Notes on the Origin of Evolutionary Computationsipper/publications/orgevol.pdf · larly characterised. This principle of preservation, I have called, for the sake of brevity, Natural

Notes on the Origin ofEvolutionary Computation

THE ORIGIN OF SPECIES

If during the long course of ages and under varying conditions of life, organicbeings vary at all in the several parts of their organisation, and I think thiscannot be disputed; if there be, owing to the high geometrical powers ofincrease of each species, at some age, season, or year, a severe struggle for life,and this certainly cannot be disputed; then, considering the infinite complex-ity of the relations of all organic beings to each other and to their conditionsof existence, causing an infinite diversity in structure, constitution, and habits,to be advantageous to them, I think it would be a most extraordinary fact if novariation ever had occurred useful to each being’s own welfare, in the sameway as so many variations have occurred useful to man. But if variationsuseful to any organic being do occur, assuredly individuals thus characterisedwill have the best chance of being preserved in the struggle for life; and fromthe strong principle of inheritance they will tend to produce offspring simi-larly characterised. This principle of preservation, I have called, for the sake ofbrevity, Natural Selection.

Chapter 4: Natural Selection

I n 1859 Charles Darwin published his book On the Origin of Species by Means ofNatural Selection, or the Preservation of Favoured Races in the Struggle for Life.The shock waves produced by Darwin’s revolutionary theory of evolution1 are

still reverberating within society at large, though science has by now come to termswith it.

One of the prominent offshoots of Darwinian evolution occurred almost a cen-tury later, with the advent of evolutionary algorithms in the 1950s. The basic idea, bynow well entrenched within scientific and engineering lore, is both simple andingenious: implementing within a computer a simulacrum of evolution by naturalselection. The basic principle of evolutionary computation was enunciated byDarwin in his 1859 book: “. . . one general law, leading to the advancement of allorganic beings, namely, multiply, vary, let the strongest live and the weakest die.”(Chapter 7)

In the following article, I wish to go back to the origin of evolutionary compu-tation, namely, The Origin of Species, with the aim being to expose some ideas thatmay be relevant to the field today. My strategy is simple: To illustrate each idea, Ishall bring forth excerpts from the book, followed by a short annotation. This latteris intended neither to rephrase nor to explain—The Origin of Species is remarkablenot only for its conceptual depth but also for its clarity of style. Rather, the anno-

MOSHE SIPPER

Moshe Sipper is a Senior Researcher inthe Logic Systems Laboratory at theSwiss Federal Institute of Technology,Lausanne, Switzerland. He received aB.A. in Computer Science from theTechnion—Israel Institute ofTechnology, and an M.Sc. and a Ph.D.from Tel Aviv University. His chiefinterests involve the application ofbiological principles to artificialsystems, including evolutionarycomputation, cellular automata (withan emphasis on evolving cellularmachines), bio-inspired systems,evolving hardware, complex adaptivesystems, artificial life, and neuralnetworks. Dr. Sipper has authored andcoauthored over 60 scientificpublications in these areas, includingthe book Evolution of Parallel CellularMachines: The Cellular ProgrammingApproach (Heidelberg: Springer-Verlag,1997). He was Program Chairman ofthe Second International Conference onEvolvable Systems: From Biology toHardware (ICES98), held in Lausannein September 1998.

© 1999 John Wiley & Sons, Inc., Vol. 4, No. 5 C O M P L E X I T Y 15CCC 1076-2787/99/05015-07

Draft

tations are intended to recast the pre-ceding passage within the modernevolutionary computation framework.This article is not intended to serve as areview of work in the field of evolution-ary computation—several good articlesand books exist for this purpose. Myaim is to recount my own views aboutsome of the relationships between Dar-win’s Origin and modern evolutionarycomputation.

LARGE POPULATIONS

I must now say a few words onthe circumstances, favourable, orthe reverse, to man’s power of se-lection. A high degree of variabil-ity is obviously favourable, asfreely giving the materials for se-lection to work on; not that mereindividual differences are not am-ply sufficient, with extreme care,to allow of the accumulation of alarge amount of modification inalmost any desired direction. Butas variations manifestly useful orpleasing to man appear only oc-casionally, the chance of their ap-pearance will be much increasedby a large number of individualsbeing kept; and hence this comesto be of the highest importance tosuccess.

