You are on page 1of 10

AGENTS AND THEIR KNOWLEDGE REPRESENTATIONS

Dr S.Srinivasan, Mr Dheeraj Kumar , Mr Vivek Jaglan


HOD Computer Applications ,PDM Bhadurgarh , India
dss_dce@yahoo.com
Assistant Professor , PDM Bhadurgarh , India
Dheerajkumar47@gmail.com
Assistant Professor , PDM Bhadurgarh ,,India.
jaglanvivek@gmail.com

ABSTRACT
The objective of this paper is to study how Memory , Knowledge and Knowledge
representation are handled by Agent architecture . Twelve architectures have been used
for this preliminary analysis representing a wide range of current architectures in
artificial intelligence (AI). The aim of the paper is to understand the various Knowledge
Representations that agents should adapt generally . Also because of the design of the
architecture of the agents that we have taken in to consideration , the representations
vary from one agent to another .

Keywords : Episodic , Truth Maintenance system , frames , STRIPS .

1 INTRODUCTION Subsumption Architecture , ATLANTIS , Theo ,


Prodigy , ICARUS , Adaptive , Intelligent
An agent is anything that can be viewed as Systems (AIS) , A Meta-reasoning Architecture
perceiving its environment through sensors and for 'X' (MAX) , Homer , Soar , Teton , RALPH-
acting upon that environment through MEA , Entropy Reduction Engine are twelve
actuator[10]. A Vacuum cleaner agent is a reactive architectures we have considered for discussion on
agent which means it responds when it senses dirt the issues of Organization , memory , knowledge ,
on the floor . Where as a Taxi Driver agent should and knowledge representation . The features of 12
behave in a different way since its environment is architectures are given in a Table 2 .
entirely different . It is a Goal directed agent .[1] it
has to keep all the percepts it receives in its 2 ORGANIZATION
knowledge base . The architectures of these two
typical examples should certainly differ greatly . An organization provides a frame work for agent
Agent properties identify and entail the techniques interaction through the definition of roles ,
and methods that were used to realize a particular behavior expectations and authority relations .
architecture or architectural component. For Organizations are in general conceptualized in
example, most architectures include some sort of terms of their structure that is the pattern of
memory[4]. An agent is said to be a knowledge- information and control relations that exit among
level system when it rationally brings to bear all agents and the distribution of problem solving
its knowledge onto every problem it attempts to capabilities among them .
solve. Thus, knowledge is the medium of
transaction at the knowledge level and the
2.2 Hierarchical Organization
behavioral law is the principle of maximum
Intelligent systems may be organized into
rationality . Agent properties characterize the
hierarchies, or levels, which correspond to
memory: Is the memory declarative, procedural,
different capabilities. Capabilities of the "lower"
episodic? Are there size limitations? Is memory levels of organization are in a sense inherited by
uniformly accessed? Is it uniformly organized? the levels above, providing both a means of
These properties have been discussed in this
modularizing individual capabilities, but also the
paper
ability to reuse methods in different branches of THEO is frame based architecture .For an
the hierarchy. architecture to be frame-based means that it is
The authority for decision making and organized around the representation of all of its
control in concentrated in a single problem solver knowledge in frames. This is important because it
at each level of hierarchy . Interaction is through allows a uniform representation of knowledge. In
vertical communication from superior to sub- a frame based system, all knowledge is
ordinate agent and vice-versa . [3,4] represented as slots of some concept which
2.2 Modular Organization describe properties of that concept
A modular organization is one in which All knowledge in PRODIGY including
different functional components are separated control rules, domain knowledge, learned
from one another, a technique adopted from knowledge is represented in the PRODIGY
software engineering. This is in contrast to a description language (PDL). PDL is based on
composite organization in which there is no first-order predicate calculus and includes
separation between functions. Modular variables, disjunction and conjunction, universal
organization is also distinct from hierarchical and existential quantification, and conditional
organization. Modular organization is chiefly representation.
concerned with the horizontal design of a system In Icarus, the Concept tree in Labyrinth stores
whereas hierarchical organization involves a all knowledge. The other components of the
consideration of the vertical nature of the design. architecture have access to this knowledge
Thus, each level in a hierarchical system may be explicitly. For example, Daedalus retrieves
sub-divided into functionally-distinct modules. explicit plans from Labyrinth. The use of
declarative representations allows each module to
3 KNOWLEDGE REPRESENTATION reason about the available knowledge . [8]
The symbolic world model or global database
Knowledge means intellectual acquaintance with in AIS is formed from a uniform and declarative
perception of fact or truth. Representation is a based conceptual network. Within this
way of describing certain fragments or representation, sub-networks can represent
information so that any reasoning system can actions, events, and control plans through
easily adopt it for inferencing purposes. symbolic abstraction. The generality of the
Knowledge Representation system should conceptual network representation permits
