Sign In
Sign Up
Home
Groups
Case Consultations
Internships/Jobs
Events
Library
Videos
Papers
Classified
Members
Melissa Bridges
Melissa Bridges's Documents
Search Documents
Melissa Bridges's Documents
Hypothesizin..
1 comment | 0 likes
Constructing ..
0 comments | 1 like
THE CONFUSION..
0 comments | 0 likes
More »
Social Share
Archives
2015
The Logical Categories of Learning and Communication
Created by
Melissa Bridges
March 12, 2015
- Category:
Other
- Tags:
#Bateson
#Learning
0 comments, 1,024 views, 0 likes
By Gregory Bateson
The Logi cal Categories of Learni ng and Communi cation Gregory Bateson “Steps to an Ecol ogy of Mi nd” ori gi nal l y publ i shed 1972 The Uni versi ty of Chi cago Press edi ti on 2000 - pag. 279-308 Gr egor y Bat es on was t he s on of pi oneer genet i c i s t Wi l l i am Bat es on and t he hus band of r enowned ant hr opol ogi s t Mar gar et Mead. Hi s wor k s panned many f i el ds , f r om ant hr opol ogy and c ommuni c at i on t heor y t o hi s s t udi es of al c ohol i s m and s c hi zophr eni a at t he Vet er ans Admi ni s t r at i on Hos pi t al i n Pal o Al t o, Cal i f or ni a. Hi s c l as s i c wor k s i nc l ude Nav en, Bal i nes e Char ac t er , c oaut hor ed wi t h Mar gar et Mead, and Mi nd and Nat ur e. The Logi cal Categories of Learni ng and Communi cation* Al l speci es of behavi oral sci enti sts are concerned wi th "l earni ng" i n one sense or another of that word. Moreover, si nce "l earni ng" i s a communi cati onal phenomenon, al l are affected by that cyberneti c revol uti on i n thought whi ch has occurred i n the l ast twenty-fi ve years. Thi s revol uti on was tri ggered by the engi neers and communi cati on theori sts but has ol der roots i n the physi ol ogi cal work of Cannon and Cl aude Bernard, i n the physi cs of Cl arke Maxwel l , and i n the mathemati cal phi l osophy of Russel l and Whi tehead. I nsofar as behavi oral sci enti sts sti l l i gnore the probl ems of Pri nci pi a Mai hemat i ca, 1 they can cl ai m approxi matel y si xty years of obsol escence. I t appears, however, that the barri ers of mi sunderstandi ng whi ch di vi de the vari ous speci es of behavi oral sci enti sts can be i l l umi nated (but not el i mi nated) by an appl i cati on of Russel l 's Theory of Logi cal Types to the concept of "l earni ng" wi th whi ch al l are concerned. To attempt thi s i l l umi nati on wi l l be a purpose of the present essay. The Theory of Logi cal Types Fi rst, i t i s appropri ate to i ndi cate the subj ect matter of the Theory of Logi cal Types: the theory asserts that no cl ass can, i n formal l ogi cal or mathemati cal di scourse, be a member of i tsel f; that a cl ass of cl asses cannot be one of the cl asses whi ch are i ts members; that a name i s not the thi ng named; that "J ohn Bateson" i s the cl ass of whi ch that boy i s the uni que member; and so forth. These asserti ons may seem tri vi al and even obvi ous, but we shal l see l ater that i t i s not at al l unusual for the theori sts of behavioral sci ence to commi t errors whi ch are preci sel y anal ogous to the error of cl assi fyi ng the name wi th the thi ng named—or eati ng the menu card i nstead of the di nner—an error of l ogi cal t ypi ng. Somewhat l ess obvi ous i s the further asserti on of the theory: that a cl ass cannot be one of those i tems whi ch are correctl y cl assi fi ed as i ts nonmembers. I f we cl assi fy chai rs together to consti tute the cl ass of chai rs, we can go on to note that tabl es and l amp shades are members of a l arge cl ass of "nonchai rs," but we shal l commi t an error i n formal di scourse i f we count the cl ass of chai rs among the i tems wi thi n the cl ass of nonchai rs. I nasmuch as no cl ass can be a member of i tsel f, the cl ass of nonchai rs cl earl y cannot be a nonchai r. Si mpl e consi derati ons of symmetry may suffi ce to convi nce the nonmathemati cal reader: (a) that the cl ass of chai rs i s of the same order of abstracti on (i . e. , the same l ogi cal type) as the cl ass of nonchai rs; and further, (b) that i f the cl ass of chai rs i s not a chai r, then, correspondi ngl y, the cl ass of nonchai rs i s not a nonchai r. Lastl y, the theory asserts that i f these si mpl e rul es of formal di scourse are contravened, paradox wi l l be generated and the di scourse vi ti ated. The theory, then, deal s wi th hi ghl y abstract matters and was fi rst deri ved wi thi n the abstract worl d of l ogi c. I n that worl d, when a trai n of proposi ti ons can be shown to generate a paradox, the enti re structure of axi oms, theorems, etc, i nvol ved i n generati ng that paradox i s thereby negated and reduced to nothi ng. I t i s as i f i t had never been. But i n the real worl d (or at l east i n our descri pti ons of i t), there i s al ways t i me, and nothi ng whi ch has been can ever be total l y negated i n thi s way. The computer whi ch encounters a paradox (due to faul ty programmi ng) does not vani sh away. The "i f . . . then . . ." of l ogi c contai ns no ti me. But i n the computer, cause and effect are used to si mul at e the "i f then . . ." of l ogi c; and al l sequences of cause and effect necessari l y i nvol ve ti me. (Conversel y, we may say that i n sci enti fi c expl anati ons the "i f . . . then . . ." of l ogi c i s used to si mul ate the "i f . . . then . . ." of cause and effect.) The computer never trul y encounters l ogi cal paradox, but onl y the si mul ati on of paradox i n trai ns of cause and effect. The computer therefore does not fade away. I t merel y osci l l ates. I n fact, there are i mportant di fferences between the worl d of l ogi c and the worl d of phenomena, and these di fferences must be al l owed for whenever we base our arguments upon the parti al but i mportant anal ogy whi ch exi sts between them. I t i s the thesi s of the present essay that thi s parti al anal ogy can provi de an i mportant gui de for behavi oral sci enti sts i n thei r cl assi fi cati on of phenomena rel ated to l earni ng. Preci sel y i n the fi eld of ani mal and mechani cal communi cati on somethi ng l i ke the theory of types must appl y. Questi ons of thi s sort, however, are not often di scussed i n zool ogi cal l aboratori es, anthropol ogi cal fi el d camps, or psychi atri c conventi ons, and i t i s necessary therefore to demonstrate that these abstract consi derati ons are i mportant to behavi oral sci enti sts. Consi der the fol l owi ng syl l ogi sm: (a) Changes i n frequency of i tems of mammal i an behavi or can be descri bed and predi cted i n terms of vari ous "l aws" of rei nforcement. (b) "Expl orati on" as observed i n rats i s a category, or cl ass, of mammal i an behavi or. (c) Therefore, changes i n frequency of "expl orati on" shoul d be descri babl e i n terms of the same "l aws" of rei nforcement. Be i t sai d at once: fi rst, that empi ri cal data show that the concl usi on (c) i s untrue; and second, that i f the concl usi on (c) were demonstrabl y true, then ei ther (a) or (b) woul d be untrue. 2 Logi c and natural hi story woul d be better served by an expanded and corrected versi on of the concl usi on (c) some-what as fol l ows: (c) I f, as asserted i n {b), "expl orati on" i s not an i t em of mammal i an behavi or but i s a cat egory of such i tems, then no descri pti ve statement whi ch i s true of i t ems of behavi or can be true of "expl orati on." I f, however, descri pti ve statements whi ch are true of i tems of behavi or are al so true of "expl orati on," then "expl orati on" i s an i tem and not a category of i tems. The whol e matter turns on whether the di sti ncti on between a cl ass and i ts members i s an orderi ng pri nci pl e i n the behavi oral phenomena whi ch we study. I n l ess formal l anguage: you can rei nforce a rat (posi ti vel y or negati vel y) when he i nvesti gates a parti cul ar strange obj ect, and he wi l l appropri atel y l earn to approach or avoi d i t. But the very purpose of expl orati on i s to get i nformati on about whi ch obj ects shoul d be approached and whi ch avoi ded. The di scovery that a gi ven obj ect i s dangerous i s therefore a success i n the busi ness of getti ng i nformati on. The success wi l l not di scourage the rat from future expl orati on of other strange obj ects. A pri ori i t can be argued that al l percepti on and al l response, al l behavi or and al l cl asses of behavi or, al l l earni ng and al l geneti cs, al l neurophysi ol ogy and endocri nol ogy, al l organi zati on and al l evol uti on—one enti re subj ect matter—must be regarded as communi cati onal i n nature, and therefore subj ect to the great general i zati ons or "l aws" whi ch appl y to communi cati ve phenomena. We therefore are warned to expect to fi nd i n our data those pri nci pl es of order whi ch fundamental communi cati on theory woul d propose. The Theory of Logi cal Types, I nformati on Theory, and so forth, are expectabl y to be our gui des. The "Learni