The Immune System and OtherCognitive Systems

I t is becoming clear that the field of immunology is approaching a paradigm shift.It is agreed by most researchers that the immune system is a complex systemboth in its composition and its behavior. However, the most popular ideas of

immune function treat the immune system in a mechanistic and reductionist man-ner. According to the clonal selection theory the immune system’s function is todefeat pathogens. The immune system identifies foreign antigens and destroysthem. The identification of the foreign is made possible by removing, in the immunesystem’s prenatal development, all receptors that recognize self. Anything that animmune receptor identifies “must be the enemy” [1]. Countering this mainstreamview are a growing number of voices that state the need to change the clonalselection theory or discard it, claiming that such a simplistic appraisal of the im-mune system’s function and mode of action is untenable because, at the molecularlevel, we are closely related to the pathogens that invade us. There is a need toconsider the immune system as a integrative system with the ability to see patternsand understand context [2,3]. It is in the context of this argument about the immunesystem that we present our theory of cognitive systems and claim that the immune

system should be seen as such a cognitive system.

The phrase “cognitive system” is used in many fields to describe the various

faculties that we and other organisms use to perceive and interact with the world.

Despite its widespread use, the phrase “cognitive systems” has not yet been definedin a way that can be applied to all of the cases in which it is used.

We suggest the following criterion to differ between cognitive and noncognitivesystems: In cognitive systems the perceptual sensitivities of the system are notpreordained only by the plan of the system but need an interaction with theirenvironment to define the system’s exact sensitivities.

URI HERSHBERG AND SOL EFRONI

Uri Hershberg is at theInterdisciplinary Center for NeuronalComputation, at the HebrewUniversity, Jerusalem, Israel. Sol Efroniis at the Computer Science and AppliedMath Department and at theImmunology Department at theWeizmann Institute of Science,Rehovot, Israel.E-mail for communication:uriher@cc.huji.ac.il.

In the following pages we propose a theory on cognitive systems and thecommon strategies of perception, which are at the basis of their

function. We demonstrate that these strategies are easily seen to be inplace in known cognitive systems such as vision and language.

Furthermore we show that taking these strategies into considerationimplies a new outlook on immune function calling for a new appraisal of

the immune system as a cognitive system.

14 C O M P L E X I T Y © 2001 John Wiley & Sons, Inc., Vol. 6, No. 5

We propose a theory of the generalunderlying principles of cognitive sys-tems, their perception regarding the en-vironment, and the way in which theydeal with the complex patterns their en-vironments present. This is essentiallydivided into two phases—a phase ofpriming (top of Figure 1), in which thesystem defines the general properties ofits environment that it knows and aphase of specific interaction (bottom ofFigure 1), in which the cognitive systemutilizes its knowledge of the generalproperties to interact with specific en-counters with its environment. Thesephases are not necessarily chronologi-cal by nature and happen with each in-teraction of the cognitive system withits environment.

C ognitive systems are innately builtwith a tendency toward perceiv-ing certain aspects of the environ-

ment. These tendencies are such thatthey cause the cognitive systems to bereceptive toward seeing certain generalproperties of the environment and ex-amples that embody them. However,the cognitive system will only acquiregeneral properties that are corroboratedin its initial interactions with the envi-ronment. The definition of generalproperties by cognitive systems issomething that is not completely prede-termined but rather defined through in-teraction with the environment; the en-vironment’s reinforcement is what de-fines the final set of general propertiesthat the cognitive system uses to knowits environment. The general propertiesare an important aspect of the shapespace,1 and so they are encountered of-ten in meaningful interactions with theshape space. Examples of such proper-ties which are both generic for and

1Protein shape space is an analogy com-monly used to describe a vector space, inwhich every point describes a configura-tion of the protein [4]. We will use this ina more general way to describe the vectorspace of the various cognitive modalities.

FIGURE 1

In the above figure we explain our theory on cognitive systems, utilizing for this purposeexamples from vision, language and the immune system. The figure shows the two phasesencompassing cognitive perception.

© 2001 John Wiley & Sons, Inc. C O M P L E X I T Y 15

ubiquitous to the relevant environment,are extremely beneficial in learning theenvironment. As such we will call them“useful examples.”