provide way of representing complex knowledge encoding complete knowledge from different
and should possess the following characteristics: problem domains, enhancing multi-faceted
1. The representation scheme should have a set expertise. The network-type representation
of well defined syntax and semantics. This facilitates explanation as well .
will help in representing various kinds of All knowledge in MAX agents exists in
knowledge . nested logic frames including the rules and
2. The knowledge representation scheme should operators. Separating rules and operators
have a good expressive capacity . A good integrates a domain theory, in which rules specify
expressive capacity will catalyze the actions and actions satisfy rules. With a
inferencing mechanism in its reasoning production memory architecture knowledge is
process. brought to bear on a problem by executing some
The philosophy behind Subsumption Architecture operator that uses that knowledge, i.e. it adds
is that the world should be its own model. symbols and it deletes symbols in some logical or
According to Brooks, storing models of the world rational fashion
is dangerous in dynamic, unpredictable Homer uses both Black-box representation
environments because representations might be and heterogeneous representation .SOAR uses
incorrect or outdated. What is needed is the ability Black-box representation .Teton , RALPH-MEA
to react quickly to the present. Thus, sensor and Entropy Reduction Engine architectures use
readings should be mapped quickly and directly to both Declarative and Procedural representation .
actuator commands in a decentralized fashion. 3.1 Symbolic World Model
Agents designed according to Subsumption For some of the architectures under consideration ,
Architecture are generally non-symbolic. They there (usually) exists a level that enables the
have no global representation, and are system to store knowledge in some basic
decentralized. framework. This framework is based on symbols
In ATALANTIS architecture the control which serve to represent relations between the
layer uses procedural knowledge while the agent and its environment, and hence knowledge,
deliberative layer uses declarative knowledge. within the system. Such symbolic abstraction
occurs at the symbol level and derives its power circuituitous. Some knowledge is shared amongst
from the physical symbol system hypothesis. all execution architectures, such as possible states
Thus, symbolic abstraction enables, or at least and actions. However, each architecture also
facilitate, many architecture capabilities including requires its own specific knowledge. This
planning and learning. Soar's productions use knowledge includes utility values of actions and
symbols to store information . Such systems have states, which are not shared amongst architectures
memory ,symbols ,operations ,interpretation and in part because each architecture needs a different
capacities to operate on symbols. [4] type of utility value. Each execution architecture
operates independently and in parallel with the
3.2 Size Of The Knowledge Base others, with no interaction of the knowledge of
There are physical limits to the size of the different architectures occurring. Each individual
knowledge base that may be supported by an execution architecture does not have its own meta-
individual agent. But, in addition to the physical knowledge in regards to what it knows. RALPH-
limit on memory , there may be limitations MEA uses this technique .
imposed by the agent's architecture and style of
control. For example, Homer experiences a 3.5 Declarative Representation
processing slow-down as its episodic knowledge A declarative representation declares every
base increases in size. Since productions can be piece of knowledge and permits the reasoning
easily added to the knowledge base or learned in system to use rules of inference to come out with
Soar, the knowledge base may grow incrementally a new piece of information .A classic example of a
to cover a wider range of behaviors. The ultimate declarative representation is logic. The primary
size of the knowledge base is dictated only by the advantages of declarative knowledge is the ability
desired functionality; no absolute limits are made to use knowledge in ways that the system designer
by the architecture or the representation. did not foresee. Often times, whether knowledge
is viewed as declarative or procedural is not an
intrinsic property of the knowledge base, but is a
3.3 Glass Box Approach function of what is allowed to read from it.
Production systems, for example, are declarative if
productions may view themselves, and are
Glass box knowledge representation may be
procedural it they cannot. (cf. glass-box/black-box
defined as the ability of rules to examine each
other. Glass box representation is useful for control knowledge).
modular architectures, so that all modules have A particular architecture may use both
declarative and procedural knowledge at different
access to all knowledge. Glass box representation
times, taking advantage of their different
allows the rules and the architecture to share
advantages. The architectures that use Declarative
responsibility to examine, activate and rewards
representation are ERE , ICARUS , MAX ,
other rules. Also, learning is facilitated by glass
box representation because learning modules may Prodigy and THEO . ERE uses Uniform
write. Related properties are uniform access to conceptual graph to store information . Homer ,
ICARUS , MAX and Theo use Frames . The
knowledge and homogeneous representation .
Prodigy use Prodigy description language .
Knowledge in Prodigy is stored in a global
3.7 Procedural Representation
