Mirroring the Theory of Genetics and Evolution in Biology in a Way that Perspectives may be Revealed to Further Develop Agents Technology
Herbert A.C. ten Thij
P.O. Box 273 – NL  5670 AG
NUENEN, The Netherlands

Abstract- There is a wide variety in the understanding of genetics and evolution. This variety can be split into two main categories, showing a clear difference in the nature of the functioning of genetics and evolution as held by the views in these categories.

    A valuation of this difference is not the issue here, but a clarification of it could help to generate new ideas, especially in the domain of Evolutionary Intelligent Agents.

   A representative view of genetics and evolution, as used by the interested computer scientists, is contrasted with a modal, theoretical biological view of genetics and evolution. Based on the differences between these views, starting points are analysed for possible new development perspectives in IA technology, particularly in its application in the fields of knowledge acquisition and cognition.

1 Introduction

There is a, sometimes unfortunately, wide variety in understanding of genetics and evolution. This variety can be split into two categories, thus uncovering a significant difference between the two.

    One category consists of the understanding of genetics and evolution in the 'Theory of Biology', being more or less the sum of all views on these matters held by most biologists and other scientists in closely-related disciplines.

    The other category is formed by what can best be called the ‘general, common understanding of genetics and evolution’, both by the rest of the learned community and by the educated part of society.

    The contrast between the two categories becomes clear when we focus on the understanding of the nature of the functioning of genetics and evolution. The resulting discrepancies resemble the usual differences between modal and factual thinking (and talking or writing).

    The further valuation of the discrepancy between these categories is not the issue here, as it is not of much interest in this context. An insight into it could help generate new ideas, however. Most people involved in Evolutionary Programming (EP), Genetic Algorithms (GA), and Evolution Strategies (ES) or Evolutionary Intelligent Agents (EIA) technology also seem very well aware of the restricted use of the terms borrowed from the biological context. Nevertheless, in using these terms this knowledge is often not clearly expressed or connotated.

   This ideally means that they are well aware that genetics and evolution form part of the biological theories that are not subject to any teleology. They should therefore keep in mind that genetics and evolution are not rationally goal-oriented, just as they are well aware that computational systems always are. EP, for instance, as a stochastic optimisation strategy placing emphasis on the behavioural linkage between parents and their offspring, is definitely goal-oriented. GA is also goal-oriented, although it differs from EP by seeking to emulate specific genetic operators. Also all Agents Technology is goal-oriented.

2 Popular Views on Genetics and Evolution

An explanation concerning evolutionary computing is given here to characterise the large category of the views on genetics and evolution that generally are held. The other category is represented in the comment from a modal point of view that follows. This will make the differences easy to understand and will bring the discrepancies between the two categories into focus.

   The chosen explanation is the one published on the EvoWeb-site. This explanation clearly represents most of the generally held notions about genetics and evolution:

What is evolutionary computing? It harnesses the power of natural selection to turn computers into automatic optimisation and design tools. The three mechanisms that drive evolution forward are reproduction, mutationand the Darwinian principle of survival of the fittest. These mechanisms enable lifeforms toadapt to aparticular environment over successive generations. Evolution generates solutions to environmental problems. These solutions bear witnessto its power as a universal optimiser. Like evolution in nature, evolutionary computing also breeds progressively better solutions to a wide variety of complex problems.”

    This description uses all the usual terms such as ‘natural selection’, ‘reproduction’, ‘mutation’, ‘survival of the fittest’, ‘adaptation’ and ‘environment’. These are all terms that are now part of everyday speech. For this reason, we also universally tend to believe that we fully understand them in the context of genetics and evolution. This may have been true originally, however, when we shift our attention to a modal biological context, this belief is false. A closer look at the statements in the explanation just cited will make this clear.

3 A Modal View of Genetics and Evolution

The given explanation states that:

The three mechanisms that drive evolution forward are reproduction, mutation and the Darwinian principle of survival of the fittest”.

The end to which these mechanisms function is indicated in the following statement:

Evolution generates solutions to environmental problems”.

In this statement, however, ‘evolution’ is narrowed to a quasi-rational process, as a result of a factual, a posteriori analysis. This metaphoric use of the term ‘evolution’ masks its proper meaning in a biological context.

   ‘Evolution’ – as understood in a biological context – is far from some rational ongoing process. Similarly, ‘environment’ is not a complexity of reality that gives rise to any ‘problem’ whatsoever to which ‘evolution’ could answer with a solution. In the end, even from a factual point of view, as a concept or theory, ‘evolution’ in biology functions only as an explanatory model of the multiform existence of the many lifeforms which we can discern, or of the multitude of organisms or categorised biological entities which we know.