Chapter 1: Variation UnderDomestication

A large amount of inheritable anddiversified variability is favour-able, but I believe mere indi-vidual differences suffice for thework. A large number of individu-als, by giving a better chance forthe appearance within any givenperiod of profitable variations, willcompensate for a lesser amountof variability in each individual,and is, I believe, an extremely im-portant element of success.


In these two passages, the first takenfrom the chapter on artificial selection(husbandry) and the second from thechapter on natural selection, Darwin ar-gues in favor of large populations. Thismay well apply to evolutionary algo-

rithms, the majority of which use verysmall populations (that is, on the orderof a few hundred individuals). Thecontinued increase in computationalpower serves to enfeeble the lack-of-computational-resources argumentgiven to support the use of small popu-lations. Large populations may thus bea simple remedy to the incessant con-cern of practitioners in the field—namely, the maintaining of variability—ultimately leading to better “organisms”(solutions).

NEUTRAL MUTATIONS, GENETIC DRIFT,AND PREADAPTATION

I am inclinded to suspect that wesee in these polymorphic generavariations in points of structurewhich are of no service or disser-vice to the species, and whichconsequently have not beenseized on and rendered definiteby natural selection, as hereafterwill be explained.

Chapter 2: Variation UnderNature

Variations neither useful nor in-jurious would not be affected bynatural selection, and would beleft a fluctuating element, as per-haps we see in the species calledpolymorphic.


We are far too ignorant, in almostevery case, to be enabled to assertthat any part or organ is so unim-portant for the welfare of a spe-cies, that modifications in itsstructure could not have beenslowly accumulated by means ofnatural selection. But we mayconfidently believe that manymodifications, wholly due to thelaws of growth, and at first in no

way advantageous to a species,have been subsequently takenadvantage of by the still furthermodified descendants of this spe-cies. We may, also, believe that apart formerly of high importancehas often been retained (as thetail of an aquatic animal by itsterrestrial descendants), though ithas become of such small impor-tance that it could not, in its pres-ent state, have been acquired bynatural selection, a power whichacts solely by the preservationof profitable variations in thestruggle for life.Chapter 6: Difficulties on Theory

One can see in these three passagesreferences to phenomena that wereelaborated over a century later by biolo-gists, for example, neutral mutations,genetic drift, and preadaptation. Suchphenomena are obviously of great in-terest to evolutionary-computationpractitioners. (NB: Darwin considersboth inter- and intra-species interac-tions. Though evolutionary algorithmsare chiefly single-“species,” I believethis is not crucial for the account givenherein, that is, the basic ideas still ap-ply. In fact, this also raises the issue ofdesigning multi-species, artificial-evolution scenarios).

COEVOLUTION

The dependency of one organicbeing on another, as of a parasiteon its prey, lies generally betweenbeings remote in the scale of na-ture. This is often the case withthose which may strictly be saidto struggle with each other for ex-istence, as in the case of locustsand grass-feeding quadrupeds.But the struggle almost invariablywill be most severe between theindividuals of the same species,for they frequent the same dis-tricts, require the same food, andare exposed to the same dangers.Chapter 3: Struggle for Existence

As the individuals of the same spe-cies come in all respects into theclosest competition with each

The continued increase incomputational power serves to

enfeeble thelack-of-computational-resourcesargument given to support the

use of small populations.

16 C O M P L E X I T Y © 1999 John Wiley & Sons, Inc.

other, the struggle will generallybe most severe between them; itwill be almost equally severe be-tween the varieties of the samespecies, and next in severity be-tween the species of the same ge-nus. But the struggle will often bevery severe between beings mostremote in the scale of nature. Theslightest advantage in one being,at any age or during any season,over those with which it comesinto competition, or better adap-tation in however slight a degreeto the surrounding physical con-ditions, will turn the balance.

