SIMULATION OP COGNITIVE PROCESSES PROJECTmx103mz5559/mx103mz5559.pdf · cormected fieldsofsimulationof cognitive processes and arti- ficialintelligence has been permanently incorporated

fr^— Cy. {jC~ Cj /_£^

University of California

SIMULATION OP COGNITIVE PROCESSES PROJECT

A Progress Report to the

Carnegie Corporation of New York

Edward A. Feigenbaum and Julian Feldman

October 1962

University" dFCalifornia"

"~"

SIMULATION OP COGNITIVE PROCESSBS PROJECT

A Progress Report to the

Carnegie Corporation of New York

1. lhe Research Program 1

1.1. Progress on the Elementary Perceivcr and Memorizcr(EPAM), a model of human verbal learning 1

1.2. Progress on a model of binary choice behavior 3

2. The Educational Program 5

2.1. A graduate student research-training course 5

2.2. Colloquia, meetings and seminars 6

2.3. Group in the Sciences 7

3. Research on List-Processing Techniques 7

3.1. TALL, a list-processing language 8

3.2. KLS, a list-processing system 8

3.3. Advanced list-processing language design 8

4. Fellowships and Assistantships 8

5. Publications and Conference Presentations 9

5.1. Working papers and publications of the project 9

5.2. Additional published reports 11

5.3. Additional conference presentations 12

6. Financial Report 12—

varsity of

iTXCM OF t 'KtfßC*

A Pr

Carnegie Corpo

Tbi® is a jre'porfe oa tte pw

Pxoeessss Fro s#por*6d a*s

:. .3 y<--.yAi ■■-'■■:::,. 'ma £isutt reee&r^MJsimted, baa as its so*l

the ad- of

isfons&t&ai processing theosrisg

edue»t£e&"«a. iez&ed* aectea fis \ Ity in.

siaalatlos of zogß.i*im proee. 6jr) *«* 1

Berkeley easpKu

i. The Reaesr«h Pasogras.

i.l. P* i m the Bi@ss& i r^Ssoae . "-7, ft

model of hum .-.^saisg

_XRftJ.l model., knora. as -

■.jie

Institute

■

of .

'

BPAM

at the Secoac. see on -

:,

,i2g

'

BP/VM HI COi

■*_""IS""

processes* association ing processes* and response processes

embedded in the previous EMM II

■ ;

£©__ the sjitssslatioa

of experiments. In other words.,

■

del was prograsrased

to be compatible with the oldes node!* &ss much duplication of

programming ms thereby avoided.s two models d plexity and gaser&liiy ©f the

potential stimuli, the discrimination nets in tahich they asesorted a__d stored, atsd! the discrimination leasaisg processes.

In addition, the association learning processes of the two rcodels

differ cosssides.__M.y_ new model these processes are essea*

tially "serial,** whereas ia the older model they ware essentially

"paired-sßsocis.te-.'* Likewise, the new response processes a:

basically serial in nature. v

lhe EPM 111 node! is described in general terns in Working

Paper #8. A more detailed description will toe prepared shortly.

omenta and ejgplori

■ ■■

ie EPAM II model also itsiAk

place during the year, Ihe learning and performance execative

routines tsere rewritten to: a) sharpen the distinction batmen

immediate memory and the discrimiaaticai net memory, asd b) sep«

aratc further the EPAM programs (as subject) from fee experimenter

and stimulus environment programs. Routines to increase theflegibility of the EPAM . ictuai - laentswere written*

Two experrlmast® in perceptual learns! --:ecogai-»

tion were run, in an apt to c ie interface b.

3

the fißAii theory sad s Lficial intelligence prc&i«ns_ At

issue was the effect of actual cues and prior learning es

machlns pattern reeogal-

Theoretical studies (mc ature surveys) were

made in two verbal learning problems currently active in ihc

psychological literature! i of the so-called "case-

trial learning," snd the Studies ©f the "effective stimulus"

isa serial learning, SPAM interpretations of soma "one-trial

learning** phenomena have been made, and will be forfhcsaaiag in

the near future.

..,__._ Progress, on a model of binary choice behavior

The Eisjor effort dm he past year -has consisted of

GsarfL&T&tim sasd dew St of 'lite hypotSiesis-testieg models

OHM) of binary choice t sea proposed earlier

(Feldßßsa, ISSSj 19&; at 52). Sosse thought and effort has

&X-s$ bees giten to c. v -3. $g_4t2_j lels in the di

tion of gi _._.tf .