  It is at a modal level that we can best explain how ‘evolution’ actually functions. At this level, it becomes more clear that ‘evolution’ is indeed situational correlated. This correlation does not primarily function with regard to the options of genetic expression or actualisation of the many lifeforms, but is intrinsically conditional.

   Without the dynamic nature of the realisation options, a process such as evolution would be unnecessary or perhaps even meaningless. In a static, unchangeable situation, there would not be a great need for successive generations with so many possibilities to realise, or for all the new possibilities of initial lifeforms which inherently accompany the production of each following generation. It could even be argued whether ‘life’ would have emerged at all in such a situation. Yet,any further reflection on this subject is beyond the scope ofthis paper.

   The dynamic nature of the real situation is inherent in a physical and chemical or, for short, ‘geological’ way in the situation itself. This dynamic nature is also increasingly influenced by all the organisms, just by their current existence and, even more so at the moment, by all the ones that previously existed. In this context, dynamic means that any state of the situation at a given moment in time is only temporary, and will be followed by the next one, a change that might occur gradually or catastrophically. One thing is certain, as long as life exists, no next state, either by evolution or by involution, will ever be quite the same as a previous one.

   In fact, another, more advanced definition of ‘species’ could be induced from this data. This advanced definition of ‘species’ would then present the relationship to the changing impact which a certain group of lifeforms themselves have -just by their existence- on their immediate situation (thus also enhancing the 'niche' concept).

    In the known situation, an inheritance mechanism in lifeforms has evolved that could also sufficiently match so far the changing options. This complex inheritance mechanism could evolve itself, of course, like all features of life, at least on the basis of the already existent and also evolved phenomenon of ‘mutation’ or change that occurs in hereditary (genetic) material. This fundamental possibility of change alone is not effective enough and is even not always sufficient, however, as is increasingly the case in a process of developing (and maintaining) more complex lifeforms. It is only most effective in relation to another evolved possibility, which is so common and evident in daily life that it can easily be neglected. This extremely vital possibility is the evolved phenomenon of the involvement of two organisms, or a pair of relatively independent biological entities of the same kind, in the production of a next generation.  In this way, the possibility of linking the hereditary (genetic) material into new combinations comes to the fore. In this context, it is essential for the understanding to note that new phenotypic features of life -most certainly in relation to matching the changing situational options - are not only produced by a change of the hereditary (genetic) material, but also by combining this material. 

    Fundamental to all of this is, of course, that a certain temporal ‘life-span’ has already proven itself as an evolutionary result, more than the possibility of an ever-lasting occurrence, for instance.

4 Combination and Genetic Combination

The possibility of ‘combination’ (and ‘recombination’) that appears by pairing as such -firstly- and then, of course, by the pairing of the genetic material -secondly-, is crucial in the inheritance mechanism. It should be evident that in the case of a single linear reproduction, which only has the offer of new existence possibilities in a next generation, differing only from the previous one by mutation, the continuity of the  lifeform concerned has a very narrow, limited horizon in an unpredictable, dynamic situation. Only a rapid multiplication of the involved organisms in a quick succession of generations could help for a time, but even then, it is clear that the continuity of such a lifeform will inevitably cease at some point, if nothing else happens. 

  It has already been known for centuries in traditional, syllogistic logic that only two valid premises can lead to a valid conclusion, and that the least necessary sufficient condition for combining is having two entities that can do this. This remark might seem somewhat trivial nowadays, but it took the learned community more than thirty five years to rightly appreciate Mendel's achievements.

   It is to Mendel’s great merit that he has proven that this way of combining is exactly the case in genetics, a process with an important implication. By combining, so to speak, 'half' the genetic material of one parent with the same quantity of the other parent, a next generation will be born with a new (mixed) variety of possibilities (that can already display new features in this generation.) This offspring with its inherited variety of possibilities can, in turn, produce together a following generation with a variety of genetic possibilities of which some combinations (genotypes) will be completely different to those of the previous generations. In this way, a secure method has been found to install, even in this specific case, completely new combinations of genetic possibilities in just two ‘steps’, by simply combining the genetic material.

   Mendel deduced the quantitative distribution of genetic factors in successive generations, reckoning with the recessive and dominance range of influences exerted by the combined genetic material. It is in this context here more interesting, however, that the number of completely new possible genetic combinations within the complete possible variety can also immediately be deduced from what he learnt.