Chapter 14: Recapitulationand Conclusion

Darwin recounts here the coevolu-tionary nature of Nature—the “coadap-tations of organic beings to each other”(Introduction chapter)—a theme thatplays a pivotal role in Darwin’s theory,recurring throughout the book. Indeed,coevolution has been receiving in-creased attention in recent years withinthe evolutionary-computation commu-nity. These passages attest to the intri-cateness of this phenomenon—the de-gree to which inter- and intra-speciescompetitions take place depends incomplex ways on the amount of simi-larity (or in some cases dissimilarity)between the competitors. The coevolu-tionary aspect of nature was very suc-cinctly worded by Darwin in Chapter12, where he speaks of “organic actionand reaction. . .” (Interestingly, this re-sembles Newton’s third law of mo-tion—for every action, there is an equaland opposite reaction—a physical issueto which Darwin alludes in Chapter 3:“Throw up a handful of feathers, and allmust fall to the ground according todefinite laws; but how simple is thisproblem compared to the action and re-action of the innumerable plants andanimals which have determined, in thecourse of centuries, the proportionalnumbers and kinds of trees now grow-ing on the old Indian ruins!”)

PUNCTUATED EQUILIBRIA?

Not that in nature the relationscan ever be as simple as this.

Battle within battle must ever berecurring with varying success;and yet in the long run the forcesare so nicely balanced, that theface of nature remains uniformfor long periods of time, thoughassuredly the merest trifle wouldoften give the victory to one or-ganic being over another.Chapter 3: Struggle for Existence

We see nothing of these slowchanges in progress, until thehand of time has marked the longlapses of ages, and then so imper-fect is our view into long pastgeological ages, that we only seethat the forms of life are now dif-ferent from what they formerlywere.


. . . I do believe that natural selec-

tion will always act very slowly,

often only at long intervals of

time, and generally on only a very

few of the inhabitants of the same

region at the same time. I further

believe, that this very slow, inter-

mittent action of natural selec-

tion accords perfectly well with

what geology tells us of the rate

and manner at which the inhab-

itants of this world have changed.


One could possibly read “punctu-

ated equilibria” into these passages, a

controversial phenomenon evoked in

the 1970s by paleontologists Eldredge

and Gould, which has also been ob-

served in some artificial-evolution ex-

periments. Interestingly, in the last

passage Darwin clearly relates “inter-

mittent action” with the geological rec-

ord—the basis of Eldredge and Gould’sobservations.

IMPLICIT VERSUS EXPLICIT FITNESS

Man selects only for his owngood; Nature only for that of thebeing which she tends.


No country can be named inwhich all the native inhabitantsare now so perfectly adapted toeach other and to the physicalconditions under which they live,that none of them could anyhowbe improved. . .


Natural selection will not pro-duce absolute perfection, nor dowe always meet, as far as we canjudge, with this high standard un-der nature.Chapter 6: Difficulties on Theory

A major difference between artificialevolution, which is more akin to hus-bandry, and natural evolution is clearlystated above. This is largely related tothe use of explicit fitness in evolution-ary computation versus nature’s im-plicit fitness. The last two passagesshow that natural evolution does notoptimize, as opposed to artificial evolu-tion—which usually does (or at least hasoptimization as the underlying goal).One should note here, nonetheless, thefew experiments that have tried toimplement a more open-ended, im-plicit-fitness scenario (the most well-known example being Tom Ray’s Tierraworld).

ARGUMENT FOR CROSSOVER

. . .that these facts alone inclineme to believe that it is a generallaw of nature (utterly ignorantthough we be of the meaning ofthe law) that no organic beingself-fertilises itself for an eternityof generations; but that a crosswith another individual is occa-sionally—perhaps at very long in-tervals—indispensable.


The slightest advantage in one being,at any age or during any season,

over those with which it comes intocompetition, or better adaptation in

however slight a degree to thesurrounding physical conditions, will

turn the balance.

© 1999 John Wiley & Sons, Inc. C O M P L E X I T Y 17

Decades before the advent of mo-lecular biology, and based solely onphenotypical observations, Darwinconcluded that crossover is of high util-ity. Practitioners in the domains of ge-netic algorithms and genetic program-ming, in which the use of crossover isubiquitous, might take heart from itsprominence in Darwin’s text.

