TASSCS : a computer simulation model for simulating organizational behavior.

SCR 4650

TASSCS

A computer simulation model

for simulating organizational behavior

Harko Verhagen

masters thesis

first supervisor: Michael Masuch

second supervisor: Bert Schijf

Department of Sociology

University of Amsterdam

Foreword

This thesis would not have seen the light, if it not were for several people

whom I wish to thank here. First of all, all the people who tried to educate me

and also my family for being there and backing me. Next I wish to thank my

wife Cindy for her belief in me and her patience with me. Several insightful

conversations with CCSOM members Maarten Marx, Zhisheng Huang, László

Pólos and Johan Henselmans helped me in my darker moments and kept me

from stumbling more than necessary. My stay in Pittsburgh was very

important for this thesis and I wish to thank Paul Fishbeck for his offer to

share his house. At Carnegie Mellon University, I was helped with my

problematic struggles with Soar and Lisp by Thomas McGinnis and Brian

Milnes. Apart from his help with these problems David Park also proved to be

a useful critic for my ideas. The intellectual help with both the work on the

model and on this thesis from Michael Masuch and Kathleen Carley, the

co-authors of the models on which my model is based, was indispensable for my

thesis. Special thanks for Dot Marsh, without whom my American adventure

would not have occurred.

Table of contents

Foreword

.

Table of contents ...

Introd uction ... " .... "' . . . 1

Chapter 1 The human problem solving paradigm. 3

1.1. The model of information processing systems . . . 3

1.2. The problem solving behavior of an information

processing system . . . 6

1.3. Human problem solving ... 7

1.4. Human problem solving in organizations . . . 8

Chapter 2 The AI-Revolution: Double AISS ... 11

2.1. The model of an actor in Double-AISS ... 11

2.2. The search space of an actor in Double-AISS . . . 12

2.3. The behavior of Double-AISS . . . 14

2.4. Critique on Double-AISS . . . 15

2.5. Conclusions on Double-AISS ... 16

Chapter 3 Using Soar: Plural Soar ... 18

3.1. The model of an actor in Plural Saar . . . 18

3.2. The search space of an actor in Plural Saar ... 19

3.3. The behavior of Plural Soar . . . 19

3.4. Critique on Plural Saar . . . 20

3.5. Conclusions on Plural Saar . . . 22

Chapter 4

'J~:/~X§~n~S ~~ -~~-~~~~ -~I-~~ -~~~ -~~~r~~

4.1. The model of an actor in TASCCS . . . 23

4.2. The search space of an actor in TASCCS ... 25

4.3. The behavior of TASCCS ... 32

Chapter 5 Results of TASCCS ... 33

Chapter 6 Conclusions ... 35

Appendix 1

What is Soar? ... 37

Appendix 2

The results of the sirnulations .... 38

2.1. Simulation results using the T ASCCS

with

organizational structure and communication ... 38

2.2. Simulation

results

us1ng

T ASCCS

without

organizational structure and communication ... 38

2.3. Simulation results

using

Plural Soar

with

communication . . . 38

2.4. Simulation results using Plural Soar without

communication . . . 39

Appendix 3

Appendix 4

The code of T ASCCS ... 40

The bugs repaired in PI ural Saar 113

Introduction

Organization theory has always used models to represent real world situations

and to predict what will happen when parameters change. This is less

problematic than experimenting in the real world. At first, these models

where mathematical. However, these mathematica! models became more and

more complex, causing great difficulties solving the complex equations. Once

the computer became available as a tool for organization theorists, it was used

to overcome these problems of tractability.

also overcame another problem,

which was not realized very often. Human intellectual capacity is limited (the

famous 'seven plus or minus two' concepts a person can at most focus on at

the same time (Miller 1956)). Since computers are less limited, they can be

used to examine more parameters at the same time, thereby making more

complex models possible.

Most initial models however still where on a very abstract level, mostly

ignoring almost entirely the human side of the organization, namely the

complex interaction of individuals which in the end make the organization

behave the way it does. The writing of computer code also proved to be hard

work and with only minor changes to the models, the gain was small.

Research on the nature and epistemology of model building made some

progress possible, but not enough to keep the interest high enough to pursue

this line of research.

A revolution changed all this, the revolution set on by Artificial

Intelligence (AI). AI showed three possible ways to improve the modelling of

organizations: first, AI showed that the modelling of human decision making

as numerical equations makes it impossible to study most of the nontrivial

aspects of the decision making process. The noncontinuity of the solution

space of real world problems defies the simplicity of numerical data structures

and equation solving algorithms. More likely, real world problems require

symbolic data structures and discontinuous solution spaces

with alternatives

constituted by application of rules (Newell and Simon 1972). A second feature

propagated by AI is the use of object driven rather than procedure driven

programming techniques. The discontinuous search spaces ask for flexible

programming techniques, able to cope with ill-defined problems. Thirdly, AI

provides better view on the epistemological conditions of modelling.

Computer models can be seen as a special kind of theory, to which the same

criteria of validation can be applied as to other empirical theories. The

advantage over other theories, posed in natural language, is the reliability.

Complex deduction system can be applied, using more than seven plus or

minus two concepts.

Double-AISS (Masuch and LaPotin 1989) was the first completed effort to

build an AI-based model of organizational decision making.

was intended as

a follow-up study of the garbage can model of organizational decision making

(Cohen, March and Olsen 1972). This theory was also implemented as a

computer model but based on a numeric model. In the Double-AISS model,

the influence of various parameters on both individual end organizational

level on the problem solving behavior of the individual actors and the

organization as a whole was studied. In 1990, yet another computer model was

developed at Carnegie Mellon University, called Plural Soar (Carley et al.