'

itractet

in the operator-, -t of the General Problem Solver.jjcdel <_!TM*-2> wjs used -as a. . ficial efc in the rep~

liestion of fswr of tl1; geaat two-choiceexperiments in tha literature, The results in general were

quite favorable and a slight parameter change (ESTM-n) resulted

ia an even better fit to the human datgt* A c&ag»arison to the

4

predictions of -the linear operator statistical learning model

shows a clear superiority of lITM in cases vjhere the linear oper-

ator model has been inadequate—negative recency and conditional

probability series (Feldman, Tonge, and Kanter, 1962)*

lIIM-II was also used to predict human behavior in an

c^jcriment of Engler (1962) « The original experiment demon-

strated clearly the ability of her subjects to "track" the

marginal and conditional probability structure of a series

independently. HTM-II also esdiibits this ability, but HTM-I

and another variant, HTM-111 are scmev_hat better predictors

of human behavior (Feldman, 1962). The linear operator model

is clearly inadequate.

In the wake of these successes, HTM-II was used as a

"subject" to predict the highly structured event series used

in the experiments of Wolin and his associates _ While the

model is able to predict some aspects of human behavior, the

discrepancies between the human behavior and the model's beha-

vior indicate that human subjects are still more sophisticated

than lITM.

Some of these failures of lITM were predictable and so

along with these empirical explorations, work was begun on

developing progracss for reeoguising temporal patterns beyond

the capabilities of HIM. The basic idea in these efforts has

been a notion of receding . lhe subject (and the model) recognize

5

pattenas in stringscf l°s and o*s by receding these strings

into other strings, e.g., runs of l*s and ogs0

g

s and/or differences(Klotz, 1961). This work on receding is being continued.

The models of individual behavior presented in Feldman(1959) have been redone so that there is essentially one basic

structure and the model for each of the three subjects is

developed as a result of changing certain subroutines. This

work has been completed and is being documented.

2. The Educational Program

During the first year, a variety of educational efforts were under-

taken, with varying degrees of success.

2.1. A graduate student research-training course

A graduate student research-training course in the closely-

cormected fields of simulation of cognitive processes and arti-

ficial intelligence has been permanently incorporated in the

Derkeley curriculum as the Spring semester of Business Administra-

tion 210, Advanced Applications of Computers in the Social

Sciences. In the Spring of 1962 the course was tau^it jointlyby Feldman and Feigenbaum, A knowledge of the list processing

computer language IPL-V was required as a prerequisite (sec

below) . A few interesting student projects in advanced listprocessing techniques and heuristic programming emerged. The

chief problem with the course, we think, is that there is not

enough time in one sersester for the students to cover the literature

6

adequately and complete an innovative, creative research

effort. As the general state of the art advances, two-month

projects appear less and less adequate and satisfying.

In addition,, we have initiated an educational program in

list processing techniques (on a non-credit basis). In the

Fall and Spring of 1961 we sponsored courses in InformationProcessing Language V, partly to allow prospective BeA„ 210

students to satisfy the prerequisite, partly to achieve wider

dissemination of these ideas on the campus. The good attendance

in the course led us to initiate a seminar in Spring, 1962

in advanced list processing techniques. This seminar included

instruction in list processing languages other than IFL-V. All

of these courses were given by Mr. Robert Hsu. Continuing inthis vein, Feldman is currently giving three parallel non-creditcourses in the use of the- FORTRAN language.

2.2. Colloquia, meetings, and seminars

Further dissemination of research goals, strategies.

and results was achieved through formal and infomal colloquia,formal presentations at professional meetings, and various

formal and informal seminars. Among these wcresa. An extended series of informal talks to the staff of

the Berkeley Center for Human Learning

b9 A series of three formal presentations to the Center

for Human JLearnlay. Collo<

7

c. Presentations to the colloquium of the Center for

Research In ."fessagemeni Science

d« A series oi entations to the BehavioralScience Stis&y Group at the l&jiversity of California

Medical Center to Sua Francisco

c* Presentations at the American Psychological Association

meetings; th«j Second ZsTcernaticnal ttKtferenee on Infor-

mation Processings the Wester® Psychological Assoei&tiea

meetings; and the tOA Simulation Conference

f. Two presentations to aa teforrcal stedy group ia 'artificialintelligence at the Caster for Advanced Study in the

Behavioral Sciences, Palo Alto.

