This contribution analyzes parallels between software and biological evolution published in software engineering literature. Several papers suggest to the software engineering community that software and biological systems share some useful features of evolution. We compare the driving forces of the software evolution with biological evolution and highlight their different nature. We conclude that searching for parallels between biological and software evolution is a challenging task but at the same time there are limits, which may be hardly observed in the impure research field of software evolution.
Writing this paper was motivated by the author’s
reading papers in software engineering journals devoted to the
parallels between the software and biological evolution, e.g.
[
The aim of this contribution is to highlight observations concerning the role of biological evolution mechanisms within the context of computer science and software engineering. In particular, it analyzes the abstractions and driving forces of the biological and software evolution. In the conclusion we summarize the possible ways of misunderstandings. 2.
The notion of software evolution, which is closely related
and often interchanged with the term maintenance was
already introduced in the middle of the seventies when Lehman
and Belady examined the growth and the evolution of a
number of large software systems [
The term software evolution has emerged in many research
papers with roots both in computer science and software
engineering disciplines [
Lehman’s second law states the following: An evolving system increases its complexity unless work is done to reduce it. Due to the consequences of this law and due to the increased computing power, research in software and related areas accelerates and very often causes confusion and inconsistency in the used terminology.
Research on software evolution is discussed in many software related disciplines. Topics of software evolution are subjects of many conferences and workshops, too. In the following paragraph we will briefly characterize the interpretation of the notion of evolution in the short history of software engineering.
The notions of program synthesis or automated program
construction are the first forerunners of the evolution
abstraction in software engineering. Papers devoted to these
topics can be found also in the research field of automated
program synthesis. Practical results achieved in the field
of programming by examples are summarized, for example
in book [
The term evolution was originally a synonym for the
automation of program construction and for the discovery of
the reusable code – i.e. searching for loops. Later on, when
programming technologies matured and program libraries
and components were established in practice, the research
field of pattern reuse [
Since it was a necessity to maintain large and more
complex legacy systems, the topic of program comprehension
came into focus and became more and more important.
Program comprehension is an activity drafted, for example, in
the paper of Rajlich and Wilde [
This is in compliance with the idea that our understanding
of the domain problem incrementally evolves and learning is
an indispensable part of program comprehension. Rajlich in
dealing with the changing paradigm of software engineering
stresses the importance of the concept location. He argues
that the volatility of the requirements is the result of the
developer’s learning. Thus learning and understanding or
comprehension is indispensably coupled with the concept of
evolution [
These ideas are the forerunners of domain engineering as
described by Bjorner in his work [
The scattered results from the above mentioned areas lead
to the attempt to establish the taxonomy of software
evolution [
It is also obvious that the evolution principle in the
biological interpretation heavily attracted and influenced the
research of the evolution in software engineering. The work
of Nehaniv et al. critically warns the software engineering
community about the non-obvious traps when the evolution
principles valid in biology are mechanically applied to the
area of software artifacts [
These short notes highlighted the interlacing of software related disciplines and how they mutually influence each other.
Our position is that abstractions and mechanisms of biological evolution have limited impact on our understanding of software evolution. It is a challenge to question and clear the abstractions derived from the parallels in the impure field of software evolution. Questioning parallels between the biological and software systems reveals the limits of their applicability.
Let us cite from the work of Svetinovic and Godfrey [
In fact this approach neglects at all the importance of the change mechanisms. That is why we cannot discuss about the differences or similarities between them.
The next sentence says:
variation, and how they get incorporated into the population is not of great importance. [...]
If we want to analyze the parallels between the biological and software evolution then it is of great importance to characterize them. First of all, we have to distinguish between the processes that model the domain and processes that select the appropriate model for the domain. The selection of the appropriate model (or software development) is a non-deterministic process and that is why it cannot be fully automated, too.
Let us put the question: How does genetic algorithms relate to software evolution?
If we accept the standpoint that software models processes of a certain domain (natural or artificial) by algorithms (not only) then those algorithms may be inspired (among others) also by natural evolution (see, for example, the Wikipedia definition of the genetic algorithm). The answer is clear, genetic algorithms belong to the world of algorithms (the abstractions applied to a certain domain may be valid or not).