1991). This model tries to examine the influence of communication and

learning capabilities on the problem solving behavior of independent actors

working on a simple task. The model proposed in this thesis tries to

incorporate some of the features of Double AISS in the Plural SOAR model of

CMU. Added was a model of communication in organizations, replacing the

role taking idea in Double AISS. Some simulations were carried out and the

results were compared to the initial Plural Soar model's outcomes as well as

theoretica! predictions.

Chapter 1 The human problem solving paradigm

The theory of human problem solving, as developed by Newell and Simon in

"Human Problem Solving" (Newell and Simon 1972) is both the underlying

paradigm of Double-AISS and of Soar. Moreover, its information processing

theory, the use of production rules (Post 1943), constitutes one of the central

paradigms of AI (Rich 1983). Connecting cognitive psychology with computer

science, the theory describes problem solving behavior of information

processing systems (IPS) in generali the human brain is one instance of such

systems.

1.1.

The model of information processing systems

According to Newell and Simon, information processing is the creation of

new

information

by applying

operators

to information already known

concerning the task at hand. Operators are rules that specify what new

information can be deduced from information already known. They can be

seen as if-then rules, i.e. if certain information is given, then applying the

operators to this information yields new information.

state 1

Figure 1:

addition to state 1, generating state x+ 1, 1 addition to state 1, gene rating state x+ 1,2

operators produce new information based on information

alread y known.

The information already known makes up the state in which the information

processing system is. Information consists of

symbol structures,

referring to

objects in the task environment.

objects

task environment state 1

0

L.Mill1----z~...-o

L....Mm=;-,..rc-~--1-11~0

L...localF---::::;l;wil~o

symbol structures

Figure 2:

the relation between task environment and symbolic

structures making up the state.

Symbol structures consist of symbols and relations between them. All possible

states together are called the

problem space.

The problem space is defined

implicitly by the

initial state,

the

goal state

and the set of operators.

_s_ta_te_x+_1_,

1--~

y

1

PS Environment Receptors Processor Effectors

~o

Figure 4:

the model of an IPS

The first type of memory contains the information about the current state.

This so-called

short-term memory

(STM) has a very limited capacity. For

instance, the human brain has a span of seven plus or minus two symbols

(Miller 1956). STM is accessed directly.

The

long-term memory

(LTM), whose size is virtually unlimited, contains

the operators which can be applied to transform the current state to a new

state. The IPS processes do not have direct access to the long-term memory,

here retrieval is necessary. This retrieval is associative.

Reception memory

is the third type of memory.

It

is a very short term

storage for information coming from the environment, such as sound, vision,

etc. For human problem solving behavior, this type of memory is not

important.

The fourth type of me1:1-ory is the

external memory.

This can take any

kind of form, e.g. a chessboard, a piece of paper, etc. The use of an external

memory can be seen as an enhancement of the short-term memory (e.g. when

using a piece of paper to multiply numbers as compared to multiplying them

without being able to write down the temporary results) or as a visual help

(such as a chessboard). The capacity of the external memory is virtually

infinite, just as the long-term memory (Newell and Simon 1972). The access

speed of both types of memory is also of the same order of magnitude, but the

access function for external memory can take any form, whereas long-term

memory is associatively accessed.

1.2.

The problem solving behavior of an information processing

system

A problem is a situation in which a goal is known

but it is not known how to

achieve it. Using the operators found in the LTM, an IPS can try to create a

path from the initial state to the goal state. However, most of the time more

than one operator is applicable. Allowing the application of more than one

operator at the same time would make the IPS work in parallel mode rather

than in sequential mode. As humans work in sequentia! mode, Newell and

Simon's theory focusses upon this mode. Working sequentially means that

only one operator can be applied at a given point in time to a state. Therefore,

only one new state can be created at that specific point. So, the IPS has to

decide which operator to apply. This is done by the ordering of preferences.

Some operators are preferred to others, because they yield higher results for

the problem solver.

If

this does not solve the decision problem, search can be

conducted. Search can be conducted basically in two modes. In the first mode,

all possible operators are applied sequentially and their results are evaluated.

If

the goal state is not reached, all possible operators are tried for all possible

new states. When the goal state is not reached, again all possible new states are

developed, etc .. This search technique is called breadth-first search. The other

search mode is called depth-first search. Here

one of the possible new states is

chosen as the new state and this procedure is repeated until no more operators

can be applied or the goal state has been reached. When there are no more

Figure 5:

breadth-first search and depth-first search. Every circle

represents a state, and every line represents an operator. The

numbers show the sequence in which all states are visited.

The left-hand diagram is searched in breadth-first mode, the

right-hand one in depth-first mode.

1.3.

Human problem solving

If

the human brain is indeed an instance of an IPS, Newell and Simon's

paradigm of Information Processing does apply to human brains. According to

Simon, human beings are characterized by limitations on their rationality

(Simon 1955). Simon's theory challenges to the model of economie or

omniscient man. Economie man is supposed to know all the possible

outcomes of all available alternatives, has a complete preference ordering and

can therefore choose the most profitable alternative. Simon's

bou nded

rationality

model outlines the limits of man's rationality. These limits

include:

- a preference ordering which is not complete, e.g. partial

- limits on the ability to predict all consequences of all alternatives

- a satisficing rather than an optimizing search for alternatives

- adaptation of the aspiration level

These limitations interact, but they all stem from one central idea: man's

cognitive capacities are limited. Since this is so, he cannot predict all possible

consequences. Because not all consequences can be foreseen, he cannot

maximize his outcomes but has to settle for a satisficing alternative. Once an

alternative is found that meets his

aspiration level,

search comes to a halt.