2.3. Group in the Information Sciences

We have joined with faculty members in business admini-

stration, lisgtsistiss, and engineering -m making th© initial

steps toward the formation of a Group ia the Information Scienceson the Berkeley campus, which would have the authority to super-

vise graduate student education leading to a H_.D.. m InformationSciences.

3. Research on List-Processing Techniques

One of the ansillssy go lie project is to . ice ihe stateof computer teetaicjues for d©> Ati in simulation of cognitive

processes Ca goal ugtich old ia eoaaam < tguea at otJ

«o_s«.

universities)" To this end., mi . following projects and pr^sess.

3.1. Peldmass completed sm XPL-iifa. language for the Pbileo 2000 ccn-puter* The language is us project was ademonstratics that a s& .y list processing : cmbe implemented ea a local machine _Ln stan-weeks instead of man-years, as is common!;.- aed.

3.2* Feigenfeaiaa css_suifced in ihe formulation sad testing of the tow

ELS list processing system. This was a language design effortwhose ultimate goaJ. is the ccsistnsctioa of a list-processingcomputer I»rdware.

3.3. We are supervising cote Ph.»«. thesis project in advanced i-

process^ Imig^zge design* Ia addition, mother p:

d-esigs. a "pr&blem-»riented" language for pure informationprocessing problems (one that could toe co riataple,into IK.-V) was begun in our seminar "

4. FeUGiashaiJS' and Assistantships

No

'

funds usre msA for ttships or fellowships during 11961-62 academic year because we were unable to obtain any good candid?in the late Spring cf 1961 or Fall of 1962* m the Spring of 1962 weposted notices and apprised appropri y members. We obtained asmall slate of qualified cahdi i-shose applications were reviewed byCratchfield, Ifcigefibawa, and Feldman, -, were made for1962-63. Max Allen has completed his written and oral exams in psychology..He will be admitted to candidacy a,c recipient of a

I

-■';■"■.■.::.

grant-in-aM frosi our project. For hi® dissertation, Allen will be doing

a siatulation nodal of mediation p. nes ia verbal learning, Threeresearch asste have also i eds Joseph Hssss will be takihis doctoral exams in Logic aasi Philosophy in ihs Spring of 1963 . RichardStauduh a degree in electrical engineering from Berkeley as_d is a

first-year graduate student i« ma' ties.* Nicholas Evegintaov has adegree is philosophy sad psychology from Cbtfosd and is now a graduate student

in business administratis.

We feel that the pool of high-quality applicants for the fellowships

and assistantships will grow as our overall campus educational effort

takes effect.

5. Publications and conference presentations

5.1. Working papers and pablicatic&s of the project

fTCRCnO PAPER #1, "Education for Research in Computer

Sinuistioss of Cognitive Processes. A Report and Pfc .arch

Summaries, ** edited by B. A., Pelgenoftua* Also sheared2S££E_££9_y__S_!&B^ TOi* 7* RO« 2» April, 1962, pp. 244-257.

WCKKIKe PAPER #2, "Bcpiricol lixploxations of a Hypothesis-

Testing Model of Biscay Choice Bc_iaviortn Julian Feldmsss (%.ith

F* Tosige and H. Kanter), Also appeared as SP-546, System Dev-elopment Corporation, Santa Msolea, California? ami w deliveredat the UCLA Simulation Conference, September, 1961. It .-.illappear in a voiuese of th to be edited by A. Hoggattaad F* B. Balderst auing 1963.

10

WORKING PAPER #3, decent Experiments with the EPAM Simu-

lation of Verbal Learning," E„ A. Feigenbaum. Delivered at the

UCLA Simulation Conference, September 1961, and will appear ina volume of those proceedings to be edited by A. Hoggatt and

F. B. Balderston, forthcoming 1963.WORKING PAPER M, "Abstract of a Generalization of an Ele-

mentary Perceiving and Memorizing Machine," B. A. Feigenbaum(with H. A. Simon). Appeared in Abstracts of Papers" of theSecond International Conf___"_■__<_;> «n Tnfnv^ation Processing.Munich, August 1962.