Through specific interactions andbased on the previous tendencies of thesystem certain general properties arecorroborated by “useful examples” ofthese properties appearing in the envi-ronment. This corroboration leads tothe formation of an achieved set of rep-resentations in the system. Specific en-counters with the environment startwith a deconstruction via a detection ofsimilarities to the useful examples, em-bodied in the achieved set. After this de-tection there is a refitting and fine tun-ing to the specific elements of the event.Having identified the specific elementsof the event, there is now a reconstruc-tion of the event together with an asso-ciation to other contextual factors to afunctional/meaningful event that can beappropriately reacted to and added tothe memory of the system. [The twophase are chronologically mingled;both happen simultaneously. Every en-counter is a specific encounter evenwhile the cognitive system is still build-ing its understanding of the generalproperties. Also in many cognitive sys-tems it is not clear if the process of de-fining new general properties evercomes to a complete stop. “Young” cog-nitive systems are less fluent in workingwith the general properties, and it isharder to teach an “old” cognitive sys-tem new tricks, but all interactions withthe environment have elements of bothphases.]

I n this discussion of cognitive sys-tems, we will start with the visualsystem. Visual systems are built ac-

cording to the niche in which they areused. A bee’s view of flowers is differentfrom ours. The limits of sensitivity towavelength and contrast are to a greatextent built in. However, seeing is notmerely a matter of light sensitivity. Wesee because we have learned how to doso; we know what things to “look” forand what they mean. This knowledge isan understanding of the natural contextof vision and the general properties ofthe visual stimulus that we encounter.

As an example, take a look at Figure 2where our knowledge will not help us.

A painting from the brush of Salva-dor Dali, this picture shows how com-plicated it is to know what we are look-ing at when we do not have the naturalcontext to tell us what we should lookfor. The title may give you a hint, “Ap-parition of a Face and Fruit Dish on aBeach.” Once we reveal that the “appa-rition” hides the figure of a dog, it isdoubtful if you could ever again look atthis picture without seeing man’s bestfriend. Our preknowledge is not of spe-cific aspects of every stimulus butrather of their general properties. Theknowledge of the general properties ofvisual stimuli is not something that weare born with; rather, it is acquiredthrough the natural exposure to the en-vironment. The process starts with thenatural visual tendencies of the visualapparatus and the innate bias that weare born with toward certain things,such as complex stimuli and move-ment; however, these innate aspects arenot enough. Given this tendency, the vi-sual system, in its first stages of devel-opment, builds up the tools that laterwill allow the act of vision.

The need for this priming stage andthe existence of a critical period for itsoccurrence is easily seen in people whohave had blocked corneas from infancy.

Such people, if their cornea is removedtoo late, will remain functionally blind;for although they can react to light, theycannot resolve the images they see [5](much like living in a painting by Dali).In those of us free to see the world, theprocess of learning the general proper-ties is unsupervised and is based on theexistence of useful examples for thegeneral properties of naturally encoun-tered visual stimuli.

The general properties are thosethings that are both ubiquitous to thedifferent stimuli and appear in manydifferent meaningful contexts. In visionthese would be things such as edges orregularities of scale [6]. Their high fre-quency and usefulness very naturallycause their corroboration, until eventu-ally we have a visual system with anachieved set of representations basedon the “useful examples” (stage 1 in Fig-ure 1). This “achieved set” is what en-ables, from then on, the resolution ofvisual stimuli: the deconstruction andreconstruction of the second stage,which allows the acquiring of theproper functional understanding of theimage and memory.

Vision starts with the general prop-erties of the image rather than its par-ticulars. In our representation this is thephase of detection (second half of Figure1). Sight, as opposed to taking a picture,

FIGURE 2

“Apparition of a Face and Fruit Dish on a Beach”: Salvador Dali (1970).

16 C O M P L E X I T Y © 2001 John Wiley & Sons, Inc.

is done while knowing what the impor-tant things in the image are. Edges,movement, and other “eye catching”things tell the visual system where tolook. Once this phase of detection anddeconstruction is done, the visual sys-tem can refit to the particular image it isdealing with and can associate it to thecurrent context. This gives the stimulusmeaning and achieves a functional im-age, while widening the knowledge ofthe visual world via memory.