ng" of Comput ers, Rat s, and Men The word "l earni ng" undoubtedl y denotes change of some ki nd. To say what ki nd of change i s a del i cate matter. However, from the gross common denomi nator, "change," we can deduce that our descri pti ons of "l earni ng" wi l l have to make the same sort of al l owance for the vari eti es of l ogi cal type whi ch has been routi ne i n physi cal sci ence si nce the days of Newton. The si mpl est and most fami l i ar form of change i s mot i on, and even i f we work at that very si mpl e physi cal l evel we must structure our descri pti ons i n terms of "posi ti on or zero moti on," "constant vel oci ty," "accel erati on," "rate of change of accel erati on," and so on. 3 Change denotes process. But processes are themsel ves subj ect to "change." The process may accel erate, i t may sl ow down, or i t may undergo other types of change such that we shal l say that i t i s now a "di fferent" process. These consi derati ons suggest that we shoul d begi n the orderi ng of our i deas about "l earni ng" at the very si mpl est l evel . Let us consi der the case of speci fi ci ty of response, or zero l earni ng. Thi s i s the case i n whi ch an enti ty shows mi ni mal change i n i ts response to a repeated i tem of sensory i nput. Phenomena whi ch approach thi s degree of si mpl i ci ty occur i n vari ous contexts: (a) I n experi mental setti ngs, when "l earni ng" i s compl ete and the ani mal gi ves approxi matel y 100 percent correct responses to the repeated sti mul us. (b) I n cases of habi tuati on, where the ani mal has ceased to gi ve overt response to what was formerl y a di sturbi ng sti mul us. (c) I n cases where the pattern of the response i s mi ni mal l y determi ned by experi ence and maxi mal l y determi ned by geneti c factors. (d) I n cases where the response i s now hi ghl y stereotyped. (e) I n si mpl e el ectroni c ci rcui ts, where t he ci rcui t st ruct ure i s not i t sel f subj ect t o change resul t i ng f rom t he passage of i mpul ses wi t hi n t he ci rcui t —i . e. , where the causal l i nks between "sti mul us" and "response" are as the engi neers say "sol dered i n." I n ordi nary, nontechni cal parl ance, the word "l earn" i s often appl i ed to what i s here cal l ed "zero l earni ng," i . e. , to the si mpl e recei pt of i nformati on from an external event, i n such a way that a si mi l ar event at a l ater (and appropri ate) ti me wi l l convey the same i nformati on: I "l earn" from the factory whi stl e that i t i s twel ve o'cl ock. I t i s al so i nteresti ng to note that wi thi n the frame of our defi ni ti on many very si mpl e mechani cal devi ces show at l east the phenomenon of zero l earni ng. The questi on i s not, "Can machi nes l earn?" but what l evel or order of l earni ng does a gi ven machi ne achi eve? I t i s worth l ooki ng at an extreme, i f hypotheti cal , case: The "pl ayer" of a Von Neumanni an game i s a mathemati cal fi cti on, comparabl e to the Eucl i dean strai ght l i ne i n geometry or the Newtoni an parti cl e i n physi cs. By defi ni ti on, the "pl ayer" i s capabl e of al l computati ons necessary to sol ve whatever probl ems the events of the game may present; he i s i ncapabl e of not performi ng these computati ons whenever they are appropri ate; he al ways obeys the fi ndi ngs of hi s computati ons. Such a "pl ayer" recei ves i nformati on from the events of the game and acts appropri atel y upon that i nformati on. But hi s l earni ng i s l i mi ted to what i s here cal l ed zero l earni ng. An exami nati on of thi s formal fi cti on wi l l contri bute to our defi ni ti on of zero l earni ng. (1) The "pl ayer" may recei ve, from the events of the game, i nformati on of hi gher or l ower l ogi cal type, and he may use thi s i nformati on to make deci si ons of hi gher or l ower type. That i s, hi s deci si ons may be ei ther strategi c or tacti cal, and he can i denti fy and respond to i ndi cati ons of both the tacti cs and the strategy of hi s opponent. I t i s, however, true that i n Von Neumann's formal defi ni ti on of a "game," al l probl ems whi ch the game may present are concei ved as computabl e, i . e. , whi l e the game may contai n probl ems and i nformati on of many di fferent l ogi cal types, the hi erarchy of these types i s stri ctl y fi ni te. I t appears then that a defi ni ti on of zero l earni ng wi l l not depend upon the l ogi cal typi ng of the i nformati on recei ved by the organi sm nor upon the l ogi cal typi ng of the adapti ve deci si ons whi ch the organi sm may make. A very hi gh (but fi ni te) order of compl exi ty may characteri ze adapti ve behavi or based on nothi ng hi gher than zero l earni ng. (2) The "pl ayer" may compute the val ue of i nformati on whi ch woul d benefi t hi m and may compute that i t wi l l pay hi m to acqui re thi s i nformati on by engagi ng i n "expl oratory" moves. Al ternati vel y, he may make del ayi ng or tentati ve moves whi l e he wai ts for needed i nformati on. I t fol l ows that a rat engagi ng i n expl oratory behavi or mi ght do so upon a basi s of zero l earni ng. (3) The "pl ayer" may compute that i t wi l l pay hi m to make random moves. I n the game of matchi ng penni es, he wi l l compute that i f he sel ects "heads" or "tai l s" at random, he wi l l have an even chance of wi nni ng. I f he uses any pl an or pattern, thi s wi l l appear as a pattern or redundancy i n the sequence of hi s moves and hi s opponent wi l l thereby recei ve i nformati on. The "pl ayer" wi l l therefore el ect to pl ay i n a random manner. (4) The "pl ayer" i s i ncapabl e of "error." He may, for good reason, el ect to make random moves or expl oratory moves, but he i s by defi ni ti on i ncapabl e of "l earni ng by tri al and error." I f we assume that, i n the name of thi s l earni ng process, the word "error" means what we meant i t to mean when we sai d that the "pl ayer" i s i ncapabl e of error, then "tri al and error" i s excl uded from the repertoi re of the Von Neumanni an pl ayer. I n fact, the Von Neumanni an "pl ayer" forces us to a very careful exami nati on of what we mean by "tri al and error" l earni ng, and i ndeed what i s meant by "l earni ng" of any ki nd. The assumpti on regardi ng the meani ng of the word "error" i s not tri vi al and must now be exami ned. There i s a sense i n whi ch the "pl ayer" can be wrong. For exampl e, he may base a deci si on upon probabi l i sti c consi derati ons and then make that move whi ch, i n the l i ght of the l i mi ted avai l abl e i nformati on, was most probabl y ri ght. When more i nformati on becomes avai l abl e, he may di scover that that move was wrong. But t hi s di scovery can cont ri but e not hi ng t o hi s f ut ure ski l l . By defi ni ti on, the pl ayer used correctl y al l the avai l abl e i nformati on. He esti mated the probabi l i ti es correctl y and made the move whi ch was most probabl y correct. The di scovery that he was wrong i n the parti cul ar i nstance can have no beari ng upon future i n-stances. When the same probl em returns at a l ater ti me, he wi l l correct l y go through the same computati ons and reach the same deci si on. Moreover, the set of al ternati ves among whi ch he makes hi s choi ce wi l l be the same set—and correctl y so. I n contrast, an organi sm i s capabl e of bei ng wrong i n a number of ways of whi ch the "pl ayer" i s i ncapabl e. These wrong choi ces are appropri atel y cal l ed "error" when they are of such a ki nd that they woul d provi de i nformati on to the organi sm whi ch mi ght contri bute to hi s future ski l l . These wi l l al l be cases i n whi ch some of the avai l abl e i nformati on was ei ther i gnored or i ncorrectl y used. Vari ous speci es of such profi tabl e error can be cl assi fi ed. Suppose that the external event system contai ns detai l s whi ch mi ght tel l the organi sm: (a) from what set of al ternati ves he shoul d choose hi s next move; and (b) whi ch member of that set he shoul d choose. Such a si tuati on permi ts two orders of error: (1) The organi sm may use correctl y the i nformati on whi ch tel l s hi m from what set of al ternati ves he shoul d choose, but choose the wrong al ternati ve wi thi n thi s set; or (2) He may choose from the wrong set of al ternati ves. (There i s al so an i nteresti ng cl ass of cases i n whi ch the sets of al ternati ves contai n common members. I t i s then possi bl e for the organi sm to be "ri ght" but for the wrong reasons. Thi s form of error i s i nevi tabl y sel f-rei nforci ng.) I f now we accept the overal l noti on that al l l earni ng (other than zero l earni ng) i s i n some degree stochasti c (i . e. , contai ns components of "tri al and error"), i t fol l ows that an orderi ng of the processes of l earni ng can be bui l t upon an hi erarchi c cl assi fi cati on of the types of error whi ch are to be corrected i n the vari ous l earni ng processes. Zero l earni ng wi l l