It should be noted that there are twodifferent forms of recombination(crossover) in nature, which seem to getconstantly conflated in the evolution-ary-computation literature: allelic re-combination, which occurs in the for-mation of a gamete within an indi-vidual, and sexual recombination,which occurs between individuals, whentwo gametes are fused into a zygote.By “crossing” Darwin was not refer-ring to the former, unknown in histime, but rather to the latter, involvingcrossing—or mating—between indi-viduals. (We note in passing that theevolutionary-computation literature of-ten speaks of the importance of sex,though the traditional one-point cross-over operator is more analogous to al-lelic recombination).

NO PURE ALTRUISM

What natural selection cannot do,is to modify the structure of onespecies, without giving it any ad-vantage, for the good of anotherspecies; and though statementsto this effect may be found inworks of natural history, I cannotfind one case which will bearinvestigation.


Natural selection will producenothing in one species for the ex-clusive good or injury of another;though it may well produce parts,organs, and excretions highlyuseful or even indispensable, orhighly injurious to another spe-cies, but in all cases at the sametime useful to the owner.Chapter 6: Difficulties on Theory

Simply put, pure altruism is not pos-sible. Fitness assignment must always

entail some advantage to the evolvingentity (be it the gene, the individual, orthe group), an important consideration,especially where more complex coevo-lutionary scenarios are concerned.

ON COMPARING ALGORITHMS

But we may thus greatly deceiveourselves, for to ascertain wheth-er a small isolated area, or a largeopen area like a continent, hasbeen most favourable for the pro-duction of new organic forms, weought to make the comparisonwithin equal times; and this weare incapable of doing.


Just a reminder of an important cri-terion when comparing different evolu-tionary methodologies—“comparisonwithin equal times”—which we are ca-pable of doing where evolutionary algo-rithms are concerned.

NATURAL SELECTION VERSUSSEXUAL SELECTION

And this leads me to say a fewwords on what I call Sexual Selec-tion. This depends, not on astruggle for existence, but on astruggle between the males forpossession of the females; the re-sult is not death to the unsuccess-ful competitor, but few or no off-spring. Sexual selection is, there-fore, less rigorous than naturalselection.


Evolutionary computation concernsby and large the artificial analog ofnatural selection, with almost no at-tention given to the second deter-minant factor: sexual selection. Thislatter would be quite interesting to

study in the context of evolutionarycomputation.

GEOGRAPHICAL CONSIDERATIONS

Isolation, also, is an important el-ement in the process of naturalselection. In a confined or iso-lated area, if not very large, theorganic and inorganic conditionsof life will generally be in a greatdegree uniform; so that naturalselection will tend to modify allthe individuals of a varying spe-cies throughout the area in thesame manner in relation to thesame conditions.


If, however, an isolated area bevery small, either from being sur-rounded by barriers, or from hav-ing very peculiar physical condi-tions, the total number of the in-dividuals supported on it willnecessarily be very small; andfewness of individuals will greatlyretard the production of newspecies through natural selec-tion, by decreasing the chance ofthe appearance of favourablevariations.


Geographical considerations perme-ate Darwin’s book; indeed, two entirechapters are devoted solely to this issue(Chapters 11 and 12). In the above pas-sages we note the importance of iso-lated niches, though not too small atthat.

In an extremely small area, espe-cially if freely open to immigra-tion, and where the contest be-tween individual and individualmust be severe, we always findgreat diversity in its inhabitants.


On the other hand, if the niche issmall, yet with a constant influx of newindividuals, then selective pressure ismaintained.

. . .that migration has played animportant part in the first appear-

In a confined or isolated area, ifnot very large, the organic and

inorganic conditions of lifewill generally be in a great

degree uniform.


ance of new forms in any onearea and formation. . .Chapter 10: On the Geological

Succession of Organic Beings

Widely-ranging species, abound-ing in individuals, which have al-ready triumphed over many com-petitors in their own widely-extended homes will have thebest chance of seizing on newplaces, when they spread intonew countries. In their newhomes they will be exposed tonew conditions, and will fre-quently undergo further modifi-cation and improvement; andthus they will become still furthervictorious, and will producegroups of modified descendants.

Chapter 11: GeographicalDistribution

Migration induces perennial compe-tition. The use of niches in evolutionaryalgorithms can thus be seen not only toexhibit computational advantages (interms of parallelism) but also more pro-found ones, deriving directly from Dar-win’s account.