  In poly-hybrid crosses, which are commonly the case, this number (of completely new possible homozygous genotypes in each second generation) is generally represented very simply by:

    (2n - 2)

    for  n2

whereas the total number of possible genetic combinations per new second generation is of course:

    (2n)2  for  n2

and n is the number of the involved genetic characteristics (or factors, or commonly referred to as ‘genes’ these days).

   (The representation of the number of all new possible heterozygous genotypes in the F2-generation of poly-hybrid crosses is then:

    {(2n  ¾   2)  x  2n}    for   n ³)

In addition to this, recombination (in the sense of crossover) can also take place, by which even more new possible combinations can occur.

    'Reproduction' is too often considered to be generally repeating or multiplying more or less the same possibilities. We have now seen, roughly but concisely and clearly, that already an exponential increase of new possible varieties can occur after only two generations. In real life, this seems to be mostly sufficient -apparently- to generally match the given situational options during that time, whether these are altering, these have altered or these have not altered already so much compared to those in the period of the ancestral generation.

5 The Survival of the Luckiest

By now, it is also clear that notions such as 'adaptation' and 'survival of the fittest' are somewhat obscure, or are rather more appropriate to an ethological or (bio-) sociological context. These notions presuppose a range of activities of organisms which they do not employ at all in the modal context of genetics (and within that perspective in evolution), notwithstanding how much they may influence a process such as 'natural selection' by their factual behaviour. That process should at most be valued -also because it is evolutionary produced- as a modifying input in the second instance on the development or non-development of certain possible combinations in the genetic variety, given that sufficient organisms are available for a real selective entry into partnership.

   It should be kept in mind that there is a major difference between the evolution theory, as understood in a biological context, and the evolution models and further strategies and techniques that are used in evolutionary and genetic computing.

  The biological evolution might only seem to have rendered the optimal possibilities for successfully actualising a certain lifeform in the circumstances at that time.

   The evolution models used in computing do function differently, as they lead to some final end, a result that either ideally or actually is the optimal solution to a given problem.

   Properly valuing what will or could be the optimal possibilities of a certain lifeform is difficult, of course. Such a valuation remains limited -and is thus always open to discussion, meaningless or not- as it is also dependant on sufficient knowledge of the future development of that specific lifeform, which is unpredictable. Valuing these possibilities therefore remains speculative, which is usually also the case when considering the development and extinction of lifeforms in the past. The set of wild theories that have been put forward to explain the ‘sudden’ extinction of the dinosaurs is a concrete example of this, for which even extraterrestrial help has been introduced these days.

   The overall nature of biological evolution – also as an ever-continuing process – is not a development towards an optimal end, although at the same time also a comparable narrowing will take place in certain lifeforms, for some specific characteristics. It is a development towards a horizon of diversity, an open and dynamic 'end', or rather a development towards a future that is difficult to determine. Indeed, the notion or paradigm ‘survival of the fittest’, so necessarily connected with the idea that the present world is perfect or is the best that it could possibly be (with a somewhat Leibnizian reference), could, from a modal point of view, better be called, ‘the survival of the luckiest’.

  Actually, the whole idea of the ‘survival of the fittest’ is rather meaningless from a modal or an a priori point of view of genetics and evolution. Yet, even factually, this paradigm verbalised as ‘survival of the fittest’ adds responsibilities to lifeforms that go beyond their capabilities in the perspective of genetics and evolution. In these words, this principle falsely adheres an ethical dimension to lifeforms. When verbalised in this way, the paradigm also places ‘evolution’ incorrectly in a (pseudo) ethical sphere. The individual lifeform is so easily considered as an end in itself, which is irrelevant, both genetically and in an evolutionary context. For a better or clearer understanding of ‘evolution’, it would therefore be better to at least rename this paradigm ‘the survival of the luckiest’.  

6 The Fittest Function

  The principle of ‘survival of the fittest’ is then also of more use as a functional evaluation concept in a goal- oriented context. This is done in the practice of GAs, for example, where the fitness function even represents the whole (problem) environment, notwithstanding all difficulties that are met in constructing this function appropriately.

Perhaps that apart from the technical problems in producing a fitness function as optimal as possible also the selected conceptual assumptions in Gas (and EAs) raise part of the encountered difficulties. Some conceptual reconsideration might lead to possible new ways to construct GAs (and EAs).

   Typical in GAs is the use of a population of a (finite) number of strings. These strings act all, regardless their composition as known nowadays, as singularities that represent a single solution to the given problem. The strings can be added to each other in various ways to form a new singularity and so possibly a new but still single solution. This makes also immediately understandable why 'mutation' has acquired in GAs a prominent role in time.