Past experience guides the adaptation of the aspiration level.

If

the aspiration

level can easily be met, aspiration increases; conversely, if it is too difficult to

meet the aspiration level, aspiration decreases. The period of time over which

experience is taken into account is called the

adaption period.

1.4.

Human problem solving in organizations

When making decisions in an organization, humans have other

considerations as opposed to when they are on their own (Baron 1983). As in

all groups, the members of the organization interact.

Interaction

consists of

cooperation

and

communication.

Cooperation is the use of other members of

the organization during the problem solving process. Communication is the

transfer of information.

Cooperation can take two forms,

problems can either be moved to

other actors or (2) problems can be attracted from other actors. Members of the

organization may be called to help in solving the problem (the moving of a

problem), or one can offer one's help to other members of the organization

(the attraction of a problem). This is known as task allocation i.e., who should

do which task? In order to make use of cooperation, transfer of information

about the task to be performed by someone else is necessary. Communication

between the members of the organization makes transfer of work possible, so,

coordination and communication go hand in hand.

In organizations, every member is in a power relationship with every

other member of the organization. This relationship may influence the

communication between members. The power relationship of the set of all

request

and

command

respectively.

The question is: when do actors try to cooperate and when do they try to

solve the problem individually, by themselves? A subfield of AI called

Distributed Artificial Intelligence (or DAI for short) is concerned with

cooperation of AI systems in solving a problem together. Task allocation is

one of the subfields of DAL Several mechanisms for task allocation are

proposed by researchers. Same authors try to use a bidding scenario to decide

who solves a certain subproblem (Smith 1980), other authors use

organizational roles as a guidance to decide which actor has to solve a certain

subproblem (Cammarata et al. 1983) and still others try to concentrate upon

skills to find out who should solve a certain subproblem (Smith and David

1981). Fox (Fox 1981) states that these different solutions for the problem of

task allocation are related to the size of the organization. A small group can

coordinate among themselves and use skills as a task allocation guide. Larger

organizations do not have enough time for the communication this costs and

instead use organizational roles to solve the task allocation problem. As the

organizations size increases, different organization structures are being used.

Still larger organizations can use a contracting or bidding scheme,

subproblems are given to the best contractor. One useful sociological theory in

this area is symbolic interactionism, the theory established by the Chicago

School and more specifically by George Herbert Mead (Mead 1934).

states that

people make decisions about what do in a social context based on the image

they have of the possible decisions of the other people in that context, given

that they choose a specific decision. This model of other people's behavior is

internalized by using a so-called 'generalized other'. Models of different

roles

are learned by playing different roles in, for instance, childplay. The

role-models are translated to a model of the general behavior of whoever occupies

that specific role. Expectations of role behavior are shared by members of a

group. An actor can act upon his image of the other actor's image of himself.

So we could postulate that people try to cooperate when they think that other

people think they would cooperate and vice versa. The model an actor has of

the other actors serves to determine with whom to cooperate.

Cooperation is not always helpful when solving problems. A distinction

can be made between well-structured problems and poorl y structured

problems (also called programmed versus non-programmed decisions by

Simon (Simon 1977)). Well-structured problems are routine and common to

the decision maker, solutions are specified in advance by organizational

policies and procedures. They are typically problems solved by lower-level

personnel and in organizations whose market and technology are relatively

stable. Poorly structured problems on the other hand, are unique and novel,

its solutions are creatively determined after the definition of the problem.

They are typically solved by higher-level personnel and are common for

organizations with poorly structured situations. Solving well-structured

problems without specialization of labor does not improve the problem

solving process, on the contrary, since group mernbers become dependant

upon each other, the time needed to solve the problem will increase.

Chapter 2 The AI-Revolution: Double AISS

Double-AISS (Masuch and Lapotin 1989) is a model of the

garbage can theory

(Cohen, March and Olsen 1972). The garbage can theory is a metaphor used to

characterize organized anarchies. Organized anarchies differ from 'normal'

organizations (i.e. organizations as they are viewed in most organization

theories: highly structured, with clear technology and clear preferences), in

the following ways: there are no clear preferences, the technology is unclear

and participation is fluid. In the garbage can model, a decision is an outcome

or interpretation of several relatively independent streams within an

organization. These streams are: problems, solutions, participants and choice

opportunities. Problems, solutions and participants move between choice

opportunities in such a way that the nature of the choice, the time it takes to

make that choice and the problems it solves are all dependant on a relatively

complex intermeshing of several elements, such as the mix of choices

available at that time, the mix of problems that have access to the

organization, the mix of solutions looking for problems, and the outside

demand on the decision maker.

Double-AISS tries to overcome the drawbacks of other models (Padgett

1980; Anderson and Fischer 1986; Carley 1986) of the garbage can theory by

using AI-techniques rather than numerical equations in order to make it

more realistic in the sense that non-continuous problem spaces could be used.

The core of the model is formed by the

actors.

They make the decisions in the

model. They are embedded in an organizational

structure,

which maps the

communication possibilities onto the actors. The content of the

communication is an

issue.

What actors can do with issues is defined by their

skills

and

actions.

The acronym Double-AISS is derived from these five

building blocks.

2.1.

The model of an actor in Double-AISS

In Double-AISS, the actors do not know all alternatives in advance, do not

anticipate all consequences of their actions, do not try to optimize but rather to

satisfice (that is, try to meet some aspiration level, which is dependent on

prior experience), do not have full y ordered preferences and their

commitment to each other or the organization may be limited. In short, their

rationality is bounded.