Thus the high importance of bar-riers comes into play by checkingmigration. . .


If the difficulties be not insuper-able in admitting that in the longcourse of time the individuals ofthe same species, and likewise ofallied species, have proceededfrom some one source; then Ithink all the grand leading facts ofgeographical distribution are ex-plicable on the theory of migra-tion (generally of the more dom-inant forms of life), together withsubsequent modification and themultiplication of new forms. Wecan thus understand the high im-portance of barriers, whether ofland or water, which separate ourseveral zoological and botanicalprovinces.

Chapter 12: GeographicalDistribution (continued)

This serves to emphasize the intri-cate interplay between isolation andmigration—barriers should allow forsome separation, though not a total oneat that.

If, for instance, a number of spe-cies, which stand in direct com-petition with each other, migratein a body into a new and after-wards isolated country, they willbe little liable to modification; forneither migration nor isolation inthemselves can do anything.These principles come into playonly by bringing organisms intonew relations with each other,and in a lesser degree with thesurrounding physical conditions.


If a large body of individuals mi-grates from one niche to another, no fit-ness benefit will result since the statusquo is maintained. One must strive tobring organisms “into new relationswith each other. . .” In simple evolu-tionary-computation scenarios, theabove passage might translate into asmall number of migrating individuals.

T o sum up the circumstancesfavourable and unfavour-able to natural selection, as

far as the extreme intricacy of thesubject permits. I conclude, look-ing to the future, that for terres-trial productions a large conti-nental area, which will probablyundergo many oscillations oflevel, and which consequentlywill exist for long periods in abroken condition, will be themost favourable for the produc-tion of many new forms of life,likely to endure long and tospread widely. For the area willfirst have existed as a continent,and the inhabitants, at this periodnumerous in individuals andkinds, will have been subjected tovery severe competition. Whenconverted by subsidence intolarge separate islands, there willstill exist many individuals of the

same species on each island: in-tercrossing on the confines of therange of each species wi l lthus be checked: after physicalchanges of any kind, immigrationwill be prevented, so that newplaces in the polity of each islandwill have to be filled up by modi-fications of the old inhabitants;and time will be allowed for thevarieties in each to become wellmodified and perfected. When,by renewed elevation, the islandsshall be re-converted into a con-tinental area, there will again besevere competition: the most fa-voured or improved varieties willbe enabled to spread: there willbe much extinction of the less im-proved forms, and the relativeproportional numbers of the vari-ous inhabitants of the renewedcontinent will again be changed;and again there will be a fair fieldfor natural selection to improvestill further the inhabitants, andthus produce new species.


This passage gives rise to an interest-ing niche algorithm in which there is atfirst but a single “area,” which is thendivided into many smaller ones that arelater reunited.

TIME CONSIDERATIONS

That natural selection will alwaysact with extreme slowness, I fullyadmit. . . I do believe that naturalselection will always act veryslowly, often only at long inter-vals of time, and generally ononly a very few of the inhabitantsof the same region at the sametime.


Though nature grants vast peri-ods of time for the work of natu-ral selection, she does not grantan indefinite period. . .


Evolution is slow. In many cases wemay expect artificial evolution to exhibit


this same characteristic (though on dif-ferent time scales). Thus, one shouldnot despair if his or her 3-week simula-tion is still churning on.

One could argue here that evolutionmight at times be quite fast, for ex-ample, where bacteria and viruses areinvolved (which, by the way, weremostly unknown in Darwin’s time). Ishould like to counterargue by sayingthat evolution is fast when the problemis easy. Novel forms of viruses canevolve quickly since often only a smallseries of mutations from an extant formis needed. I believe this holds for artifi-cial evolution as well: Truly hard prob-lems do require intensive computationand long evolution times; if not, thenthey were probably not that hard to be-gin with, or we have managed to renderthem easier by custom tailoring theevolutionary algorithm (the genomicrepresentation or the genetic operators,for example).

ONTOGENY

There are many laws regulatingvariation, some few of which canbe dimly seen, and will be here-after briefly mentioned. I willhere only allude to what may becalled correlation of growth. Anychange in the embryo or larva willalmost certainly entail changes inthe mature animal.