This practice differs substantially, as we have seen already, from the real life situation where in the development and the functioning of a lifeform a duality generally executes a 'solution'. A duality that consists of (two) inheritable 'elements' and each 'element' can form a new combination.

  A reconsideration of GAs and EAs in a duality context in stead of the present singularity concepts might perhaps result in a new and more effective start in which also other genetic features (as dominance, incomplete dominance, co-dominance, etc.) can play a useful role.

   The use of selection mechanisms in GAs also makes clear that we do not really deal with 'natural selection', but entirely with artificial selection. Characteristic of artificial selection is traditionally the attempt to select for mating from the phenotypes at hand the probably best suited towards an ideal phenotype or to select the phenotypes with the best liked properties according to a breeder's criteria at the time. Also the replacement policies in GAs in order to keep the size of the population constant reminds us to the long tradition of the breeding practices. Practices that resulted, for example, in the variety and quantity of all the food available these days.

   The fitness function reveals that in GAs at the operational level the difference between genotype and phenotype is hardly relevant. The selection mechanisms in GAs perform their duties on base of the evaluation of the strings by the applied fitness function, also called 'the environment'. Clearly, this 'environment' could have than, from a modal point of view, controlling advantages. Perhaps this awareness could lead to considerations of the usefulness of constructing fitness functions or the selection mechanisms in a way (but still in a heuristic context, of course) that the course of the possible development structure of the strings towards an advanced next generation (population) also is valued or even is directed.

7 Interactive Instruction Systems

Evolutionary Intelligent Agents (EIAs) have come forward in the development of neural networks and agents technology. The application of EIAs in the teaching and learning environment, in particular, seems promising for the improvement of Interactive Instruction Systems (IISs).

  Probably the most interesting, and definitely the most convenient, in rendering such a system in a general way, is the development of interactive instruction systems that are independent of any specific discipline to be taught. Of course, a requirement of such a system (IIS), besides those for any software system such as robustness, maintainability and economy, is that it is generative in the sense that it can be applied to particular disciplines.

  The agents in an IIS can commonly play different and, when needed, parallel roles such as topic search and presentation, or giving explanation, feedback and exercises, while remaining adaptive to a learner’s abilities and progress.

Again, this adaptability, being here more or less synonymous to ‘learning’, is different from the a posteriori adaptation notion as used in the popular biological theory of evolution, and it denominates the active functioning of the agents.

It also means that by keeping track of the learning progress of the user in the instruction process, they do not only show pre-programmed behaviour, but they have the ability to learn how to behave adaptively and, by doing so, they can improve their behaviour in relation to a particular user. It is clear that EIAs could be functional in this, in a distinctive way, if they could provide the relevant real-time behaviour, and if their application made it possible to develop them at acceptable expenses. With respect to the options of providing an optimal functioning by a duality -surviving in a dynamic situation-, these could be further applied in IISs. A start could be made in the mixed-locus aspect of control, either by the user, the agents or by both, in the choices of the topics and depending on the ability of a user to monitor the learning process himself. It will be even harder to introduce the use of the combination options on the ‘genotype’ level, but it should be considered to see whether this could further advance EIAs.

8 Conclusions

   Genetics and evolution form part of the biological theories that are not subject to any teleology. It should therefore be kept in mind that genetics and evolution are not rationally goal-oriented. Computational systems definitely are goal-oriented. Nevertheless, an improved awareness of the nature of the functioning of genetics and evolution will lead to an advanced use of these theories in Evolutionary Computing or in Evolutionary Intelligent Agents technology.

   To witness to this awareness some of the used notions should be renamed or replaced. So, for example, the 'survival of the fittest' paradigm should be replaced with 'survival of the luckiest'.

   The advantage will be a proper thinking to deal better with the problems in Evolutionary Computing and conceiving more appropriate ways to solve them. Also in this way a secure road is opened to advanced designs of Evolutionary Intelligent Agents in the future.  



Mendel, G.J. (1865) "Versüche über Pflanzen - Hybriden", in Verh. Naturf. Ges. in Brünn, 4.

Ten Thij, H.A.C. (1999) "Mendel's Lessons", Unpublished Paper.

Ten Thij, H.A.C. (1999) "Let's get more bio-logical", Contribution to the 'Beagle' discussion on 'The Origin of Life'.

Ten Thij, H.A.C. (1996) "Urbanisation. An Essayistic Sketch Concerning Human Formalization", Gayles Stone Man Press, U.K.