced through virtual mutation of the “parents.”
The creatures all began as rectangular blocks and “grew”
additional blocks as limbs according to their “genotypes.” In some experiments,
all limbs were grown at the beginning, while in others, growth took place in
several developmental steps over the lifespan. Some experiments also included
behavior in the genotype, such that the creatures could place objects in the
valleys, making them easier to cross. Finally, some experiments kept a selected
percentage of added objects in the environment from one generation to the next.
The greatest evolutionary change occurred when the creatures
had developmental stages and were able to place objects. In that case, the most
successful creatures were those who grew a new limb close enough to the first
valley to be launched across it by the force of the growing limb against the
ground and those who had the behavior of placing objects in the second valley
for easier crossing. When added objects persisted in the environment, the
creatures were less successful in general because their object-adding behavior
did not immediately adapt to the objects already present, resulting in additional
objects forming obstacles.
Further Exploration
Simulations have been used for decades to demonstrate
concepts in evolution. I remember seeing simulations of allele frequencies in a
population back when I was a TA for freshman biology (see https://www.biologysimulations.com/population-genetics).
If you know the starting allele frequencies and relative probabilities of survival/reproduction
for different genotypes, it’s just a matter of doing a lot of math really fast,
which computers are great at.
The paper made me wonder if simulations are actually used
for hypothesis testing in evolution. It turns out they are! (see https://theconversation.com/simulating-evolution-how-close-do-computer-models-come-to-reality-57538
and https://www.nature.com/articles/srep08242.)
It’s impossible to model all the complexity of the real world, but researchers
can devise their simulations to ask how selection for a trait would look, all
other things being equal. In that sense, it’s a perfectly controlled experiment.
If other factors play an important role, the results of the experiment won’t
hold up in the long run, and the model will have to be refined.
I was surprised to find that the sort of virtual creatures
described in the paper were developed in the 1990s (see https://www.karlsims.com/evolved-virtual-creatures.html.)
The information I found seemed to focus on the use of virtual creature
evolution to make better robots or AI, not so much on the study of biological
evolution. I liked the way the paper explored the feedback loop in which inherited
behavioral tendencies cause organisms to alter their environment, leading to
new selection pressures that could change behavior. I wonder what kinds of hypotheses
could be tested with these systems, but that’s a rabbit hole for another day!
No comments:
Post a Comment
Please be nice.