The fact that our visual ability isbased on and inseparable from our per-ception of the general properties of vi-sual stimuli can easily be seen in thekinds of illusions that fool us. Movies,animation, and paintings are all onlygood copies of natural images in thesense that they capture the generalproperties of the visual world.

Note the different scales involved inthe two phases that we have discussedso far regarding vision. The first stage,the stage of priming and obtaining theachieved set, is a stage that responds tosome external input, but does so in aduration orders of magnitude longerthan the time needed to process newinput once the system is fully func-tional. Moreover, the second stage, spe-cific interaction, is a continuous pro-cess going on all the time, whereas thefirst stage is a once in a lifetime oppor-tunity. It takes a long time to put to-gether an organism, but, once it exists,the organism has to use its mechanismsimmediately.

O ne cognitive system where thisseparation in time is clearly seenis in the development of language

in children. As infants and children, welearn to speak, and it is a long and la-borious process. However, once lan-guage is acquired, when we add newwords to our vocabulary, learning themand their correct use is something thatwe can do almost instantaneously. Wewill deal with only one aspect of the ac-quisition of language—the learning ofcorrect syntactic combinations—howwe know the correct order of words thatmake a meaningful sentence. This stepis made possible by the previous stagesof auditory and cognitive development,

which are influenced by various innateand learned biases. The exact amountof learned and innate influence is thescene of many an argument in the cog-nitive sciences that we will not go into[7]. In any case, it is obvious that at thestage of syntactic development the lan-guage system already has a “tendency”that sets the stage for the learning ofsyntactic combination. This can be seenin the fact that the learning of sentenceformation is always preceded, in chil-dren, by a stage of rapid growth in vo-cabulary or “vocabulary explosion,”that brings about the creation of the vo-cabulary necessary for syntactic combi-nations [8].

In studies concerning the use of in-transitive and transitive verbs in syntac-tic combinations, in parents’ conversa-tions with their children, it was seenthat parents use a very small subset ofverbs at a very high frequency whentalking to their children. Words such aswant, come, go, and make account fora high fraction of the verbs used inparental conversation. All these high-frequency verbs are very general, haveuses that are almost empty semanti-cally, and can be said to be generic of

the verb subcategories to which they

belong. In return, all the first verbs used

by children are drawn from this group

of verbs (though individually each

child’s first verb need not be the most

commonly said word of his parent).

Further, once the first verbs are learned

in a certain syntactic construction, the

speed of learning other verbs in the

same syntactic construction, but not

necessarily in other constructions, is

greatly enhanced. This could be indica-

tive of a scenario where the child learns

the first 2 or 3 examples, after which the

others are greatly facilitated (Reference

9 and Figure 3).

In effect, because of the shape space

of languages, in the course of normal

conversation, children are exposed to

the useful example of the different types

of syntactic combinations and the cor-

rect use of language. The first words are

very common and have many uses (they

embody various ideas). In this fashion

the useful examples of language are cor-

roborated, bringing about the forma-

tion of an achieved set of representa-

tions that greatly facilitates the latter

identification of similarities to the use-

FIGURE 3

Figure 3. Cumulative number of different verbs in VO and SVO word combinations produced bya subject as a function of age [9].

© 2001 John Wiley & Sons, Inc. C O M P L E X I T Y 17

ful examples and the formation of cor-rect functional syntactic combinations.

The studies mentioned above dealwith the learning of intransitive verb-dependent combinations and of transi-tive verbs in verb–object (VO) and sub-ject–verb–object (SVO) relations. Theview of the first examples as useful ex-amples is further strengthened by thefact that the first words learned are eas-ily seen to be generic of some relation ofthe speaker to the world. In learning in-transitive verbs children deal with ani-mate objects, and the first few verbsspoken in word combinations (i.e.,learned) always cover categories both ofactive (come, go) and passive/static(fall, sleep) relations [9]. VO syntacticcombinations are learned before SVO.The improvement in one syntactic com-bination is separate from the improve-ment in the other. In 13 of 16 children,the Hebrew verb “raza” (want) was oneof the first VO verbs (and if not, it wasgenerally the first SVO verb). In VO,84.4% of first utterances were requests,whereas in SVO 25.9% were requestsand 44.4% were descriptions of creationor consumption of objects. In learningboth syntactic combinations what is be-ing learned, along with the correct useof VO and SVO, are useful examples ofthe child’s relation to alienable objects(things that can be part of me or notpart of me). The child is learning to add,remove, or maintain objects in his “per-sonal space” [10] and in essence is ac-quiring an achieved set of representa-tions to deal with objects and their ab-stract reflections.