The structure of the organization in Double-AISS is implemented as a

communication network, it defines which actor can reach which other

actor(s). The content of the communication is an issue. Issues are

multidimensional sets of interrelated subtasks. In Double-AISS, the categorical

task is the production of a memo. This consists of six subtasks: writing,

drafting, typing, editing, approving and copying. Every subtask is a dimension

of the memo -task. Issues belong to one actor.

Skills are the qualifications of an actor. They can be applied to problems, so

if

an actor can draft a memo and has an issue which contains the need for

drafting, then his skill can be applied to that issue, thereby reducing the

number of dimensions of the issue. An actor can choose between several

strategies to solve the memo-task. The possible actions are: reducing an issue,

moving an issue to another actor, attracting an issue from another actor,

combining two issues to one new issue (which has less dimensions than the

two old issues has together), splitting one issue into two new issues (who

have more dimensions together than the old one had) or do nothing.

2.2.

The search space of an actor in Double-AISS

The organization consists of ten actors. These individual decision makers

search through their own problem space depending on seven factors:

1

the search strategy

2

cognitive capacity

foresee the future by conceiving trees of alternatives. In this process,

the actor may think about what another actor would do, given that

the alternative under consideration is chosen. In breadth-first

search, an actor tries all possible alternatives in turn. Both search

modes search until either a satisficing alternative is reached, or

until the cognitive capacity of an actor is exhausted. The choice fora

search alternative is made according to a preference for one of the

search modes.

If

the preferred search strategy gives no result, the

other strategy is given a try.

If

no satisficing alternative is found,

then the last found tolerable alternative is chosen.

If

there also is no

tolerable alternative, then the action "do nothing" is chosen.

2.

the cognitive capacity of an actor is the number of search steps an

actor can make per decision cycle.

3.

The aspiration level of an actor is the number of tasks an actor

wants to reduce in one decision cycle. An actor also has a tolerance

level.

If

the aspiration level is not met when the time has come to

make a decision, but tolerance level is met, the corresponding

action is chosen.

4.

The preferences an actor has concern various parts of the decision

process. There are preferences for subtasks, skills, actions, other

actors, search strategy and decision outcome (whether or not to

reduce one's own workload at the expense of other actors'

workload). They are implemented as a "strength", which is the

probability with which an alternative is chosen (by tossing a

multiple loaded die).

5.

The workload of the organizations is the number of issues that

come into the organization at one decision cycle multiplied with

the number of tasks per issue. The workload of an actor is his part of

the total workload of the organization, the total number of issues

present in the organization multiplied with the number of tasks the

issues consist of.

6.

The commitment of an actor to the organization can be one of the

following two alternatives: the actor is concerned with only his own

workload or he is concerned with the workload of the organization

as a whole (also called individualistic versus cooperative actors

(Baron 1983)). The commitment of an actor determines the ordering

of the strategie alternatives. E.g., an egoistic actor (concerned only

with his own workload) thinks of the moving of an issue to

another actor as a solution, since his own workload decreases.

Attracting an issue from another actor is a non-solution according

to an egoistic actor, since his own workload increases. Altruistic

actors think of moving as a non-solution since the overall

workload is not reduced, the same goes for attracting.

7.

2.3.

The structure of the organization determines the communication

possibilities of an actor.

The behavior of Double-AISS

The behavior of the organization in Double-AISS is the interactive effect of

the decisions made by the individual actors. The order in which rules and data

are accessed is individualized and dependent on the factors described in

section 2.2.

In the simulation, various parameters were varied in order to study their

effect upon several measures. The variation was done in a blocking design

with two variables per block. The independent variables were: cognitive

capacity, workload, structure, aspiration level, adaptation period, the strength

of various preferences and the maximum depth of the search. The dependent

2.4.

Critique on Double-AISS

Some critical points can be raised with respect to Double-AISS. These include:

1.

the implementation of subtasks and skills in Double-AISS

2. the reduce-calculus in Double-AISS

3. the communication structure in Double AISS

4. the thinking-ahead in Double-AISS

5. the authority relations in Double-AISS

1.

The subtasks in Double-AISS are implemented as characters, which

are abbreviations of the task itself. E.g. 'a' stands for 'approve'. The

skills use the same notation. This is not very insightful. When one

uses AI-techniques in order to prevent the sloppiness of numerical

equations, one should avoid the situation that it becomes obscure

what character stands for what subtask or skill, this should be done

more appropriate. Also, subtasks are not related to a subject, i.e. one

cannot know

if

'a' stands for 'approve memo X' or 'approve memo

Y'. Therefor, it is not possible to keep an eye on the order in which

subtasks are carried out for a specific memo, but only in an abstract

sense, which is not always necessary. Approving memo X and

typing memo Y do not have to be carried out in a specific order, but

approving and typing memo X do.

2.

The "reduce" calculus used in Double-AISS is very peculiar indeed.

The garbage can theory states that whether something is a problem

or a solution is dependent upon the point of view taken. This is

implemented in Double-AISS as the use of a minus-sign in front of

a subtask to indicate that it is a solution and a subtask without the

minus-sign indicates a problem. Both skills and issues consist of a

mixture of solutions and problems, but it is difficult

(if

not

impossible) to grasp what it means to have the skill 'a' (which

stands for a memo that needs approving). The reduce-calculus

consists of the meeting of a problem and a solution for the same

task, so '-a' as a skill and 'a' in an issue can reduce each other, but

the same goes of course for 'a' as a skill and '-a' in an issue.

problems and solutions) could for example be conceptualized as the

influence of the preferences. The preference to 'approve',

strong

enough, puts strain on the decision making process, leaving other

tasks 'on hold'. In this way, skills could consist of solutions and

issues of problems, but the preference for a skill or action can create

problems at the decision making level.