If we accept the standpoint that changes ”are introduced by intentional human actions” (as stated by the authors in the same paper) then we also have to admit that the changes are non-deterministic (it is also possible to select other algorithms than genetic for modeling a particular problem).
The conclusion is that we have to be aware and not to confuse non-deterministic human activities with the deterministic processes of the natural evolution.
The following sentence of the authors says:
same as long as the whole process is the same.
Accepting the fact that the processes are not the same intrinsically (software development is non-deterministic process – based on creative thinking and understanding) then we have to avoid temptation (very comfortable) to write about ”some common principles, mechanisms between the software and biological evolution”.
Otherwise we come to the conclusion that software development is a deterministic process (some natural process), which is contradiction with our initial standpoint (it is nondeterministic).
To sum up, we have to differentiate strictly between the processes that we model and the thinking-understanding (this was not yet modeled). These processes belong to two ”distinct worlds”. If we superficially ”merge” these two worlds (intentionally or unintentionally) then we deliberately give up seeking the truth and may construct false theories.
The same work states also:
If one wants to make the parallel between biological and software evolution, one could argue about the similarity of mutation to the incorporation of new, relatively origin features into the product. Recombination could be treated similar to the product combination and gene flow could be compared to the improvements due to the flow of knowledge similar.
We note that it is easy to find common abstractions on the observational level and, for example, to draw an analogy between the biological mutation and the incorporation of new features into the product. At the same time we have to declare that here we arrived to our boundaries. We have to fix that gene flow is deterministic but flow of knowledge is unknown and non-deterministic.
In other words; software is modified by actions of human
beings, while biological evolution is not the result of
actions by human beings. Biological evolution is forced by
”some natural” forces. Processes that drive the organism’s
genotypes and phenotypes [
This idea is expressed from another aspect by Jazayeri
[
[...] Not the software itself evolves, but our understanding and the comprehension of the reality. [...]
This approach stresses the importance of program
comprehension and the tools that support comprehension (see
also the work of Jackson [
In summary, software evolution evolves according to our understanding of the domain, which we want to model by the computer. From this point of view knowledge about biological evolution does not support our understanding of the domain under consideration (or universe of discourse).
Considering a software’s life cycle phases, we may observe (and have to observe) in all steps (as domain engineering, requirements engineering, specification, design, implementation, test, and operation) learning activities, which in fact means the evolution of our understanding of the modeled phenomenon.
Much more effective, than to seek parallels between
biological and software evolution is to follow the ideas in works
of Bjorner [
[...] Software cannot be designed without a robust understanding of the requirements and the requirements cannot be prescribed without a robust understanding of the domain. [...]
As far as we know, the only way of achieving a robust understanding of the domain is attentive work in the domain in order to gain experiences by trial and error. This is also called inductive inference and the product is our intangible understanding of the idea to be modelled. Following this thread of thought; the result of inductive inference is a deductive system (software or model), which has to be mapped on the computer.
In fact, it is not possible to build a software system without thorough understanding of the domain (which we do anyhow through all phases of software development life cycle). The more deeply we understand the reality the more the software could be evolved.
We must note that the Software Engineering Method and
Theory initiative [
Summing up, the existing disciplines of software engineering approach the phenomenon of evolution from various aspects. These approaches have their own history and theoretical roots; they are in various branches of computer/computing science and are treated from the philosophical point of view, too. It can be guaranteed that new techniques and research areas will emerge in the near future and will further deal with the phenomenon of evolution.
We hope that the above-discussed ideas will focus the attention of software engineers toward the reconsideration of the obviously claimed statements, which are against the facts observed in reality. In other words, to provoke a discussion about the feasibility of biological and software evolution parallels in software development. 5.
This work was supported by the project KEGA 040TUKE4/2011: Modern Software Engineering in Education – Proposal of the Structure and Realization of Actual Software Engineering Subjects for Informatics Study Program at Technical Universities. 6.