then be the l abel for the i mmedi ate base of al l those acts (si mpl e and compl ex) whi ch are not subj ect to correcti on by tri al and error. Learni ng I wi l l be an appropri ate l abel for the revi si on of choi ce wi thi n an unchanged set of al ternati ves; Learni ng I I wi l l be the l abel for the revi si on of the set from whi ch the choi ce i s to be made; and so on. Learni ng I Fol l owi ng the formal anal ogy provi ded by the "l aws" of moti on (i . e. , the "rul es" for descri bi ng moti on), we now l ook for the cl ass of phenomena whi ch are appropri atel y descri bed as changes i n zero l earni ng (as "moti on" descri bes change of posi ti on). These are the cases i n whi ch an enti ty gi ves at Ti me 2 a di fferent response from what i t gave at Ti me 1, and agai n we encounter a vari ety of cases vari ousl y rel ated to experi ence, physi ol ogy, geneti cs, and mechani cal process: (a) There i s the phenomenon of habi tuati on—the change from respondi ng to each occurrence of a repeated event to not overtl y respondi ng. There i s al so the exti ncti on or l oss of habi tuati on, whi ch may occur as a resul t of a more or l ess l ong gap or other i nterrupti on i n the sequence of repeti ti ons of the sti mul us event. (Habi tuati on i s of especi al i nterest. Specifi ci ty of response, whi ch we are cal l i ng zero l earni ng, i s characteri sti c of al l protopl asm, but i t i s i nteresti ng to note that "habi tuati on" i s perhaps the onl y form of Learni ng I whi ch l i vi ng thi ngs can achi eve wi thout a neural ci rcui t.) (b) The most fami l i ar and perhaps most studi ed case i s that of the cl assi cal Pavl ovi an condi ti oni ng. At Ti me 2 the dog sal i vates i n response to the buzzer; he di d not do thi s at Ti me 1. (c) There i s the "l earni ng" whi ch occurs i n contexts of i nstrumental reward and i nstrumental avoi dance. (d) There i s the phenomenon of rote l earni ng, i n whi ch an i tem i n the behavi or of the organi sm becomes a sti mul us for another i tem of behavi or. (e) There i s the di srupti on, exti ncti on, or i nhi bi ti on of "compl eted" l earni ng whi ch may fol l ow change or absence of rei nforcement. I n a word, the l i st of Learni ng I contai ns those i tems whi ch are most commonl y cal l ed "l earni ng" i n the psycho-l ogi cal l aboratory. Note that i n al l cases of Learni ng I , there i s i n our descri pti on an assumpti on about the "context." Thi s assumpti on must be made expl i ci t. The defi ni ti on of Learni ng I assumes that the buzzer (the sti mul us) i s somehow the "same" at Ti me 1 and at Ti me 2. And thi s assumpti on of "sameness" must al so del i mi t the "context," whi ch must (theoreti cal l y) be the same at both ti mes. I t fol l ows that the events whi ch occurred at Ti me 1 are not, i n our descri pti on, i ncl uded i n our defi ni ti on of the context at Ti me 2, because to i ncl ude them woul d at once create a gross di fference between "context at Ti me 1" and "context at Ti me 2." (To paraphrase Heracl i tus: "No man can go to bed wi th the same gi rl for the fi rst ti me twi ce.") The conventi onal assumpti on that context can be repeated, at l east i n some cases, i s one whi ch the wri ter adopts i n thi s essay as a cornerstone of the thesi s that the study of behavi or must be ordered accordi ng to the Theory of Logi cal Types. Wi t hout t he assumpti on of repeatabl e context (and the hypothesi s that f or t he organi sms whi ch we study the sequence of experi ence i s real l y somehow punctuated i n thi s manner), i t woul d fol l ow that al l "l earni ng" woul d be of one type: namel y, al l woul d be zero l earni ng. Of the Pavl ovi an experi ment, we woul d si mpl y say that the dog's neural ci rcui ts contai n "sol dered i n" from the begi nni ng such characteri sti cs that i n Context A at Ti me 1 he wi l l not sal i vate, and that i n the total l y di fferent Context B at Ti me 2 he wi l l sal i vate. What previ ousl y we cal l ed "l earni ng" we woul d now descri be as "di scri mi nati on" between the events of Ti me 1 and the events of Ti me 1 pl us Ti me 2. I t woul d then fol l ow l ogi cal l y that al l questi ons of the type, "I s thi s behavi or 'l earned' or 'i nnate?" shoul d be answered i n favor of geneti cs. We woul d argue that wi thout the assumpti on of repeatabl e context, our thesi s fal l s to the ground, together wi th the whol e general concept of "l earni ng." I f, on the other hand, the assumpti on of repeatabl e context i s accepted as somehow true of the organi sms whi ch we study, then the case for l ogi cal typi ng of the phenomena of l earni ng necessari l y stands, because the noti on "context" i s i tsel f subj ect to l ogi cal typi ng. Ei ther we must di scard the noti on of "context," or we retai n thi s noti on and, wi th i t, accept the hi erarchi c seri es— sti mul us, context of sti mul us, context of context of sti mul us, etc. Thi s seri es can be spel l ed out i n the form of a hi erarchy of l ogi cal types as fol l ows: Sti mul us i s an el ementary si gnal , i nternal or external . Context of sti mul us i s a metamessage whi ch cl assi f i es the el ementary si gnal . Context of context of sti mul us i s a meta-metamessage whi ch cl assi fi es the metamessage. And so on. The same hi erarchy coul d have been bui l t up from the noti on of "response" or the noti on of "rei nforcement." Al ternati vel y, fol l owi ng up the hi erarchi c cl assi fi cati on of errors to be corrected by stochasti c process or "tri al and error," we may regard "context" as a col l ecti ve term for al l those events whi ch tel l the organi sm among what set of al ternati ves he must make hi s next choi ce. At thi s poi nt i t i s conveni ent to i ntroduce the term "context marker." An organi sm responds to the "same" sti mul us di fferentl y i n di fferi ng contexts, and we must therefore ask about the source of the organi sms's i nformati on. From what percept does he know that Context A i s di fferent from Context B? I n many i nstances, there may be no speci fi c si gnal or l abel whi ch wi l l cl assi fy and di fferenti ate the two contexts, and the organi sm wi l l be forced to get hi s i nformati on from the actual congeri es of events that make up the context i n each case. But, certai nl y i n human l i fe and probabl y i n that of many other organi sms, there occur si gnal s whose maj or functi on i s to cl assi f y contexts. I t i s not unreasonabl e to sup-pose that when the harness i s pl aced upon the dog, who has had prol onged trai ni ng i n the psychol ogi cal l aboratory, he knows from thi s that he i s now embarki ng upon a seri es of contexts of a certai n sort. Such a source of i nformati on we shal l cal l a "context marker," and note i mmedi atel y that, at l east at the human l evel , there are al so "markers of contexts of contexts." For exampl e: an audi ence i s watchi ng Haml et on the stage, and hears the hero di scuss sui ci de i n the con-text of hi s rel ati onshi p wi th hi s dead father, Ophel i a, and the rest. The audi ence members do not i mmedi atel y tel ephone for the pol i ce because they have recei ved i nformati on about the context of Haml ets context. They know that i t i s a "pl ay" and have recei ved thi s i nformati on from many "markers of context of context"—the pl aybi l l s, the seati ng arrangements, the curtai n, etc, etc. The "Ki ng," on the other hand, when he l ets hi s consci ence be pri cked by the pl ay wi thi n the pl ay, i s i gnori ng many "markers of context of context." At the human l evel , a very di verse set of events fal l s wi thi n the category of "context markers." A few exampl es are here l i sted: (a) The Pope's throne from whi ch he makes announcements ex cat hedra, whi ch announcements are thereby endowed with a speci al order of val i di ty. (b) The pl acebo, by whi ch the doctor sets the stage for a change i n the pati ents subj ecti ve experi ence. (c) The shi ni ng obj ect used by some hypnoti sts i n "i nduci ng trance." (d) The ai r rai d si ren and the "al l cl ear." (e) The handshake of boxers before the fi ght. (f) The observances of eti quette. These, however, are exampl es from the soci al l i fe of a hi ghl y compl ex organi sm, and i t i s more profi tabl e at thi s stage to ask about the anal ogous phenomena at the pre-verbal l evel . A dog may see the l eash i n hi s master*s hand and act as i f he knows that thi s i ndi cates a wal k; or he may get i nformati on from the sound of the word "wal k" that thi s type of context or sequence i s comi ng. When a rat starts a sequence of expl oratory acti vi ti es, does he do so i n response to a "sti mul us?" Or i n response to a context? Or i n response to a context marker? These questi ons bri ng to the surface formal probl ems about the Theory of Logi cal Types whi ch must be di scussed. The theory i n í ts ori gi nal form deal s onl y wi th ri g- orousl y di gi