The statistical shape of language issuch that when speaking to those whodo not speak, we will use those wordsthat are useful examples and enable thecognitive task of learning. Languagesare “built” so that when talking insimple language to children, we willcorroborate the general properties ofcorrect speech and sentence formation,thus enabling the acquisition of lan-guage. Possibly, we have here also evi-dence for the priming phase of the mostcognitive of modalities: our ability toexercise abstract thoughts. Togetherwith the general properties of language,we see here that children learn useful

examples of the mental interaction with

alienable objects and define some of the

general properties of abstract thought.

H aving shown how our theory

works in systems, commonly

agreed upon to be cognitive, we

come now, finally, to the immune sys-

tem. It is in treating the immune system

as cognitive that we believe that our

theory is most controversial and also of

the greatest benefit compared with the

present paradigm, the clonal selection

theory. We hope to show, citing various

sources of contemporary research, that

present knowledge of the immune sys-

tem and its interaction with the anti-

genic/molecular patterns of our body

calls for the treatment of the immune

system as a cognitive system.

The immune system has an ability to

identify specific events and changes in

the body. The immune system’s envi-

ronment is the body. It interacts on the

cellular/molecular level. To do this, it

has many types of cells as well as effec-

tor and signaling substances, many of

which are yet to be identified and un-

derstood. However, in general the

population of cells that make the im-

mune system can be characterized as

the populations of cells known as lym-

phocytes. The two most important

groups of lymphocytes are called B cells

and T cells.

Both of these cell families have a

unique ability the create receptors,

which, though they all originate from the

same genetic material, use different com-

binations of this material to create an im-

mense variability in their final form. The

shape of the receptor, which like all pro-

teins is based on the sequence of a cer-

tain gene, implies the shape and type of

molecule that will activate the receptor.

Therefore, this genetic variability gives

the immune system the potential ability

to have receptors that can identify a near

infinite number of molecular shapes. The

molecules that immune receptors iden-tify are commonly known as antigens.The region within the antigen to whichthey attach is known as an epitope. Asingle antigen may have several differentepitopes.

The receptors of B cells identify ex-tracellular substances. The receptors ofT cells identify intracellular substancesby interacting with specialized antigen-presenting proteins known as majorhistocompatability complex (MHC) re-ceptors [11], which are expressed on thesurface of every one of the body’s cells.MHCs present fragments of intracellu-lar proteins, in effect mirroring the in-ternal state of the cell. Together, T cellsand B cells can identify most intra- andextracellular substances. The immunesystem’s identification and reaction to apathogen or other immune events is de-pendent on mutual reaction by both Tcells and B cells to that event [12].

In trying to fit the immune system toour theory of cognitive systems, we aremaking a remark on the kind of recep-tor repertoire that the immune systemforms out of its potential variability.

The potential repertoire of receptorsis immense, between 1011 for B cellsand 1016 for T cells [4]. Because (inmice) the immune system contains onlyabout 108 of each of the types of cellsand every single cell has only one typeof receptor, it is obvious that the actualrepertoire is smaller. If the immunesystem were to have a repertoire builtof every potential receptor it can gener-ate, then in a rat, for example, thiswould necessitate having a spleen 70times the size of the rat’s entire body[13]. What kind of repertoire actually ex-ists, and what factors are important inits formation?

As we mentioned above, immunol-ogy is in the midst of a paradigm shift.There is an especially widespread de-bate on the way in which the immunesystem differentiates between the mo-lecular patterns of the body and foreignpathogens [3]. We will not go into allaspects of this discussion. In plainwords, the current textbook outlook onimmunity, the clonal selection theory,states that anything that an immunecell receptor identifies is a foreignpathogen. According to the clonal selec-tion theory, this state of affairs isbrought about in the following way:During embryonic development im-mune cells are created randomly, eachreactive to a different antigen. Those

18 C O M P L E X I T Y © 2001 John Wiley & Sons, Inc.

cells bearing receptors that bind to self-antigens at a certain level of affinity orabove are eliminated. This is known asnegative selection. At the end of thisprocess any receptors that remain canonly be activated by foreign pathogens[1,14].

