Review of "The evolutionary origin of complex features"
Harsha V. Madhyastha

This paper aims to study the intricacies of how Darwin's theory of evolution
by natural selection works. This study is carried out in a test environment
wherein each organism is a self-replicating computer program. Each program
comprises of a sequence of instructions, each of which is one of 26 chosen
instructions. The replication is subject to errors so as to simulate
mutations. The "health" of each organism is evaluated on the basis of how many
logical operations it can perform, from those possible with the available 
instruction set. The merit derived from any logical operation is determined
by the minimum number of NAND gates required to implement that logical
operation. The authors then study the ancestral hierarchy of the organisms
produced and make an attempt to pinpoint the reasons for development of the
best possible organism, which in the given test setting is a program that can
perform an EQUALS operation.

The major strength of the paper is clearly the elaborate test environment that
the authors setup for this study. Being able to simulate the process of
replication and natural selection helps the authors to get a complete trace of
the sequence of generations produced. This is the key strength of this study
as being able to look at the exact sequence of organisms produced helps the
authors reason out the how and why the evolution process proceeded the way it
did. In the real-world, one can only make guesses about how evolution might
have proceeded due to the lack of complete information. The other main
strength of this paper is the depth to which the authors have explored the
reasons behind evolution of the "best" organism. The thorough study performed
addresses almost all questions that might arise in a reader's mind and leaves 
hardly anything to speculation.

The main bone of contention in this paper is its representativeness. Though
the environment used for testing is pretty elaborate, the authors had to make 
various choices such as the choice of instructions, the possible ways in which 
mutations can occur and the means of evaluating the health of any organism.
Whether these choices are in anyway representative of evolution as it occurs
in the real-world is anybody's guess. Though the conclusions drawn in this
paper appear possible explanations to evolution in the real-world, the
question remains as to whether there could more intricacies involved which
were lost in the test environment chosen. One question that I felt the authors
could have addressed more is study the genealogical sequence of organisms with
moderate health that lingered on for several generations. The authors
restricted much of their study to the evolution of the best possible organisms
and though they did a pretty thorough job of it, it would have been
interesting to see whether there are some kinds of features which linger on
for several generations even though they might not necessarily be the best and
look at the reasons behind such evolution.

The process of evolution can be modeled as a search problem but it would
extremely complex. Each state would be representative of a generation, which
is a set of organisms. However, unlike in a normal search, where one can
define either probabilistically or deterministically the successor(s) to each
state, in this case, the successor to any state also depends on the
interactions between the organisms in any state. The new state produced is
dependent on the competition between the organisms in a given state for the
available resources. This also cannot be accounted for by factoring the state
into a set of variables because the number of variables in each state is not
fixed. On the whole, it does not look like evolution can be modeled as a
search problem to which any normal search algorithm can be applied.