tal communi cati on, and i t i s doubtful how far i t may be appl i ed to anal ogue or i coni c systems. What we are here cal l i ng "context markers" may be di gi tal (e. g. t the word "wal k" menti oned above); or they may be anal ogue si gnal s — a bri skness i n the master's movements may i ndi cate that a wal k i s pendi ng; or some part of the comi ng context may serve as a marker (the l eash as a part of the wal k); or i n the extreme case, the wal k i tsel f i n al l i ts compl exi ty may stand for i tsel f, wi th no l abel or marker between the dog and the experi ence. The percei ved event i tsel f may communi cate i ts own occurrence. I n thi s case, of course, there can be no error of the "menu card" type. Moreover, no paradox can be generated because i n purel y anal ogue or i coni c communi cati on there i s no si gnal for "not." There i s, i n fact, al most no formal theory deal i ng wi th anal ogue communi cati on and, i n parti cul ar, no equi val ent of I nformati on Theory or Logi cal Type Theory, Thi s gap i n formal knowl edge i s i nconveni ent when we l eave the rari fi ed worl d of l ogi c and mathemati cs and come face to face wi th the phenomena of natural hi story. I n the natural worl d, communi cati on i s rarel y ei ther purel y di gi tal or purel y anal ogi c Often di screte di gi tal pi ps are combi ned together to make anal ogi c pi ctures as i n the pri nter's hal ftone bl ock; and someti mes, as i n the matter of context markers, there i s a conti nuous gradati on from the ostensi ve through the i coni c to the purel y di gi tal . At the di gi tal end of thi s scal e al l the theorems of i nformati on theory have thei r ful l force, but at the ostensi ve and anal ogi c end they are meani ngl ess. I t seems al so that whi l e much of the behavi oral communi cati on of even hi gher mammal s remai ns ostensi ve or anal ogi c, the i nternal mechani sm of these creatures has become di gi tal i zed at l east at the neuronal l evel . I t woul d seem that anal ogi c communi cati on i s i n some sense more pri mi ti ve than di gi tal and that there i s a broad evol uti onary trend toward the substi tuti on of di gi tal for anal ogi c mechani sms. Thi s trend seems to operate faster i n the evol uti on of i nternal mech- ani sms than i n the evol uti on of external behavi or. Recapi tul ati ng and extendi ng what was sai d above: (a) The noti on of repeatabl e context i s a necessary premi se for any theory whi ch defi nes "l earni ng" as change. (b) Thi s noti on i s not a mere tool of our descri pti on but contai ns the i mpl i ci t hypothesi s that for the organi sms whi ch we study, the sequence of l i fe experi ence, acti on, etc, i s somehow segmented or punctuated i nto subsequences or "contexts" whi ch may be equated or di fferenti ated by the organi sm. (c) The di sti ncti on whi ch i s commonl y drawn between percepti on and acti on, afferent and efferent, i nput and out put, i s for hi gher organi sms i n compl ex si tuati ons not val i d. On the one hand, al most every i tem of acti on may be reported ei ther by external sense or endocepti ve mechani sm o the C.N.S., and i n thi s case the report of thi s i tem becomes an i nput. And, on the other hand, i n hi gher organi sms, percepti on i s not by any means a process of mere passi ve recepti vi ty but i s at l east partl y determi ned by efferent control from hi gher centers. Percepti on, notori ousl y, can be changed by experi ence. I n pri nci pl e, we must al l ow both for the possi bi l i ty that every i tem of acti on or output may create an i tem of i nput; and that percepts may i n some cases partake of the nature of output. I t i s no acci dent that al most al l sense organs are used for the emi ssi on of si gnal s between organi sms. Ants communi cate by thei r antennae; dogs by the pri cki ng of thei r ears; and so on. (d) I n pri nci pl e, even i n zero l earni ng, any i tem of experi ence or behavi or may be regarded as ei ther "sti mul us" or "response" or as both, accordi ng to how the total sequence i s punctuated. When the sci enti st says that the buzzer i s the "sti mul us" i n a gi ven sequence, hi s utterance i mpl i es an hypothesi s about how the organi sm punctuates that sequence. I n Learni ng I , every i tem of percepti on or behavi our may be sti mul us or response or rei nf orcement accordi ng to how the total sequence of i nteracti on i s punctuated. Learni ng II What has been sai d above has cl eared the ground for the consi derati on of the next l evel or l ogi cal type of "l earni ng" whi ch we shal l here cal l Learni ng I I . Vari ous terms have been proposed i n the l i terature for vari ous phenomena of thi s order. "Deutero-l earni ng," 4 "set l earni ng," 5 "l earni ng to l earn," and "transfer of l earni ng" may be menti oned. We recapi tul ate and extend the defi ni ti ons so far gi ven: Zero l earni ng i s characteri zed by speci f i ci t y of response, whi ch—ri ght or wrong—i s not subj ect to correcti on. Learni ng I i s change i n speci f i ci t y of response by correcti on of errors of choi ce wi thi n a set of al ternati ves. Learni ng II i s change i n t he process of Learni ng /, e. g. , a correcti ve change i n the set of al ternati ves from whi ch choi ce i s made, or i t i s a change i n how the sequence of experi ence i s punctuated. Learni ng III i s change i n t he process of Learni ng II, e. g. , a correcti ve change i n the system of set s of al ternati ves from whi ch choi ce i s made. (We shal l see l ater that to demand thi s l evel of performance of some men and some mammal s i s someti mes pathogeni c.) Learni ng IV woul d be change i n Learni ng III, but probabl y does not occur i n any adul t l i vi ng organi sm on thi s earth. Evol uti onary process has, however, created organi sms whose ontogeny bri ngs them to Level I I I . The combi nati on of phyl ogenesi s wi th ontogenesi s, i n fact, achi eves Level I V. Our i mmedi ate task i s to gi ve substance to the defi ni ti on of Learni ng I I as "change i n Learni ng I ," and i t i s for thi s that the ground has been prepared. Bri efl y, I bel i eve that the phenomena of Learni ng I I can al l be i ncl uded under the rubri c of changes i n the manner i n whi ch the stream of acti on and experi ence i s segmented or punctuated i nto contexts together wi th changes i n the use of context markers. The l i st of phenomena cl assi fi ed under Learni ng I i ncl udes a consi derabl e (but not exhausti ve) set of di fferentl y structured contexts. I n cl assi cal Pavl ovi an contexts, the conti ngency pattern whi ch descri bes the rel ati on between "sti mul us" (CS), ani mal s acti on (CR), and rei nforcement (UCS) i s profoundl y di fferent from the conti ngency pattern characteri sti c of i nstrumental contexts of l earni ng. I n the Pavl ovi an case: If sti mul us and a certai n l apse of ti me: t hen rei nforcement. I n the I nstrumental Reward case: // sti mul us and a parti cul ar i tem of behavi or; t hen rei nforcement. I n the Pavl ovi an case, the rei nforcement i s not conti ngent upon the ani mal s behavi or, whereas i n the i nstrumental case, i t i s. Usi ng thi s contrast as an exampl e, we say that Learni ng I I has occurred i f i t can be shown that experi ence of one or more contexts of the Pavl ovi an type resul ts i n the ani mal s acti ng i n some l ater context as though thi s, too, had the Pavl ovi an conti ngency pattern. Si mi l arl y, i f past experi ence of i nstrumental sequences l eads an ani mal to act i n some l ater context as though expecti ng thi s al so to be an i nstrumental context, we shal l agai n say that Learni ng I I has occurred. When so defi ned, Learni ng I I i s adapti ve onl y i f the ani mal happens to be ri ght i n i ts expectati on of a gi ven conti ngency pattern, and i n such a case we shal l expect to see a measurabl e l earni ng t o l earn. I t shoul d requi re fewer tri al s i n the new context to establ i sh "correct" behavi or. I f, on the other hand, the ani mal i s wrong i n hi s i denti fi cati on of the l ater conti ngency pattern, then we shal l expect a del ay of Learni ng I i n the new context. The ani mal who has had prol onged experi ence of Pavl ovi an contexts mi ght never get around to the parti cul ar sort of tri al -and-error behavi or necessary to di scover a correct i nstrumental response. There are at l east four fi el ds of experi mentati on where Learni ng I I has been careful l y recorded: (a) I n human rote l earni ng. Hul l 6 carri ed out very careful quanti tati ve studi es whi ch reveal ed thi s phenomenon, and constructed a mathemati cal model whi ch woul d si mul ate or expl ai n the curves of Learni ng I whi ch he recorded. He al so observed a second-order phenomenon whi ch we may cal l "l earni ng to rote l earn" and publ i shed the curves for thi s phenomenon i n the Appendi x to hi s book. These curves were separated from the mai n body