This is a classical reductionist theorythat is extremely elegant and simple. Itexplains how we create a repertoire,which on the one hand can identifypathogens and yet does not react falselywith our body’s molecular patterns. It isalso a very mechanistic way of viewingthe workings of the immune system.One major implication of this theory isthat autoimmunity—the reaction of im-mune receptors to self antigens—issomething that exists only as a pathol-ogy and never in a properly functioningimmune system or a healthy body. Evenif we were not about to show that theimmune system works like a cognitivesystem, there is a major problem withthis theory. The elimination of self-repertoire completely ignores otherfunctions of the immune system thatare essentially involved with self, suchas wound healing and combatingcancer [2,15].

Several generally known aspects ofimmune detection, agreed on even bythe most ardent supporters of the clonalselection theory, seem to imply that theimmune system is working as a cogni-tive system: First, the need for costimu-lation of B cells and T cells for immunereaction [12], and second, the fact thatB cells are reacting to extracellular in-formation, whereas T cells react to in-tracellular information. Together theseappear to imply an immune reaction topatterns and context.

Treating the immune system as acognitive system, the idea of building arepertoire in the way suggested by theclonal selection theory becomes lessplausible. The immune system’s envi-ronment is built completely of cellsboth endogenous and exogenous,which at the time of encounter are re-siding in the body. Also, all of this cel-lular and viral life is built of similarbuilding blocks. There is no intrinsicmolecular signal that differentiates be-tween the organic substances of our

body and those of other organisms. Re-moving all receptors to self amounts toremoving all receptors to all of thethings that are common to all cellularlife. Building a system that needs to rec-ognize the important aspects of this en-vironment but is blind to the generalproperties (which are those things thatare ubiquitous in the environment) islike building a human visual systemthat can not become aware of edges.

We are suggesting a model of the im-mune system as a cognitive system.This implies several things about theway the immune system is primed andhow it detects its environment (see Fig-ure 1). As we showed for vision and lan-guage, the priming of the system andbuilding of the achieved set of represen-tations starts by fulfilling innate sys-temic biases or tendency. In this case,this would probably be a geneticallytransferred tendency to present certainprotein examples that are used to buildthe receptor repertoires. These usefulexamples would, as in vision and lan-guage, be examples of the general prop-erties of the living molecular environ-ment. They should, therefore, be ex-amples of self that cause a positiveselection of receptors with at least someminimal affinity to these examples.

We would, therefore, expect to findthat, from the first randomly generatedstock of receptors, the adult repertoireof immune receptors are created by acombination of positive and negativeselection using specific molecular ex-amples of self.

Part of the reason that the clonal se-lection theory is being forced to changeis that positive selection is apparentlyimportant for the creation of the maturerepertoires in both B cells and T cells[16,17]. This is especially evident in Tcells. One form of positive selection isagreed on for T cells, even by the clonalselection theory. T cells must have aminimal affinity for at least one self an-tigen—MHC receptors—if they are tofunction. However it appears that therecognition of self-MHCs is not the onlykind of self-recognition that is neces-sary for proper T-cell development, infact the proteins nested in the MHCs,while selection is in process, affect the

positive selections outcome [18]. Fur-ther, it has been shown, using geneticengineering techniques, that an im-mune system built with fewer kinds offragments presented by MHCs, withinthe context of positive selection, willhave a less diverse T-cell repertoire [18].The selection of T cells within certainboundaries of affinity to MHC receptorsand the fact that MHC receptors onlypresent the fragments of certain pro-teins [11], together show a possiblemechanism by which the immune sys-tem creates an important bias toward acertain population of examples whilecreating the repertoire of receptors.