3.

The communication in Double-AISS is no communication in the

ordinary sense. Normally, comrnunication consists of the flow of

information. In Double-AISS, the communication consists of the

flow of issues.

4.

The thinking-ahead in Double-AISS is one-sided. The actor who is

trying to find an alternative does all the thinking, including the

thinking-ahead, alone. This includes the thinking-ahead which

involves thinking what another actor would do if the alternative

being studied is indeed chosen. This seerns consistent with Mead's

role-playing theory, but

it

is not. In Double-AISS, the thinking like

another actor is implernented as being that other actor temporarily.

There is no model of the other actor used, so there is no distortion

of the view of the other actor possible. This is hard to rnaintain

if

we look at the real world, where perfect models do not exist.

5.

The authority relations in Double-AISS are promised by the

authors, but are not implemented. Instead, if an actor can reach

another actor, he can choose to move or attract whether the other

actor likes it or not. The only constraint is the availability of an

issue to move or attract.

seems to me that this means that

if

one

2.5.

Conclusions on Double-AISS

The drawbacks of Double-AISS are: the unclear notation used to represent

subtasks and skills, the "reduce" calculus used to implement the viewpoint

dependency of the garbage can theory as to whether something is a problem or

a solution, the lack of flow of information, the one-sidedness of both the

thinking-ahead procedure and the flow of subtasks. The pros of the

Double-AISS model include: the use of bounded rationality, the use of authority

relations, the coupling of skills and subtasks, the use of different problem

solving strategies (i.e. moving, attracting or reducing issues) and the use of

different commitment modes.

Chapter 3 Using Soar: Plural Soar

Soar (Laird, Newell and Rosenbloom 1987) is an implementation of the

human problem solving paradigm.

is a system capable of general intelligent

behavior and can perform the full range of cognitive tasks, employ the full

range of problem solving methods and learn about all aspects of the tasks and

its performance on them. Plural-Soar (Carley et al. 1991) is a model of a small

organization implemented in Soar. Here also, the core of the model is formed

by the actors (or agents as they are called by the authors). The organization

structure is flat, there are no hierarchical levels. Communication does exist,

consisting of the moving of a subtask to other actors. Every actor can perform

the whole task, i.e. all actors have the same skills. There are no strategy

alternatives for the actors. The name Plural-Soar is derived from the

contraction of plural agents and Soar.

3.1.

The model of an actor in Plural Soar

In Plural Soar, there are no strategy alternatives among which the actor has to

choose. The organization structure is flat, all actors are on the same

hierarchical level. Not all actors can communicate. Those who can, can pose

questions to all other actors. Only those who can listen, hear the question. The

original actor tries to find the answer himself at the same time as the others

try to answer his question, so one could say this particular problem is solved

in parallel mode by the organization. Once the answer is found, it is

communicated to the original actor.

4

move item to right stack

5

move item to conveyer belt

6

ask order

7

ask question

8

answer question

9

wait

The choice for one of these actions is dependent upon the state in which the

actor is. E.g., in order to move an item on the conveyer belt, the actor must be

immediately in front of a stack, the top item of the stack must be the item

which is ordered. After this action, the order is taken to be filled.

If

there can

be more than one action applied, preferences govern the choice in this

impasse.

The physical properties of the warehouse sometimes cause actors to wait

for their turn. Only one actor can look at a particular stack at the same time, so

if more than one actor is in front of a stack, the actors not immediately in

front of it have to wait for their turn.

3.2.

The search space of an actor in Plural Soar

The organization consists of (at most) five actors. Every actor is implemented

as a separate Soar program and is autonomous in his choice of action. Every

actor has an idea of the warehouse, he knows the number of stacks, their

location and the location of the conveyer belt. The search through the

problem space is only influenced by the behavior of other actors when they

have to wait or when there is a question answered or a question to be

answered. When an actor has filled his order, he goes to the order stack to get

a new one. Every item is unique, it is only once ordered and is only once

available in the warehouse. The warehouse itself contains ten item stacks each

containing 3 items and one order stack, as well as a conveyer belt in front of all

the stacks and a walkway between the conveyer belt and the stacks to move

through the warehouse.

3.3.

The behavior of Plural Soar

The simulation model (Carley et al. 1991) studied the influence of two

cognitive capabilities. The ability to memorize a stack, thereby extending the

model the actor has of the warehouse, and the ability to communicate. The

influence of the number of actors was also studied.

More actors did not always mean quicker results for the organization.

However, more actors does mean less working time per actor.

Communication also lowered the working time per actor, provided that there

were enough actors who could communicate. Waiting time increased when

the number of actors was increased, and this effect was reduced a bit by adding

communication and memory skills. Because communication was preferred

over search, communicating actors waited more than non-communicating

actors. A side-effect was noted, once an actor with communication skills took

an order, he first asked his question before moving away from the front of the

order stack, thereby obstructing the view on the order stack for the other

actors. Too much actors also reduced the number of answers given, because

the waiting time increased.

3.4.