of the book because, as he states, hi s mathemati cal model (of Rote Learni ng I ) di d not cover thi s aspect of the data. I t i s a corol l ary of the theoreti cal posi ti on whi ch we here take that no amount of ri gorous di scourse of a gi ven l ogi cal type can "expl ai n" phenomena of a hi gher type. Hul l 's model acts as a touchstone of l ogi cal typi ng, automati cal l y excl udi ng from expl anati on phenomena beyond i ts l ogi cal scope. That thi s was so—and that Hul l percei ved i t—i s testi moni al both to hi s ri gor and to hi s perspi caci ty. What the data show i s that for any gi ven subj ect, there i s an i mprovement i n rote l earni ng wi th successi ve sessi ons, asymptoti cal l y approachi ng a degree of ski l l whi ch vari ed from subj ect to subj ect. The context for thi s rote l earni ng was qui te compl ex and no doubt appeared subj ecti vel y di fferent to each l earner. Some may have been more moti vated by fear of bei ng wrong, whi l e others l ooked rather for the sati sfacti ons of bei ng ri ght. Some woul d be more i nfl uenced to put up a good record as compared wi th the other subj ects; others woul d be fasci nated to compete i n each sessi on wi th thei r own previ ous showi ng, and so on. Al l must have had i deas (correct or i ncorrect) about the nature of the experi mental setti ng, al l must have had "l evel s of aspi rati on," and al l must have had previ ous experi ence of memori zi ng vari ous sorts of materi al . Not one of Hul l 's subj ects coul d have come i nto the l earni ng context uni nfl uenced by previ ous Learni ng I I . I n spi te of al l thi s previ ous Learni ng I I , and i n spi te of geneti c di fferences whi ch mi ght operate at thi s l evel , al l showed i mprovement over several sessi ons. Thi s i mprovement cannot have been due to Learni ng I because any recal l of the speci fi c sequence of syl l abl es l earned i n the previ ous sessi on woul d not be of use i n deal i ng wi th the new sequence. Such recal l woul d more probabl y be a hi ndrance. I submi t, therefore, that the i mprovement from sessi on to sessi on can onl y be accounted for by some sort of adaptati on to the cont ext whi ch Hul l provi ded for rote l earni ng. I t i s al so worth noti ng that educators have strong opi ni ons about the val ue (posi ti ve or negati ve) of trai ni ng i n rote l earni ng. "Progressi ve" educators i nsi st on trai ni ng i n "i nsi ght," whi l e the more conservati ve i nsi st on rote and dri l l ed recal l . (b) The second type of Learni ng I I whi ch has been experi mental l y studi ed i s cal l ed "set l earni ng." The concept and term are deri ved from Harl ow and appl y to a rather speci al case of Learni ng I I . Broadl y, what Harl ow di d was to present rhesus monkeys wi th more or l ess compl ex gest al t en or "probl ems." These the monkey had to sol ve to get a food reward. Harl ow showed that i f these probl ems were of si mi l ar "set,” i . e. , contai ned si mi l ar types of l ogi cal compl exi ty, there was a carry-over of l earni ng from one probl em to the next. There were, i n fact, two orders of conti ngency patterns i nvol ved i n Harl ow*s experi ments: fi rst the overal l pattern of i nstrumental i sm (i f the monkey sol ves the probl em, t hen rei nforcement); and second, the conti ngency patterns of l ogi c wi thi n the speci fi c probl ems. (c) Bi tterman and others have recentl y set a fashi on i n experi mentati on wi th "reversal l earni ng." Typi cal l y i n these experi ments the subj ect i s fi rst taught a bi nary di scri mi nati on. When thi s has been l earned to cri teri on, the meani ng of the sti mul i i s reversed. I f X i ni ti al l y "meant" R1 and Y i ni ti al l y meant R2, then after reversal X comes to mean R2, and Y comes to mean R1. Agai n the tri al s are run to cri teri on when agai n the meani ngs are reversed. I n these experi ments, the cruci al questi on i s: Does the subj ect l earn about the reversal ? I . e. , after a seri es of reversal s, does the subj ect reach cri teri on i n fewer tri al s than he di d at the begi nni ng of the seri es? I n these experi ments, i t i s conspi cuousl y cl ear that the questi on asked i s of l ogi cal type hi gher than that of questi ons about si mpl e l earni ng. I f si mpl e l earni ng i s based upon a set of tri al s, then reversal l earni ng i s based upon a set of such sets. The paral l el i sm between thi s rel ati on and Russel l s rel ati on between "cl ass" and "cl ass of cl asses" i s di rect. (d) Learni ng I I i s al so exempl i fi ed i n the wel l -known phenomena of "experi mental neurosi s." Typi cal l y an ani mal i s trai ned, ei ther i n a Pavl ovi an or i nstrumental l earni ng con-text, to di scri mi nate between some X and some Y ; e. g. , between an el l i pse and a ci rcl e. When thi s di scri mi nati on has been l earned, the task i s made more di ffi cul t: the el l i pse i s made progressi vel y fatter and the ci rcl e i s fl attened. Fi nal l y a stage i s reached at whi ch di scri mi nati on i s i mpossi bl e. At thi s stage the ani mal starts to show symptoms of severe di sturbance. Notabl y, (a) a nai ve ani mal , presented wi th a si tuati on i n whi ch some X may (on some random basi s) mean ei ther A or B, does not show di sturbance; and (b) the di sturbance does not occur i n absence of the many context markers characteri sti c of the l aboratory si tuati on. 7 I t appears, then, that Learni ng I I i s a necessary preparati on for the behavi oral di sturbance. The i nformati on, "Thi s i s a context for di scri mi nati on," i s communi cated at the begi nni ng of the sequence and underl i ned i n the seri es of stages i n whi ch di scri mi nati on i s made progressi vel y more di ffi cul t. But when di scri mi nati on becomes i mpossi bl e, the structure of the context i s total l y changed. The context markers (e. g. , the smel l of the l aboratory and the experi mental harness) now become mi sl eadi ng because the ani mal i s i n a si tuati on whi ch demands guesswork or gambl i ng, not di scri mi nati on. The en-ti re experi mental sequence i s, i n fact, a procedure for putti ng the ani mal i n the wrong at the l evel of Learni ng I I . I n my phrase, the ani mal i s pl aced i n a typi cal "doubl e bi nd," whi ch i s expectabl y schi zophrenogeni c. 8 I n the strange worl d outsi de the psychol ogi cal l aboratory, phenomena whi ch bel ong to the category Learni ng I I are a maj or preoccupati on of anthropol ogi sts, educators, psychi atri sts, ani mal trai ners, human parents, and chi l dren. Al l who thi nk about the processes whi ch determi ne the character of the i ndi vi dual or the processes of change i n human (or ani mal ) rel ati onshi p must use i n thei r thi nki ng a vari ety of assumpti ons about Learni ng I I . From ti me to ti me, these peopl e cal l i n the l aboratory psychol ogi st as a consul tant, and then are confronted wi th a l i ngui sti c barri er. Such barri ers must al ways resul t when, for exampl e, the psychi atri st i s tal k-i ng about Learni ng I I , the psychol ogi st i s tal ki ng about Learn- i ng I , and nei ther recogni zes the l ogi cal structure of the di fference. Of the mul ti tudi nous ways i n whi ch Learni ng I I emerges i n human affai rs, only three wi l l be di scussed i n thi s essay: (a) I n descri bi ng i ndi vi dual human bei ngs, both the sci enti st and the l ayman commonl y resort to adj ecti ves descri pti ve of "character." I t i s sai d that Mr. J ones i s dependent, hosti l e, fey, fi ni cky, anxi ous, exhi bi ti oni sti c, narci ssi sti c, passi ve, competi ti ve, energeti c, bol d, cowardl y, fatal i sti c, humorous, pl ayful , canny, opti mi sti c, perfecti oni st, carel ess, careful , casual , etc. I n the l i ght of what has al ready been sai d, the reader wi l l be abl e to assi gn al l these adj ecti ves to thei r appropri ate l ogi cal type. Al l are descri pti ve of (possi bl e) resul ts of Learni ng I I , and i f we woul d defi ne these words more careful l y, our defi ni ti on wi l l consi st i n l ayi ng down the conti ngency pattern of that context of Learni ng I whi ch woul d expectabl y bri ng about that Learni ng I I whi ch woul d make the adj ecti ve appl i cabl e. We mi ght say of the "fatal i sti c" man that the pattern of hi s transacti ons wi th the envi ronment i s such as he mi ght have acqui red by prol onged or repeated experi ence as subj ect of Pavl ovi an experi ment; and note that thi s defi ni ti on of "fatal i sm" i s speci fi c and preci se. There are many other forms of "fatal i sm" besi des that whi ch i s defi ned i n terms of thi s parti cul ar context of l earni ng. There i s, for exampl e, the more compl ex type characteri sti c of cl assi cal Greek tragedy where a man's own acti on i s fel t to ai d the i nevi tabl e worki ng of fate. (b) I n the punctuati on of human i nteracti on. The cri ti cal reader wi l l have observed that the