Chapter 1: Variation UnderDomestication

Correlation of Growth. I mean bythis expression that the whole or-ganisation is so tied together dur-ing its growth and development,that when slight variations in anyone part occur, and are accumu-lated through natural selection,other parts become modified.This is a very important subject,most imperfectly understood. . .any malconformation affectingthe early embryo, seriously af-fects the whole organisation ofthe adult.

Chapter 5: Laws of Variation

But it is by no means obvious, onthe ordinary view, why the struc-

ture of the embryo should be

more important for this purpose

than that of the adult, which

alone plays its full part in the

economy of nature.

Chapter 13: Mutual Affinities of

Organic Beings: Morphology:

Embryology: Rudimentary

Organs

Ontogeny is the process by which a

single mother cell, the zygote, gives rise,

through successive divisions, to a com-

plete organism, possibly containing tril-

lions of cells (for example, in humans).Darwin discussed this issue at somelength in Chapter 13 (“Mutual Affinitiesof Organic Beings: Morphology: Embry-ology: Rudimentary Organs”). Essen-tially, ontogeny is the process that mapsthe genotype to the phenotype. Thismapping is quite simple in most currentevolutionary algorithms, though somemore complex (ontogenetic) ones havebeen proposed. In such cases, one in-deed notes the intricate intra-genomicrelations, that is, “when slight variationsin any one part occur. . . other parts be-come modified.” One of the advantagesof ontogeny concerns information com-pression—the ability to represent largephenotypes by much smaller genotypes(as is the case, for example, with hu-mans). Such compression leads, amongothers, to improved scalability. Notethat it is the phenotype, that is, the re-sult of the ontogenetic process, that“plays its full part in the economy ofnature.”

ECONOMY OF STRUCTURE

I suspect, also, that some of thecases of compensation whichhave been advanced, and likewisesome other facts, may be mergedunder a more general principle,namely, that natural selection is

continually trying to economisein every part of the organisation.If under changed conditions oflife a structure before useful be-comes less useful, any diminu-tion, however slight, in its devel-opment, will be seized on bynatural selection, for it will profitthe individual not to have its nu-triment wasted in building up anuseless structure.


Economy of structure may be ob-served when using evolutionary algo-rithms in which the genotypic structureis subject to change (for example, ge-netic programming)—if a cost is associ-ated with structural complexity, as innature (for example, a simple way of do-ing this in genetic programming wouldbe by penalizing larger programs). Notethat the case for economy is not clear-cut since “a part formerly of high im-portance has often been retained. . .though it has become of such small im-portance. . .” (Chapter 6).

VARIABILITY BEGETS VARIABILITY

Any part or organ developed toan extraordinary size or in an ex-traordinary manner, in compari-son with the same part or organin the allied species, must havegone through an extraordinaryamount of modification since thegenus arose; and thus we can un-derstand why it should often stillbe variable in a much higher de-gree than other parts; for varia-tion is a long-continued and slowprocess, and natural selectionwill in such cases not as yet havehad time to overcome the ten-dency to further variability and toreversion to a less modified state.


It is only in those cases in whichthe modification has been com-paratively recent and extraordi-narily great that we ought to findthe generative variability, as itmay be called, still present in ahigh degree.


One of the advantages of ontogenyconcerns information

compression—the ability torepresent large phenotypes by much

smaller genotypes.


. . .and this implies in most casesthat there has been recent vari-ability; and therefore we mightexpect that such variability wouldoften continue and be super-added to that arising from themere act of crossing.

Chapter 8: Hybridism

This is interesting in that it suggeststhat variability itself is inheritable, andnot a mere byproduct of genetic opera-tions. Thus, one can try to engineer“generative variability” into one’s evo-lutionary algorithm.

LEVELS OF SELECTION

If we look at the sting of thebee. . . for if on the whole thepower of stinging be useful to thecommunity, it will fulfill all therequirements of natural selec-tion, though it may cause thedeath of some few members.Chapter 6: Difficulties on Theory

This difficulty, though appearinginsuperable, is lessened, or, as Ibelieve, disappears, when it is re-membered that selection may beapplied to the family, as well as tothe individual, and may thus gainthe desired end.