Critique on Plural Soar

There is also some critique possible on Plural Soar, albeit of a somewhat

different kind than the critique on Double-AISS. These critical points include:

1

the cooperation in Plural Soar

2

the autonomy of the actors in Plural Soar

3

the unbounded rationality of the actors in Plural Soar

4

the implementation of the warehouse task

unrealistic. Every actor can perform the whole task

by

himself, the

only help sometimes asked (but not needed) is to help locate the

item. A more realistic model would be one where there is a flow of

subtasks; actors should not be perfect, their skills should be limited.

3.

Although humans are thought to be bounded rational (Simon

1955), in Plural Soar the rationality of the actors is unbounded.

Every alternative is evaluated, and the optimal alternative is

chosen. Because of the order in which the warehouse task has to be

performed, no optimal solutions are possible. The only

non-optimal choice would be to move away from an item instead of

closer to it. But this is also impossible, because the movement is

always directed towards an item. This follows from the fact that

searching for an item starts when the order is taken, and all item

stacks are in one row on the same side of the order stack. This is

illustrated in figure 6.

~ ~

---· §

order stack stack 1 stack 2 stack 1 O

actor X

conveyterbelt

Figure 6:

layout of the warehouse

walkway

The moving of subtasks to actors in another location than the order

stack can make non-optimal alternatives possible. The addition of

other strategies would also make non-optimal alternatives possible.

4.

In order to be able to move subtasks, these have to be identified and

the program has to be rewritten. In the Plural Soar model, some of

the actions are implicit. E.g.

if

one is immediately in front of the

stack, one sees all of its contents in one glance. Also,

if

one is

immediately in front of the stack in where one has located the

ordered item, one moves items around until the ordered item is on

top.

If

it is on top, it is immediately moved to the conveyer belt.

3.5.

Conclusions on Plural Soar

The drawbacks of Plural Soar are: the minimality of the communication, the

total autonomy of the actors and their skills, the unbounded rationality, the

absence of clear defined subtasks and skills and the lack of strategical choices.

The pros of plural Soar include: the use of clear notation for subtasks

(operators) and the use of a cognitively based view on actors.

Chapter 4 The merging of Double AISS and Plural

Soar: T ASCCS

The new model is intended to use the strong points of both Double-AISS and

Plural Soar and at the same time to overcome their weaknesses. The starting

point was the Plural Soar code, to which several additions were made. Skills

were added, with their corresponding subtasks and completion markers,

problem solving strategies were added and organizational structure was

added, implemented as a communication network. Evaluation of cooperation

requests depends upon the commitment of the receiving actor, another

addition to the Plural Soar model. The preference ordering for one of the

three strategies is set by the commitment of the actor. However, these

preferences are absolute and bounded rationality implemented as in

Double-AISS (using aspiration and tolerance level combined with an adaption period)

is not implemented in this model. Also there are no models of other actors

used to guide communication. All actors that can be reached by an actor are

alike to that actor. Currently, the evaluation of commands or requests is very

shallow, in future models other evaluation criteria should be developed. The

acronym TASCCS stands for: tasks, actors, structure, commitment,

communication and skills.

4.1.

The model of an actor in T ASCCS

In TASCCS, the problem solving is conducted by actors. Every actor is

implemented as a separate Soar program. The actors are characterized by the

skills they have, their place in the organizational hierarchy (implemented as a

communication network) and their commitment to the organization. A skill

is the ability to perform a (sub)task. The place in the organizational hierarchy

is implemented as the place in a communication network.

Two

communication modes are possible: command and request. Organizational

levels correspond to communication modes, communication within an

organizational level is implemented as making use of 'request' as the

communication mode. Communication from one level towards a level below

the actor is implemented as making use of 'command' as the communication

mode. Cooperation requests or commands to alevel in the organization above

the actor are not possible in the model. Every actor has a list of actors that can

be reached with requests and who can be reached with commands. The list can

of course be empty. The actor' s commitment is either altruistic or egoistic. The

commitment of an actor is decisive for the ordering of the strategy alternatives

and for the evaluation of received requests. Three strategy alternatives (or

problem solving strategies) exist: reduce (that is, solve) subtasks, move

subtasks to another actor or attract subtasks from another actor. Altruistic

actors look at the workload of the whole organization in order to decide which

problem solving strategy they prefer. I.e. they prefer the reduce strategy over

the move and attract strategy since these do not solve a subproblem. Egoistic

actors try to reduce their own workload, therefor they prefer the move strategy

over the reduce strategy.

The communication modes and problem solving strategies can be

combined to five different alternatives: reduce, a request to move, a command

to move, a request to attract and a command to attract. A request to move can

be read as: "Can you do (subtask) for me concerning item (item)?" and a

command to attract like "I want to do (subtask) for you!". In general,

commands are not negotiable, the only prerequisite for a command to be

obeyed is the possibility of the action, a command to attract to an actor without

the asked subtask would result in a negative answer. Requests on the other

hand have to be evaluated by the receiving actor. Here also the possession of

the subtask is a prerequisite, but even if the attraction of a subtask is possible,

the commitment of an actor decides the evaluation. Since altruistic actors are

concerned with the organization as a whole, they will answer yes to a move

request. Egoistic actors on the other hand will answer no, for otherwise their

workload would increase. Attract requests will be answered positively by the

egoistic actors because they are glad to get rid of a subtask and negatively by

by doing as much work as possible themselves. 'Command' is chosen when a

negative answer is not wanted, 'request' is chosen when a negative answer is

hoped for. Thus, an egoistic actor tries to move a subtask, at first using

'command' and if that' s not possible, using 'request' as the communication

mode, before trying to reduce it, and

if

that's not possible, to attract with the

hope of a negative answer and

if

that's not possible too, attract using the

'command' communication mode. Altruistic actors prefer to reduce, then to

attract an item, preferring 'command' over 'request' to increase the chance of

a successful attraction and then to move with a preference for a negative

answer, so 'request' is preferred over 'command'.