adj ecti ves above whi ch purport to descri be i ndi vi dual character are real l y not stri ctl y appl i cabl e to the i ndi vi dual but rather descri be t ransactions between the i ndi vi dual and hi s materi al and human envi ronment. No man i s "resourceful " or "dependent" or "fatal i sti c" i n a vacuum. Hi s characteri sti c, whatever i t be, i s not hi s but i s rather a characteri sti c of what goes on between hi m and somethi ng (or somebody) el se. Thi s bei ng so, i t i s natural to l ook i nto what goes on between peopl e, there to fi nd contexts of Learni ng I whi ch are l i kel y to l end thei r shape to processes of Learni ng I I . I n such systems, i nvol vi ng two or more persons, where most of the i mportant events are postures, acti ons, or utterances of the l i vi ng creatures, we note i mmedi atel y that the stream of events i s commonl y punctuated i nto contexts of l earning by a taci t agreement between the persons regardi ng the nature of thei r rel ati onshi p—or by context markers and taci t agreement that these context markers shal l "mean" the same for both parti es. I t i s i nstructi ve to attempt anal ysi s of an ongoi ng i nterchange between A and B. We ask about any parti cul ar i tem of A's behavi or: I s thi s i tem a sti mul us for B? Or i s i t a response of A to somethi ng B sai d earl i er? Or i s i t a rei nforcement of some i tem provi ded by B? Or i s A, i n thi s i tem, consummati ng a rei nforcement for hi msel f? Etc. Such questi ons wi l l reveal at once that for many i tems of A's behavi or the answer i s often qui te uncl ear. Or i f there be a cl ear answer, the cl ari ty i s due onl y to a taci t (rarel y ful l y expl i ci t) agreement between A and B as to the nature of thei r mutual rol es, i. e. , as to the nature of the contextual structure whi ch they wi l l expect of each other. I f we l ook at such an exchange i n the abstract: a 1 b 1 a 2 b 2 a 3 b 3 a 4 b 4 a 5 b 5 . . . . . , where the a' s refer to i tems of A's behavi or, and the b' s to i tems of B's behavi or, we can take any a 1 and construct around i t three si mpl e contexts of l earni ng. These wi l l be: i . (a i b i a i +1 ), i n whi ch a i i s the sti mul us for b i i i . (b i - 1 a i b i ), i n whi ch a i i s the response to b i - 1 , whi ch response B rei nforces wi th b i i i i . (a i - 1 b i - 1 a i ), i n whi ch a i i s now A’ s rei nforcement of B's b i - 1 , whi ch was response to a i - 1 . I t fol l ows that a i may be a sti mul us for B or i t may be A's response to B, or i t may be A's rei nforcement of B. Beyond thi s, however, i f we consi der the ambi gui ty of the noti ons "sti mul us" and "response," "afferent" and "efferent"—as di scussed above— we note that any a i may al so be a sti mul us for A; i t may be A’ s rei nforcement of sel f; or i t may be A’ s response to some previ ous behavi or of hi s own, as i s the case i n sequences of rote behavi or. Thi s general ambi gui ty means i n fact that the ongoi ng sequence of i nterchange between two persons i s structured onl y by the person's own percepti on of the sequence as a seri es of contexts, each context l eadi ng i nto the next. The parti cul ar manner i n whi ch the sequence i s structured by any parti cul ar person wi l l be determi ned by that person’ s previ ous Learni ng I I (or possi bl y by hi s geneti cs). I n such a system, words l i ke "domi nant" and "submi ssi ve", "succori ng" and "dependent" wi l l take on defi nabl e meani ng as descri pti ons of segments of i nterchange. We shal l say that "A domi nates B" i f A and B show by thei r behavi or that they see thei r rel ati onshi p as characteri zed by sequences of the type a 1 b 1 a 2 where a 1 i s seen (by A and B) as a si gnal defi ni ng condi ti ons of i nstrumental reward or puni shment; b 1 as a si gnal or act obeyi ng these condi ti ons; and a 2 as a si gnal rei nforci ng b 1 . Si mi l arl y we shal l say that "A i s dependent on B" i f thei r rel ati onshi p i s characteri zed by sequences a 1 b 1 a 2 , where a 1 i s seen as a si gnal of weakness; b 1 as a hel pi ng act; and a 2 as an acknowl edgement of b 1 But i t i s up to A and B to di sti ngui sh (consci ousl y or unconsci ousl y or not at al l ) between "domi nance" and "dependence." A “command” can cl osel y resembl e a cry for "hel p." (c) I n psychotherapy, Learni ng I I i s exempl i fi ed most conspi cuousl y by the phenomena of "transference." Orthodox Freudi an theory asserts that the pati ent wi l l i nevi tabl y bri ng to the therapy room i nappropri ate noti ons about hi s rel ati on- shi p to the therapi st. These noti ons (consci ous or unconsci ous) wi l l be such that he wi l l act and tal k i n a way whi ch woul d press the therapi st to respond i n ways whi ch woul d resembl e the pati ent's pi cture of how some i mportant other person (usual l y a parent) treated the pati ent i n the near or di stant past. I n the l anguage of the present paper, the pati ent wi l l try to shape hi s i nterchange wi th the therapi st accordi ng to the premi ses of hi s (the pati ent’ s) former Learni ng I I . I t i s commonl y observed that much of the Learni ng I I whi ch determi nes a pati ent's transference patterns and, i n-deed, determi nes much of the rel ati onal l i fe of al l human bei ngs, (a) dat es f rom earl y i nf ancy, and (b) i s unconsci ous. Both of these general i zati ons seem to be correct and both need some expl anati on. I t seems probabl e that these two general i zati ons are true because of the very nature of the phenomena whi ch we are di scussi ng. We suggest that t hat i s l earned i n Learni ng I I i s a way of punct uat i ng event s. But a way of punct uat i ng i s not true or fal se. There i s nothi ng contai ned i n the proposi ti ons of thi s l earni ng that can be tested agai nst real i ty. I t i s l i ke a pi cture seen i n an i nkbl ot; i t has nei ther correctness nor i ncorrectness. I t i s onl y a way of seei ng the i nkbl ot. Consi der the i nstrumental vi ew of l i fe. An organi sm wi th thi s vi ew of l i fe i n a new si tuati on wi l l engage i n tri al -and-error behavi or i n order to make the si tuati on provi de a posi ti ve rei nforcement. I f he fai l s to get thi s rei nforcement, hi s purposi ve phi l osophy i s not thereby negated. Hi s tri al -and- error behavi or wi l l si mpl y conti nue. The premi ses of "purpose" are si mpl y not of the same l ogi cal type as the materi al facts of l i fe, and therefore cannot easi l y be contradi cted by them. The practi ti oner of magi c does not unl earn hi s magi cal vi ew of events when the magi c does not work. I n fact, the proposi ti ons whi ch govern punctuati on have the general characteri sti c of bei ng sel f-val i dati ng. 9 What we term "con-text" i ncl udes the subj ect's behavi or as wel l as the external events. But thi s behavi or i s control l ed by former Learni ng I I and therefore i t wi l l be of such a ki nd as to mol d the total context to fi t the expected punctuati on. I n sum, thi s sel f-val i dati ng characteri sti c of the content of Learni ng I I has the effect that such l earni ng i s al most i neradi cabl e. I t fol l ows that Learni ng I I acqui red i n i nfancy i s l i kel y to persi st through l i fe. Conversel y, we must expect many of the i mportant characteri sti cs of an adul t’ s punctuati on to have thei r roots i n earl y i nfancy. I n regard to the unconsci ousness of these habi ts of punctuati on, we observe that the "unconsci ous" i ncl udes not onl y repressed materi al but al so most of the processes and habi t s of gestal t percepti on. Subj ecti vel y we are aware of our "dependency" but unabl e to say cl earl y how thi s pattern was constructed nor what cues were used i n our creati on of i t. Learni ng III What has been sai d above about the sel f-val i dati ng character of premi ses acqui red by Learni ng I I i ndi cates that Learni ng I I I i s l i kel y to be di ffi cul t and rare even i n human bei ngs. Expectabl y, i t wi l l al so be di ffi cul t for sci enti sts, who are onl y human, to i magi ne or descri be thi s process. But i t i s cl ai med that somethi ng of the sort does from ti me to ti me occur i n psychotherapy, rel i gi ous conversi on, and i n other sequences i n whi ch there i s profound reorgani zati on of character. Zen Buddhi sts, Occi dental mysti cs, and some psychi atri sts assert that these matters are total l y beyond the reach of l anguage. But, i n spi te of thi s warni ng, l et me begi n to specul ate about what must (l ogi cal l y) be the case. Fi rst a di sti ncti on must be drawn: i t was noted above that the experi ments i n reversal l earni ng demonstrate Learni ng I I whenever there i s measurabl e l earni ng about the fact of reversal . I t i s possi bl e to l earn (Learni ng I ) a gi ven premi se at a gi ven ti me and to l earn the converse premi se at a l ater ti me wi thout acqui ri ng the knack of reversal l earni