What are the “useful examples” pre-sented by the MHCs bias? The exacttypes of proteins, fragments of whichare presented by MHC, have not yetbeen characterized. However, let usconsider, given a tendency, what wouldconstitute the useful examples that arecorroborated, and what is the environ-ment for which they create an achievedset? As we mentioned before, the envi-ronment of the immune system is thebody’s cellular life. Candidates for “use-ful examples” would have to have thefollowing properties: they would haveto be part of every cell; they would haveto have been there all through the his-tory of the development of the immunesystem for them to be such a major partof its function; they should be relevantin times of stress (or otherwise theycannot serve extreme conditions usu-ally common in immune response).

Indeed, it is possible to find such aset. The set corresponds to a group ofantigens Cohen has named “homuncu-lar antigens” [2], and all belong to agroup called housekeeping or mainte-nance proteins. Housekeeping proteinsare essential in all cells, because theyare responsible for ongoing energy me-tabolism, protein construction, and ba-sic genetic manipulations.

One group of housekeeping proteinsof special interest is called heat shockproteins (HSPs). HSPs are part of alarger group of proteins called chaper-ones, which are essential in correct pro-tein construction and folding. HSPshelp cells maintain the proper form(and function) of proteins in various

© 2001 John Wiley & Sons, Inc. C O M P L E X I T Y 19

states of emergency and stress. Thesituations when HSPs are most ex-pressed, emergency and stress, are alsothe ones where you may usually findimmune response. That alone sufficesto mark HSPs as useful examples to theimmune system, but more than that,because states of emergency abound inall environments, HSPs are essentialand are expressed in all cells. In fact,these proteins are ubiquitous in cells intimes of stress and are highly preservedthroughout evolution, from prokaryotesto multicellular organisms [19]. All thissuggests HSPs to be part of the usefulexamples that build the achieved setthrough which detection is carried out.

Now that we have proposed this pos-sible achieved set (the homuncular an-tigens), we can go on to the mechanismof detection. If the immune system be-haves according to our suggestedtheory, we would expect several pointsof immune behavior. We have justshown that housekeeping genes in gen-eral and HSPs in particular would begood candidates as “useful examples,”which would imply that the immunesystem and its receptors would have abias in its reaction to these proteinswhen reacting to the environment andperforming immune functions. At leastin HSPs this appears to be the case. Re-ceptors to different types of endog-enous and exogenous HSPs have beenshown to be important to immune re-action. In Cohen and Young’s review ofthe immunological reaction to HSPs,they show that during almost any reac-tion to bacteria and parasites, the rec-ognized antigen detected is a type ofHSP [20]. In other immune activities, Tcells reactive to specific self-proteins,for example, HSP60 and MBP (an essen-tial factor in nervous tissue), have beenshown to enhance the regeneration ofskin and nerve tissue [21–24].

If the immune system is indeedworking as a cognitive system, as wehave described it, then the immunemechanisms of detection should in-clude a means for the detection of thegeneral properties of the environmentalong with a mechanism for refitting to-ward specific encounters. This shouldallow for the deconstruction and recon-

struction of the immune “image.”

Therefore, we are expecting an adult

repertoire of receptors built to react de-

generately, at a median level of affinity,

to a few self-antigens. This repertoire

will not change much during our life-

time. Along side this, we expect a rep-

ertoire of non-self-reactive receptors,

which changes and evolves over time as

the immune system encounters patho-

gens and evolves throughout our life.

Both T cells and B cells have adult

receptor repertoires that are highly

cross-reactive and react degenerately to

self-antigens. These repertoires remain

at a steady level throughout life [13,25].

B cells have an added ability: when ac-

tivated by the innate immune reper-

toire, they start a process of fine tuning,

known as affinity maturation [26], by

which they create highly specific and

accurate receptors that react to a spe-

cific antigens. In B cells we find a sharp

distinction between a self-reactive rep-

ertoire, which remains permanent

throughout life, and a changing popu-

lation of receptors that is a result of the

immune system’s interactions with dif-

ferent immune events [25]. B cells,

while acquiring a growing repertoire

toward specific pathogens, during the

lifetime of the organism, maintain a

permanent repertoire of cross-reactive

receptors, possibly through inter-

action with the less changeable T-cell

repertoire.

The stability of T-cell repertoires and

fluidity of B-cell repertoires could very

well be a way to allow the process of

detection and refitting, while keeping

focused on the achieved set of molecular

life (homuncular antigens).