WORKING PAPER #5, "Computer Siraulation of Cognitive Processes,'Julian Feldman_ Also appeared in Computer Applications in the Beha-vioral Sciences 01. Borko, cd.), Englctrood Cliffs, N. J.: Prentice-Hall, 1962.

WORKING PAPER #6, "TftLL-A List Processor for the Philco2000 Computer," Julian Feldman. Also appeared as Field Note#6200, System Development Corporation, Santa Monica, California,January 1962. and in Cqr^unications of the Association for Cora-

£Hi2___£__i__£_}J££_cL. vol- 5 8 no. 9, September 1962, 454-455.WORKING PAPER ff7, "'Brief Notes on the BPAM Theory of Verbal

Learning," E. A. Feigenbaum (with 11. A, Simon). Forthcoming inVerbal Behavior, and Learning! Problems and Proc..^.,^ cofer andMusgrave, eds„, McGraw-Hill, 1963.—W3.K__NGJ»APBR[_#s, "Generalization of an Elementary Perceivingand Memorizing Machine," l^l^Zgmxmn-Z^ 11. A_ Simon). Al*.

11

appeared in Preprints of the Second International Conference

on Information Processing. Munich, August 1962, and is forth-

coming in the volume of conference proceedings.

WORKING PAPER #9, "Recognition of Patterns in Sequences

of Symbols," Michael Klotz.

WORKING PAPER #10, "Empirical Explorations of the Hypothesis-

Testing Machine: The Engler Experiment," Julian Feldman.

WORKING PAPBR #11, "Ibpirical Explorations of the Hypothesis-

Testing Machine. The Wolin Experiment," Julian Feldman.

5.2 Additional published reports

"A Theory of the Serial Position Effect," B. A. Feigenbaum

(with 11. A. Simon). Appeared as P-2375, The RAND Corporation,

Santa Monica, California, 1961. and in The British Journal of"■■ i am i r in r _ni-n

t ,it

iiir im

_■■!

Psychology, vol. 53, August 1962, 307-320.

"Soviet Cybernetics and Computer Sciences, 1960," B. A.

Feigenbaum. Appeared in Communications of Ihe Association for"" * ' ■■■-■ ... ..

—■

. - -...-! . , ,

t

mm mm

Computing Machinery, vol. 4, no. 12, December 1961, 566~579j

and in the Transactions on Electronic Commuters of the Institute" * ' "*'" **""

" "* " ' *— ' ""~ -"

iT hi i hi i _l»i iimniii ii Wli___»jnm.jn.i

of Radio Engineers, vol. EC-10, no. 4, December 1961, 759-776."Performance of a Reading Task by an Elementary Perceiving

and Memorizing Program," E. A. Feigenbaum (with 11. A. Simon).

appeared as P-2355, The RAND Corporation, Santa Monica, California,1961; and Is forthcoming in Behavioral Science. January 1963.

"Forgetting in an Association Memory," E. A. Feigenbaum.

Appeared as P-2311, The RAND Corporation, Santa Monica, California,

12

6

1961j and was presented at the annual meeting of the Associationfor Computing Machinery, Los Angeles, September 1961, where it

.»-

also appears in the Preprints of the 1961 National Conference of

the Association for Computing Machigerg_t vol. 16, 2C2-2C5."A Note on a Class of Probability Matching Models," Julian

Feldman (with Allen Newell). Pgydiqne.tritca. 1961, vol. 26, September.

5.3. Additional conference presentations

Julian Feldman participated as a panel member in a discussionon computers in the behavioral sciences at the annual meeting ofthe Western Psychological Association, San Francisco, April 1962.

At the annual meeting of the American Psychological Associa-tion held in St. Louis in September 1962, Feldman presented

"Sapirical Explorations of a Hypothesis-Testing Model of BinaryChoice Behavior."

Financial Report

1 July 1961 - 30 June 1962

Academic Salaries

Grants-in-Aid, Research Agsts.

General Assistance 6,410.54 3,026.82 3,383.72Supplies and E_qx»nse

Equipment

3,025.00 1,789.91 1,232.09

Travel

*To be carried over to the fiscal year 1 July 1962 - 30 June 1963

.

Appropriated Expended Balance

13,189.46 13,189.46 -0-

3,250.00 -0- 3,250.00

125.00 125.00 ~0-

—■I*ooo*oo 421.20 _ 578.80--^ggOgOO — IS.352.39 SffiOl*