ng. I n such a case, there wi l l be no i mprovement from one reversal to the next. One i tem of Learni ng I has si mpl y repl aced another i tem of Learni ng I wi thout any achi evement of Learni ng I I . I f, on the other hand, i mprovement occurs wi th successi ve reversal s, thi s i s evi dence for Learni ng I I . I f we appl y the same sort of l ogi c to the rel ati on between Learni ng I I and Learni ng I I I , we are l ed to expect that there mi ght be repl acement of premi ses at the l evel of Learni ng I I wi t hout the achi evement of any Learni ng I I I . Prel i mi nary to any di scussi on of Learni ng I I I , i t i s there-fore necessary to di scri mi nate between mere repl acement wi thout Learni ng I I I and that faci l i tati on of repl acement whi ch woul d be trul y Learni ng Hl . That psychotherapi sts shoul d be abl e to ai d thei r pati ents even i n a mere repl acement of premi ses acqui red by Learni ng I I i s al ready no mean feat when we consi der the sel f-val i dati ng character of such premi ses and thei r more or l ess unconsci ous nature. But that thi s much can be done there i s no doubt. Wi thi n the control l ed and protected setti ng of the therapeuti c rel ati onshi p, the therapi st may attempt one or more of the fol l owi ng maneuvers: (a) to achi eve a confrontati on between the premi ses of the pati ent and those of the therapi st—who i s careful l y trai ned not to fal l i nto the trap of val i dati ng the ol d premi ses; (b) to get the pati ent to act, ei ther i n the therapy room or outsi de, i n ways whi ch wi l l confront hi s own premi ses; (c) to demonstrate contradi cti on among the premi ses whi ch currentl y control the pati ent's behavi or; (d) to i nduce i n the pati ent some exaggerat i on or cari cat ure (e.g., i n dream or hypnosi s) of experi ence based on hi s ol d premi ses. As Wi l l i am Bl ake noted, l ong ago, "Wi thout Contrari es i s no progressi on." (El sewhere I have cal l ed these contradi cti ons at l evel I I "doubl e bi nds.") But there are al ways l oophol es by whi ch the i mpact of contradi cti on can be reduced. I t i s a commonpl ace of l earni ng psychol ogy that whi l e the subj ect wi l l l earn (Learni ng I ) more rapi dl y i f he i s rei nforced every ti me he responds correctl y, such l earni ng wi l l di sappear rather rapi dl y i f rei nforcement ceases. I f, on the other hand, rei nforcement i s onl y occasi onal , the subj ect wi l l l earn more sl owl y but the resul ti ng l earni ng wi l l not easi l y be exti ngui shed when rei nforcement ceases al together. I n other words, the subj ect may l earn (Learni ng I I ) that the context i s such that absence of rei nforcement does not i ndi cate that hi s response was wrong or i nappropri ate. Hi s vi ew of the context was, i n fact, correct unti l the experi menter changed hi s tacti cs. The therapi st must certai nl y so support or hedge the contrari es by whi ch the pati ent i s dri ven that l oophol es of thi s and other ki nds are bl ocked. The Zen candi date who has been assi gned a paradox (koan) must l abor at hi s task "l i ke a mosqui to bi ti ng on an i ron bar." I have argued el sewhere ("Styl e, Grace, and I nformati on i n Pri mi ti ve Art," see p. 128) that an essenti al and necessary functi on of al l habi t formati on and Learni ng I I i s an economy of the thought processes (or neural pathways) whi ch are used for probl em-sol vi ng or Learni ng I . The premi ses of what i s commonl y cal l ed "character"—the defi ni ti ons of the "sel f" —save the i ndi vi dual from havi ng to exami ne the abstract, phi l osophi cal , aestheti c, and ethi cal aspects of many sequences of l i fe. "I don*t know whether i t's good musi c; I onl y know whether I l i ke i t." But Learni ng I I I wi l l throw these unexami ned premi ses open to questi on and change. Let us, as was done above for Learni ng I and I I , l i st some of the changes whi ch we shal l be wi l l i ng to cal l Learni ng I I I . (a) The i ndi vi dual mi ght l earn to form more readi l y those habi ts the formi ng of whi ch we cal l Learni ng I I . (b) He mi ght l earn to cl ose for hi msel f the "l oophol es" whi ch woul d al l ow hi m to avoi d Learni ng I I I . (c) He mi ght l earn to change the habi ts acqui red by Learni ng I I . (d) He mi ght l earn that he i s a creature whi ch can and does unconsci ousl y achi eve Learni ng I I . (e) He mi ght l earn to l i mi t or di rect hi s Learni ng I I . (f ) I f Learni ng I I i s a l earni ng of the contexts of Learni ng I , then Learni ng I I I shoul d be a l earni ng of the contexts of those contexts. But the above l i st proposes a paradox. Learni ng I I I (i . e. , l earni ng about Learni ng I I ) may l ead ei ther to an i ncrease i n Learni ng I I or to a l i mi tati on and perhaps a reducti on of that phenomenon. Certai nl y i t must l ead to a greater fl exi bi l i ty i n the premi ses acqui red by the process of Learni ng I I —a f reedom from thei r bondage. I once heard a Zen master state categori cal l y: "To become accustomed to anythi ng i s a terri bl e thi ng." But any freedom from the bondage of habi t must al so denote a profound redefi ni ti on of the sel f. I f I stop at the l evel of Learni ng I I , "I " am the aggregate of those characteri sti cs whi ch I cal l my "character." "I " am my habi ts of acti ng i n context and shapi ng and percei vi ng the contexts i n whi ch I act. Sel fhood i s a product or aggregate of Learni ng I I . To the degree that a man achi eves Learni ng I I I , and l earns to percei ve and act i n terms of the contexts of contexts, hi s "sel f" wi l l take on a sort of i rrel evance. The concept of "sel f" wi l l no l onger functi on as a nodal argument i n the punctuati on of experi ence. Thi s matter needs to be exami ned. I n the di scussi on of Learni ng I I , i t was asserted that al l words l i ke "dependency," "pri de," "fatal i sm," refer to characteri sti cs of the sel f whi ch are l earned (Learni ng I I ) i n sequences of rel ati onshi p. These words are, i n fact, terms for "rol es" i n rel ati onshi ps and refer to somethi ng arti fi ci al l y chopped out of i nteracti ve sequences. I t was al so suggested that the correct way to assi gn ri gorous meani ng to any such words i s to spel l out the formal structure of the sequence i n whi ch the named characteri sti c mi ght have been l earned. Thus the i nteracti ve sequence of Pavl ovi an l earni ng was proposed as a paradi gm for a certai n sort of "fatal i sm," etc. But now we are aski ng about the contexts of these contexts of l earni ng, i . e. , about the l arger sequences wi thi n whi ch such paradi gms are embedded. Consi der the smal l i tem of Learni ng I I whi ch was menti oned above as provi di ng a "l oophol e" for escape from Learni ng I I I . A certai n characteri sti c of the sel f—cal l i t "persi stence"—i s generated by experi ence i n mul ti pl e sequences among whi ch rei nforcement i s sporadi c. We must now ask about the l arger context of such sequences. How are such sequences generated? The questi on i s expl osi ve. The si mpl e styl i zed experi mental sequence of i nteracti on i n the l aboratory i s generated by and partl y determi nes a network of conti ngenci es whi ch goes out i n a hundred di recti ons l eadi ng out of the l aboratory i nto the processes by whi ch psychol ogi cal research i s desi gned, the i nteracti ons between psychol ogi sts, the economi cs of research money, etc, etc. Or consi der the same formal sequence i n a more "natural " setti ng. An organi sm i s searchi ng for a needed or mi ssi ng obj ect. A pi g i s rooti ng for acorns, a gambl er i s feedi ng a sl ot machi ne hopi ng for a j ackpot, or a man must fi nd the key to hi s car. There are thousands of si tuati ons where l i vi ng thi ngs must persi st i n certai n sorts of behavi or preci sel y because rei nforcement i s sporadi c or i mprobabl e. Learni ng I I wi l l si mpl i fy the uni verse by handl i ng these i nstances as a si ngl e category. But i f Learni ng I I I be concerned wi th the contexts of these i nstances, then the categori es of Learni ng I I wi l l be burst open. Or consi der what the word "rei nforcement" means at the vari ous l evel s. A porpoi se gets a fi sh from the trai ner when he does what the trai ner wants. At l evel I , the fact of the fi sh i s l i nked wi th the "ri ghtness" of the parti cul ar acti on. At l evel I I , the fact of the fi sh confi rms the porpoi se's under-standi ng of hi s (possi bl y i nstrumental or dependent) rel ati onshi p wi th the trai ner. And note that at thi s l evel , i f the porpoi se hates or fears the trai ner, pai n recei ved from the l atter may be a posi ti ve rei nforcement confi rmi ng that hate. ("I f i t*s not the way I want i t, Ti l prove i t.") But what of "rei nforcement" at l evel I I I (for porpoi se or for man) ? I f, as I have suggested above, the creature i s dri ven to l evel I