The existence of self-reactive recep-

tors does not mean that an autoim-

mune reaction of the immune system

is something that happens in health.

It means that such an occurrence

is avoided, not through the lack of

self-reactive immune receptors, butrather through their heavily controlledexistence.

Although we have not described ex-act cellular mechanisms of detectionand interaction, we feel that we havegiven enough of an argument to justify

treating the immune system as a cogni-tive system.

The immune system’s ability tofunction is dependent on its under-standing the environment in a cognitivemanner and its ability to discern thecomplex patterns it encounters. Theworld of immune function is the bodyin which it resides. This is the source ofthe examples it uses to see the generali-ties of cellular life. These points imply anew outlook at the receptor repertoireof the immune system, which enables itto react to changing and unexpectedpatterns.

Our self is the source of examples,and yet in the end the immune systemknows not to react to the patterns ofself. This could be because the immunesystem has no receptors for self and sothe immune system can not “see” it. Wewould suggest that this is not so, butrather the immune system has many re-ceptors with a sensitivity for self; how-ever, their interaction is such that it rec-ognizes the patterns of self as what theyare—the background on which the pic-ture of immune events are painted. Self-immunity is the basis not the nemesisof immune function.

I n closing, although we have used vi-sion and language as examples to ex-plain something about the immune

system, still in doing so we have alsoput together principles that are at thebase of all cognitive systems. Thesethree systems are all different in theirparticular components and in the spe-cific fashion in which they interact withtheir part of the environment. Still theyall share the common traits of dealingwith the specifics of their environmentthrough an acquired sensitivity to thegeneral properties of their environment.We would say that this is what makesthem all cognitive systems.

Cognitive systems have the commontrait of functioning in an existing envi-ronment constantly on the brink ofchange. Noncognitive systems are builtwith a limited hardwired perceptualability capable of reacting to certainstimulants and no others. In cognitivesystems, the existence of a built in (ge-netic) tendency promises the acquisi-

20 C O M P L E X I T Y © 2001 John Wiley & Sons, Inc.

tion of generalities proven to be impor-tant over several evolutionary steps.However, because of the need for cor-roboration, the cognitive system retainsan ability to change, over time, the defi-nition of the generalities with thechanging of the environment. The abil-ity to modify the innate tendencies andthe need to build the understanding ofgeneral properties relative to the envi-ronment for every generation is whatsets apart the function of cognitive andnoncognitive systems.

Cognitive understanding deals withspecific things and the ability to recog-nize a great amount of the specific as-pects of its environment. However, theability to deal with all of the varying and

changing aspects of the environmentstarts not from the specifics of the worldbut from generalities. First are learnedthe general metaphors, the useful ex-amples of general properties, which areapplicable in many different ways to theworld. Once they are grasped, it is easyto apply them in many different ways tothe world around us. The fact that gen-eralities take on many different aspectsis not a drawback to learning them. It isone of the main reasons they arelearned first. We have dealt here withonly three systems and only in a super-ficial manner; still, we are sure thatstudying other cognitive systems willshow similar patterns. Furthermore, webelieve that in studying cognitive sys-

tems discovering the nature of theiruseful examples and the general prop-erties they represent will tell us much asto those systems basic function andabilities.

ACKNOWLEDGMENTSWe thank Professor Irun Cohen, of theImmunology Department at the Weiz-mann Institute in Rehovot, and Profes-sor Anat Ninio, of the Psychology De-partment at the Hebrew University inJerusalem. Without their insights andhelp, none of these ideas would havestarted forming, nor could they havereached their final shape. This researchwas supported by a grant from the Min-istry of Science, Israel.