Chapter 7: Instinct

Darwin is discussing here an issuethat has since received wide attentionin biology—the level at which selectionforces act. This can range from selectionat the gene level (an extremist view,mainly championed by Richard Daw-kins in The Selfish Gene), through theindividual level, group level, specieslevel, and higher. Some biologists havebeen advocating a more eclectic viewthat involves selection at several levels.This issue has been given little attentionwithin the evolutionary-computationdomain, with selection performed pre-dominantly at the individual level.

CONDITIONS OF LIFE

It is an old and almost universalbelief, founded, I think on a con-

siderable body of evidence, thatslight changes in the conditionsof life are beneficial to all livingthings.

Chapter 8: Hybridism

This may translate to slight changesin the fitness-function definition duringthe run of the evolutionary algorithm.

RARITY PRECEDES EXTINCTION

It is most difficult always to re-member that the increase of ev-ery living being is constantly be-ing checked by unperceived inju-rious agencies; and that thesesame unperceived agencies areamply sufficient to cause rarity,and finally extinction. We see inmany cases in the more recenttertiary formations, that rarityprecedes extinction. . .

Chapter 10: On the GeologicalSuccession of Organic Beings

Where evolutionary algorithms areconcerned, this may translate into aform of anti-elitist strategy—in order tomaintain diversity one can retain (atleast) some of these rare individuals,before they become extinct.

CONCLUDING REMARKSThe annotated Origin of Species pre-sented above is but a sampling—thereader is encouraged to delve into thebook and seek his own pearls. The ge-stalt effect of Darwin’s work from thestandpoint of evolutionary computa-tion is to remind us of the primitivenessof our algorithms with respect to na-ture. We may, from time to time, incor-porate this feature or that, yet we stillend up with but a simplified simula-crum. Furthermore, nature admits allthe characteristics delineated here (andmany more), which renders currentevolutionary algorithms even further re-moved from reality.

Darwin’s book is by now 140 yearsold, naturally giving rise to the questionof its relevance today. Surprisingly, ithas withstood the test of time, not onlyas concerns the general framework, butalso in many of the details (though ob-

viously not all). One major body ofknowledge that was entirely lacking inDarwin’s time is the molecular under-pinnings of evolution, which have beenelaborated over the past 50 years.Where evolutionary algorithms are con-cerned this lack may have less dire con-sequences than for biology. One canview the Origin as a “high-level” ac-count, with the lower level remainingunspecified. While this may cause bi-ologists to frown, it perfectly servespractitioners in the field of evolutionarycomputation: These practitioners seekexactly such high-level ideas, which arethen implemented in an entirely differ-ent medium than that of molecular bi-ology. This also means that one is “per-mitted” to borrow an idea that has notwithstood the (biological) test of time, ifit proves useful in the digital medium.

It would perhaps be most fitting toend this account with Darwin’s ownending: “There is grandeur in this viewof life, with its several powers, havingbeen originally breathed into a fewforms or into one; and that, whilst thisplanet has gone cycling on according tothe fixed law of gravity, from so simple abeginning endless forms most beautifuland most wonderful have been, and arebeing, evolved.”

ACKNOWLEDGMENTSI am grateful to David B. Fogel, MarcoTomassini, and the anonymous review-ers for their helpful remarks. This workwas supported in part by Grant 2000-049349.96 from the Swiss National Sci-ence Foundation.

NOTE1. In point of historical fact, one should

refer to the Darwin-Wallace (or Wal-lace-Darwin) theory of evolution. Aspointed out by Darwin himself in theintroduction to his book: “. . .I havebeen urged to publish this Abstract. Ihave more especially been induced todo this, as Mr. Wallace, who is nowstudying the natural history of the Ma-lay archipelago, has arrived at almostexactly the same general conclusionsthat I have on the origin of species.”Indeed, both Darwin and Wallace hadpresented their works to the LinneanSociety on the same day.


Documents

Notes on the Origin of Evolutionary Computationsipper/publications/orgevol.pdf · larly characterised. This principle of preservation, I have called, for the sake of brevity, Natural