I I by "contrari es" generated at l evel I I , then we may expect that i t i s the resol vi ng of these contrari es that wi l l consti tute posi ti ve rei nforcement at l evel I I I . Such resol uti on can take many forms. Even the attempt at l evel I I I can be dangerous, and some fal l by the waysi de. These are often l abel ed by psychi atry as psychoti c, and many of them fi nd themsel ves i nhi bi ted from usi ng the fi rst person pronoun. For others, more successful , the resol uti on of the contrari es may be a col l apsi ng of much that was l earned at l evel I I , reveal i ng a si mpl i ci ty i n whi ch hunger l eads di rectl y to eat-i ng, and the i denti fi ed sel f i s no l onger i n charge of organi zi ng the behavi or. These are the i ncorrupti bl e i nnocents of the worl d. For others, more creati ve, the resol uti on of contrari es reveal s a worl d i n whi ch personal i denti ty merges i nto al l the processes of rel ati onshi p i n some vast ecol ogy or aestheti cs of cosmi c i nteracti on. That any of these can survi ve seems al most mi racul ous, but some are perhaps saved from bei ng swept away on oceani c feel i ng by thei r abi l i ty to focus i n on the mi nuti ae of l i fe. Every detai l of the uni verse i s seen as proposi ng a vi ew of the whol e. These are the peopl e for whom Bl ake wrote the famous advi ce i n the "Auguri es of I nnocence:" To see t he Worl d i n a Grai n of Sand, And a Heaven i n a Wi l d Fl ower, Hol d Inf i ni t y i n t he pal m of your hand, And Et erni t y i n an hour. The Rol e of Genet i cs i n Psychol ogy Whatever can be sai d about an ani mal s l earni ng or i n-abi l i ty to l earn has beari ng upon the geneti c make-up of the ani mal . And what has been sai d here about the l evel s of l earni ng has beari ng upon the whol e i nterpl ay between geneti c make-up and the changes whi ch that i ndi vi dual can and must achi eve. For any gi ven organi sm, there i s an upper l i mi t beyond whi ch al l i s determi ned by geneti cs. Pl anari ans can probabl y not go beyond Learni ng I . Mammal s other than man are probabl y capabl e of Learni ng I I but i ncapabl e of Learni ng I I I . Man may someti mes achi eve Learni ng I I I . Thi s upper l i mi t for any organi sm i s (l ogi cal l y and presumabl y) set by geneti c phenomena, not perhaps by i ndi vi dual genes or combi nati ons of genes, but by whatever factors control the devel opment of basi c phyl ar characteri sti cs. For every change of whi ch an organi sm i s capabl e, there i s the f act of that capabi l i ty. Thi s fact may be geneti cal l y determi ned; or the capabi l i ty may have been l earned. I f the l atter, then geneti cs may have determi ned the capabi l i ty of l earni ng the capabi l i ty. And so on. Thi s i s i n general true of al l somati c changes as wel l as of those behavi oral changes whi ch we cal l l earni ng. A man's ski n tans i n the sun. But where does geneti cs enter thi s pi cture? Does geneti cs compl etel y determi ne hi s abi l i t y to tan? Or can some men i ncrease thei r abi l i ty to tan? I n the l atter case, the geneti c factors evi dentl y have effect at a hi gher l ogi cal l evel . The probl em i n regard to any behavi or i s cl earl y not "I s i t l earned or i s i t i nnate?" but "Up to what l ogi cal l evel i s l earni ng effecti ve and down to what l evel does geneti cs pl ay a determi nati ve or partl y effecti ve rol e?" The broad hi story of the evol uti on of l earni ng seems to have been a sl ow pushi ng back of geneti c determi ni sm to l evel s of hi gher l ogi cal type. A Not e on Hi erarchi es The model di scussed i n thi s paper assumes, taci tl y, that the l ogi cal types can be ordered i n the form of a si mpl e, unbranchi ng l adder. I bel i eve that i t was wi se to deal fi rst wi th the probl ems rai sed by such a si mpl e model . But the worl d of acti on, experi ence, organi zati on, and l earni ng cannot be compl etel y mapped onto a model whi ch excl udes proposi ti ons about the rel ati on bet ween cl asses of di fferent l ogi cal type. I f C x i s a cl ass of proposi ti ons, and C 2 i s a cl ass of proposi ti ons about the members of Cj ; C 3 then bei ng a cl ass of proposi ti ons about the members of C 2 ; how then shal l we cl assi fy proposi ti ons about the rel ati on bet ween these cl asses? For exampl e, the proposi ti on "As members of C x are to members of C 2) so members of C 2 are to members of C 3 " cannot be cl assi fi ed wi thi n the unbranchi ng l adder of types. The whol e of thi s essay i s bui l t upon the premi se that the rel ati on between C 2 and C 3 can be compared wi th the rel ati on between C 1 and C 2 . I have agai n and agai n taken a stance to the si de of my l adder of l ogi cal types to di scuss the structure of thi s l adder. The essay i s therefore i tsel f an exampl e of the fact that the l adder i s not unbranchi ng. I t fol l ows that a next task wi l l be to l ook for exampl es of l earni ng whi ch cannot be cl assi fi ed i n terms of my hi erarchy of l earni ng but whi ch fal l to the si de of thi s hi erarchy as l earni ng about the rel ati on between steps of the hi erarchy. I have suggested el sewhere ("Styl e, Grace, and I nformati on i n Pri mi ti ve Art") that art i s commonl y concerned wi th l earni ng of thi s sort, i . e. , wi th bri dgi ng the gap between the more or l ess unconsci ous premi ses acqui red by Learni ng I I and the more epi sodi c content of consci ousness and i mmedi ate acti on. I t shoul d al so be noted that the structure of thi s essay i s i nduct i ve i n the sense that the hi erarchy of orders of l earni ng i s presented to the reader from the bottom upward, from l evel zero to l evel I I I . But i t i s not i ntended that the expl anati ons of the phenomenal worl d whi ch the model affords shal l be uni di recti onal . I n expl ai ni ng the model to the reader, a uni di recti onal approach was necessary, but wi thi n the model i t i s assumed that hi gher l evel s are expl anatory of l ower l evel s and vi ce versa. I t i s al so assumed that a si mi l ar refl exi ve rel ati on—both i nducti ve and deducti ve—obtai ns among i deas and i tems of l earni ng as these exi st i n the l i ves of the creatures whi ch we study. Fi nal l y, the model remai ns ambi guous i n the sense that whi l e i t i s asserted that there are expl anatory or determi nati ve rel ati ons between i deas of adj acent l evel s both upward and downward, i t i s not cl ear whether di rect expl anatory rel ati ons exi st between separated l evel s, e.g., between l evel I I I and l evel I or between l evel zero and l evel I I . Thi s questi on and that of the status of proposi ti ons and i deas col l ateral to the hi erarchy of types remai ns unexami ned. Notes: * Thi s essay was wri tten i n 1964 whi l e the author was empl oyed by the Communi cati ons Research I nsti tute, under a Career Devel opment Award (K3-NH-21, 931) from the Nati onal I nsti tute of Mental Heal th. I t was submi tted as a posi ti on paper to the "Conference on Worl d Vi ews" sponsored by the Wenner-Gren Foundati on, August 2-11, 1968. The secti on on "Learni ng I I I " was added i n 1971. 1 A. N. Whi tehead and B. Russel l , Pri nci pi a Mat hemat i ca, 3 vol s., 2nd ed., Cambri dge, Cambri dge Uni versi ty Press, 1910-13. 2 I t i s concei vabl e that the same words mi ght be used i n descri bi ng both a cl ass and i ts members and be true i n both cases. The word "wave" i s the name of a cl ass of movements of parti cl es. We can al so say that the wave i tsel f "moves,” but we shal l be referri ng to a movement of a cl ass of movements. Under fri cti on, thi s metamovement wi l l not l ose vel oci ty as woul d the movement of a parti cl e. 3 The Newtoni an equati ons whi ch descri be the moti ons of a "parti cl e" stop at the l evel of "accel erati on." Change of accel erat i on can onl y happen wi th progressi ve deformati on of the movi ng body, but the Newtoni an "parti cl e" was not made up of "parts" and was therefore (l ogi cal l y) i ncapabl e of deformati on or any other i nternal change. I t was therefore not subj ect to rate of change of accel erati on. 4 G. Bateson, "Soci al Pl anni ng and the Concept of Deutero-Learni ng," Conf erence on Sci ence, Phi l osophy and Rel i gi on, Second Symposi um, New Y ork, Harper, 1942. 5 H. E. Harl ow, "The Formati on of Learni ng Sets," Psychol ogy. Revi ew, 1949, 56: 51-65. 6 E. L. Hul l , et al . , Mat hemat i co-deduct i ve Theory of Rot e Learni ng, New Haven, Y al e Uni versi ty, I nsti tute of Human Rel ati ons, 1940. 7 H. S. Li ddel l , "Refl ex Method and Experi mental Neurosi s," Personal i t y and Behavi or Di sorders, New Y ork, Ronal d Press, 1944. 8 G. Bateson, et al , "Toward a Theory of Schi zophreni a," Behavi oral Sci ence, 1956, 1: 251-64. 9 J . Ruesch and G. Bateson, Communi cat i on: The Soci al Mat ri x of Psychi at ry, New Y ork, Norton, 1951.
0 comments
Share