The content of 'move' and 'attract' is a subtask for a certain item.

When moving a subtask to another actor, the sending actor decides which

subtask concerning what item is moved. E.g. the actor sends the message "Can

you FIND item A for me?". When attracting the sending actor chooses the

subtask, but the receiving actor chooses the item. The message being send

looks like this "Can I FIND an item for you?".

The warehouse task is split into different subtasks: take order and fill

order. The filling of the order is again split into subtasks. These are the

subtasks that can be communicated and which concern the skills of the actors.

They are: finding an item in the warehouse, getting the item from the stack

and putting the item on the conveyer belt. In the current model, the skills of

the actor are not ordered by preferences, so no skill is preferred over other

skills. Subtasks can be handled in two ways: subtasks can be reduced or

moved. The attraction of a subtask is an alternative for the taking of an order.

take order

reduce subtask

warehouse task

attract subtask

move subtask

In order to implement this new model of the warehouse task, looking at a

stack was made explicit, as well as putting an item on the conveyer belt.

Finding an item consists of moving through the warehouse and examining

the stacks. When an item is found, it's location is remembered. Getting hold

of an item consists of going to the item location, manipulating the stack and

taking the item from the stack. This is signalled by the completion marker 'in

possession'. An item can be put on the conveyer belt by an actor when it is in

possession of that actor. The item location has to be communicated when the

'get' subtask is moved to or attracted from another actor. Also, once the

subtask 'get' is reduced by another actor, the item has to be handed over to the

original subtask owner. The same goes for the moving or attracting of the

'put' subtask. Once again, communication is needed.

4.2. The search space of an actor in TASCCS

The search space of an actor in TASCCS consists of several layers. The top

layer is the main decision cycle, illustrated in figure 8.

Figure 8:

continu search through problem space

take order

wait

the main cycle of the new model

end

First, the actor tries to obtain an order. This is the first action of every actor.

Then, the actor tries to decide on whether to reduce the first subtask or to

move it. This of course depends on his commitment. When reduce is chosen,

the skill to perform that subtask is needed in order to be able complete the

reduction of the subtask, see figure 9 a, b and c.

move to next stack yes remember item location end

Figure 9a:

the find problem space

yes

yes take the item from the stack

end

Figure 9b:

goto the item location

move the top item to another stack

yes

put the item on the conveyer belt and remove it from the working on list

end

no

no

end

goto the location where the item owner is and receive the item

Figure 9c:

the put problem space

When move is chosen, the actor to move the subtask to has to be decided on.

This is illustrated in figure 10.

whether to try to attract a subtask from another actor or take an order.

An actor receiving a subtask move or attract message has to evaluate

that message (see figure 11).

end

Figure 11:

the problem space for communication receival

Once the actor agrees on the move or attract, dependent on the subtask

involved the item location has to be exchanged or the item has to be handed

whether to try to attract a subtask from another actor or take an order.

An actor receiving a subtask move or attract message has to evaluate

that message (see figure 11).

end end answer end evaluate request send answer yes send no as an answer end

over. After this, the actor receiving the subtask has to go through the same

cycle himself. Shall I reduce it or move it?

If

an actor has reduced a received

subtask, the necessary information has to be fed back to the actor who send the

subtask.

4.3. The behavior of TASCCS

Due to limitations in the computer facilities and the time needed to finish the

model, the organization consisted of only two actors, with one actor

hierarchial above the other actor in the organization. The top level actor is

egoistic and the bottom level actor altruistic. This was chosen because the

attract problem space could not be finished in time. In this particular

organization, attract is not an alternative which is viable. Using this

organization in TASCCS, the influence of communication on the problem

solving process of the organization was studied. The dependent variables

studied are the amount of cycles waited by an actor, the times an actor moves

to another stack and the number of items that is moved to another stack. The

outcomes of the new model are compared to the results of the old Plural Soar

model. The results will be given in full detail in appendix 2 and discussed in

chapter 5, the conclusions will be drawn in chapter 6.

Chapter 5

H.esults of

TASCCS

The results of TASCCS are given in full detail in appendices 2.1 and and are

summarized in table 1.

model using communication model without communication

actor 1 actor 2

actor 1 actor 2

I

actor moves

0

158

158

88

70

158

item moves

0

50

50

26

19

45

waited cydes 1621

577

2198

10

62

72

Table 1:

summary of the results of TASCCS

The results for the same model using Plural Soar are given in full detail in

appendices 2.3 and 2.4 and are summarized in table 2.

model with communication model without communication

actor 1 actor 2

I

actor 1 actor 2

I

actor moves

746

82

158

79

81

160

item moves

20

15

35

17

19

36

waited cydes

18

29

5

32

37

Table 2:

summary of the results of Plural Soar

It

is dear that the difference between the organization modelled here with

cooperation and communication and that same organization without

warehouse (i.e., only one actor can look at a stack at the same time). In Plural

Soar, only the physical properties of the warehouse cause actors to wait.

The number of moves an actor makes are not influenced by the

communication and cooperation variation (that is, of the organization as a

whole, for the individual actors there is of course a difference). The only

deviation is the Plural Soar model without communication, where the

number of actor moves is slightly higher than in all other simulations.

The increase of the waiting time in the TASCCS model with

communication is in accordance with the theory that well-structured

problems need labor specialization in order to make cooperation useful. The

cooperation makes the actors dependant on each other, they have to wait for

the other actor to perform a subtask or answer to communication.