REFERENCES1. Burrnet, F.M. A modification of Jerne’s theory of antibody production using the concept of clonal selection. Aust J Sci 20, 1957, 67–69.2. Cohen, I.R. Tending Adam’s garden. Academic Press, San Diego, CA, 2000.3. Seminars In Immunology. Vol. 12, No. 3, (June 2000, R.E. Langman, M. Cohn, eds.).4. A.S. Perelson, and G. Weisbuch. Immunology for physicists. Rev. Mod. Phys. 69/4, 1997, pp. 1219–1267.5. Clyde, W.O. The human eye. Sinauer, Sunderland, MA, 1999.6. Cutting, J.E. Perception with an eye for motion. MIT, 1986.7. Ingram, D. Child Language Acquisition: Method, description, and explanation. Cambridge University Press, 1989.8. Bates, E., Bretherton, I., Snyder, L., and Beeghly, M. From first words to grammar: individual differences and dissociable mechanisms. Cambridge

University Press, 1988.9. Ninio, A. Model learning in syntactic development: Intransitive verbs. International Journal of Bilingualism, 3, 111–131. (Invited article for a special

issue on Cross-linguistic Studies of Early Grammar, edited by M.M. Vihman, 1999).10. Ninio, A. Pathbreaking verbs in syntactic development and the question of prototypical transitivity. Journal of Child Language. 26/3, 1999, pp.

619–653.11. Germain, R.N., Stefanova, I. The dynamics of T cell receptor signaling: complex orchestration and the key roles of tempo and cooperation. Annu Rev

Immunol 17: 1999, pp. 467–522.12. Langman, R.E., and Cohn, M. A minimal model for the self/non-self discrimination: a return to the basics. Seminars in Immunology 12(3): 2000, pp.

189–195.13. Mason, D. A very high level of crossreactivity is an essential feature of the T-cell receptor. Immunology Today 19(9): 1998, 395–404.14. Janeway, C.A., Travers, P., Walport, M., and Capra, J.D. Basic concepts in immunology. Immuno-biology: the immune system in health and disease.

Chapter 1, especially Figure 1.13. (4th ed., 1999).15. Paul, W.E. Tumor immunology: effector mechanisms in cancer immunity. Fundamental Immunology Chapter 37 (4th ed., 1998).16. Hayakawa, K., Asano, M., Shinton, S.A., Gui, M., Allman, D., Stewart, C.L., Silver, J., Hardy, R.R. Positive selection of natural autoreactive B cells.

Science 285(5424): 1999, 113–116.17. Sebzda et al. Selection of the T cell repertoire. Annu Rev Immunol 17, 1999, 829.18. Barton, G.M., Rudensky, A.Y. Requirement for diverse, low-abundance peptides in positive selection of T cells. Science 283(5398): 1999, 67–70.19. Gupta Redhay, S. Protein phylogenies and signature sequences: a reappraisal of evolutionary relationships among archaebacteria, eubacteria, and

eukaryotes. Microbiol Mol Biol Rev 62(4): 1998, 1435–1491.20. Cohen, I.R., Young, D.B. Autoimmunity, microbial immunity and the immunological homunculus. Immunol Today 12(4): 1991, 105–110.21. Moalem, G., Yoles, E., Leibowitz-Amit, R., Muller-Gilor, S., Mor, F., Cohen, I.R., Schwartz, M. Autoimmune T cells retard the loss of function in injured

rat optic nerves. J Neuroimmunol 106(1–2): 2000, 189–197.22. Schwartz, M., Cohen, I.R. Autoimmunity can benefit self-maintenance. Immunol Today 21(6): 2000, 265–268.23. Abulafia-Lapid, R., Elias, D., Raz, I., Keren-Zur, Y., Atlan, H., Cohen, I.R. T cell proliferative responses of type 1 diabetes patients and healthy

individuals to human hsp60 and its peptides. J Autoimmun 12(2): 1999, 121–129.24. Kipnis, J., Yoles, E., Porat, Z., Cohen, A., Mor, F., Sela, M., Cohen, I.R., Schwartz, M. T cell immunity to copolymer 1 confers neuroprotection on

the damaged optic nerve: possible therapy for optic neuropathies. Proc Natl Acad Sci USA 97(13): 2000, 7446–7451.25. Lacroix-Desmazes, S., Kaveri, S.V., Mouthon, L., Ayouba, A., Malanchere, E., Coutinho, A., Kazatchkine, M.D. Self-reactive antibodies (natural

autoantibodies) in healthy individuals. J Immunol Methods 216(1–2): 1998, 117–137.26. Paul, W.E. Affinity maturation. Fundam Immunol Chapter 25. (4th ed., 1998).

© 2001 John Wiley & Sons, Inc. C O M P L E X I T Y 21