If

labor

specialization was implemented (e.g. if an actor did not have all skills or

prefer some subtask over another), the problem solving process could have

been less tedious. However, that also calls for another approach of the

communication of subtasks and the taking of orders. Subproblems will have

to be processed in parallel mode, so that the moving of the find subtask

permits an actor to take another order, and also that an actor can work on

several find subtasks at the same time. A different implementation of the

actions following the completion of the get subtask and the moving of the put

subtask could also speed up the problem solving process. Not using the

original subtask owner as an intermediary item owner but only as a

communication manager could be useful. I.e. after the completion of the get

subtask, the item does not have to be handed over to the original subtask

owner. Instead, the completion can be communicated, after which the

communication manager (i.e. the original subtask owner) can send a message

as to who the item should be handed over to (this is the actor that accepts the

moving of the put subtask). Another important improvement can be made.

Instead of letting the actors decide when they take an order, orders can be

distributed by an order manager, like the inflow of items in Double-AISS. In

this way, the attract strategy is not an alternative for the taking of an order, but

one of the three possible action strategies.

Chapter 6

Conclusions

The new model was designed to overcome certain shortcomings of its

predecessors Plural Soar and Double AISS. The mixing of these models,

together with additions of my own, has proven to be quite time-consuming

but seems to work now. The preliminary results are in accordance with

theoretical predictions about the efficiency of cooperation when solving

well-structured problems. To make full use of the benefits of the new model, much

more work is necessary. Future work should focus on the implementation of

the attract strategy, the implementation of bounded rationality, the use of

revisable belief models of other actors to serve as a guideline for

communication and better evaluation of communication (e.g. based on skills,

workload and preferences instead of only based on commitment). Labor

specialization should be added and subtask reduction in parallel mode to

make the advantages of communication and cooperation more clear. Less

physical movement and more use of communication will make the problem

solving of the organization more effective, both from the viewpoint of

physical resources used by actors and a better use of the possibilities

communication offers an organization. This can be accomplished by allowing

the subtask reduction in parallel mode and by replacing the handing over of

items after a moved or attracted "get" subtask to the original subtask owner by

the handing over of the item to the actor who reduces the "put" subtask.

Some bugs have to be repaired also. For instance, some file updates are

not clone correctly. The working-on file is not updated, but overwritten, and

so it seems that only one item is being worked on. In the model without

communication, the actors work independent and most of the time, two items

"know" what the solution is. Learning concerning a task transforms beginners

into experts. Analogous to this, one can say that routine decisions are

decisions made by an expert on that task, and that nonroutine decisions are

decisions that beginners, in that specific task, have to make. As experts on a

task

can

predict all consequences and choose the best alternative, through

their experience, it is questionable if their rationality, concerning that task, is

indeed bounded. The distinction between standardized and nonstandardized

work can be used as a basis to judge

if

bounded rationality is applicable to the

task and the individuals working on that task and consequently,

if

cooperation is useful.

If

the effect of cooperation and bounded rationality are

studied, it is therefor necessary that the individual actors do not have all skills

and have to find out how subproblems can be solved, i.e. the task must not be

routine to them. Learning can replace the cooperation in the long run, but

learning capabilities are also limited and cooperation will be necesarry to solve

problems that are to complex for an actor to solve by himselve.

will be

interesting to see

if

the learning capabilities of Soar can indeed create such an

effect.

Appendix 1 What is Soar?

Soar is the computer equivalent of the human problem solving paradigm.

Using Soar, one can imitate the behavior of humans solving a particular

problem. The underlying structure and the use of operators are all taken from

the theory as posed in (Newell and Simon 1972). Soar, which is derived from

the cycle of taking a state, applying an operator and generating a result,

searches in problem spaces using operators to change the state, problem space

or preferences. Problem spaces, operators and states are generated and selected

in order to pursue the goals of the system. In order to find correct and efficient

paths from the initial state of the system to the desired state, knowledge is

needed. This knowledge can be of two forms: directly available as operators or

indirectly available through problem resolution. The operators are stored in

long-term memory. Problems occur when either more than one or less than

one decisions can be made (e.g. two operators can be applied, among which the

system then has to choose). The problem resolution becomes the new goal of

the system, a technique called 'subgoaling'. When the problem is solved, the

knowledge used to solve the subproblem can be stored as a learned production

in long-term memory (this is called 'chunking').

Soar's decision cycle consists of two parts.

first tries to elaborate the

current situation with relevant information retrieved from the long-term

memory. When all fireable operators have been found, the decision procedure

starts. Preferences for operators are processed, and in the end either an

impasse is reached or the selected production is fired. The impasse creates a

subgoal (as described above). The result of this subgoaling can be stored as a

learned operator, with the objects in working memory that caused the impasse

as the antecedent and the results of the search in the subgoal problem space in

the consequent of the new operator.

Appendix 2 The results of the simulations

2.1. Simulation results using the TASCCS with organizational

structure and communication

Agent X (the commanding agent):

Agent movements:

0

Item movements:

0

Orders taken:

15

Cycles waited:

1621

Agent Y (the agent who is being commanded):

Agent movements:

158

Item IJ.'lovements:

50

Questions answered: 0

Cycles waited:

577

2.2. Simulation results using T ASCCS without organizational

structure and communication

Agent X:

Agent movements:

88

Item movements:

26

Orders taken:

8

Cycles waited:

Agent Y:

Agent rnovernents:

Item rnovements:

19

Orders taken:

Cycles waited:

62

2.3. Simulation results using Plural Soar with communication