database which is freely available to all other
Architectures with procedural representations
modules. Knowledge is represented uniformly to
simplify access by other modules. The declarative encode how to do some task. In other words,
procedural knowledge is skill knowledge. One
representation of knowledge allows the
advantage of procedural representations is
architecture to analyze its own actions and
possibly faster usage in a performance system.
decisions.
Productions are a common means of representing
procedural knowledge.
3.4 Black Box Approach Use of procedural knowledge in an agent
Black box knowledge representation may be raises the questions of whether the agent can
defined as the inability of rules to examine other "know what it knows" . That the agent can
rules. Architectures with this property are limited demonstrate that it "knows what it knows" is
in their direct inferencing ability to see what other illustrated in a Soar system which includes the
rules are up to. Such representations are less ability to explain its actions. Procedural
fragile to changes and more modular and easier to representation is where knowledge is intrinsically
understand. bound up in the routines and procedures which use
Black-box knowledge representation does not rule it . These procedures and routines know how to do
out meta-reasoning, but it would make it more
a particular task which would be regarded as false statements) may be derived. Statements that
intelligent . can't be derived from the basis set are assumed
3.8 Global Representation and Uniform Access to false (or true). This is the closed-world
Knowledge assumption. The primary advantage of knowledge
In architectures with global knowledge, consistency is ability to store less statements.
different modules may read and/or write to An architecture that tolerates knowledge
common database(s). Often this knowledge is inconsistency generally treats its knowledge base
used to representing a world-view of what the as a set of competing hypotheses, or as a set of
agent believes is true in its environment. An statements that it has varying amounts of
advantage of having global knowledge is that confidence in. Often there is a numerical measure
different modules may share their data and of belief. The flexibility in representation and
abilities for more intelligent combined behavior. flexibility in learning and reasoning are some of
This makes modular architectures more effective. the advantages .
Also, such knowledge is necessary for Prodigy is a consistent knowledge
representing world-views: without it an architecture while MAX is a inconsistent
architecture may only react to its present sensor knowledge architecture .
readings (e.g. Subsumption Architecture.)[9]. 3.10 Homogeneous (Uniform) Knowledge
A related property is knowledge Representation
homogeneity, which is a measure of how similar Architectures with global access to
knowledge is represented by the architecture. knowledge may store it in a uniform format in a
Homer's memory modules contains state central database, or may have it in a non-uniform
information which is globally available. The format in a distributed fashion. The uniform
planning module retrieves goals from the generic method is often employed in systems with a
memory. The interpreter and perception modules general knowledge representation scheme like
enter events into memory. Although the world frames or first order predicate calculus. The non-
state is uniformly accessible, other types of uniform method is often used in loosely-coupled
knowledge in Homer is restricted architectures for storing special knowledge not for
3.8 Efficiency of Knowledge Access general use . Access all knowledge easily ,
As the knowledge bases of these architectures interfaces not required to be changed by addition
grow increasingly large, an efficient way of or change of modules and easy to modify module
accessing specific knowledge in the knowledge to accommodate new knowledge are the some of
bases becomes increasingly important. While this the advantages of knowledge uniformity . The
is an issue for almost all architectures, MAX has design limitation , inefficiency to work on a few
good and bad aspects for efficient knowledge modules and wastage of memory are some of the
access. The good aspects are that all knowledge is advantages .
homogeneously represented and uniformly In Prodigy, all types of rules- inference rules,
accessed using frames. The bad aspects are that all operators, and control rules, are represented
knowledge is declaratively stored.. In the case of uniformly. This allows the control rules to apply
SOAR , with a very large knowledge base to both inferences and operators, guiding the
containing perhaps a million or more productions, application of rules. This representation allows
accessing the knowledge in an efficient manner is Prodigy to glass box approach.
an important issue for the architecture. During the
Elaboration Phase in Soar, all the productions in 3.11 Heterogeneous Knowledge Representation
the memory that apply are fired in parallel. This The non-uniform method is often used in
parallel operation allows the architecture to access loosely-coupled architectures for storing speciality
all the relevant knowledge in the long-term knowledge not for general use. A heterogeneous
memory in an efficient, tractable way. knowledge representation is based on associating
3.9 Knowledge Consistency a finite automaton with first order logic. A number
Knowledge consistency is the property that a of knowledge representation problems raised by
knowledge database contain no contradictions. It the electromyography test features are examined
is extremely important for knowledge in this study and the expert system architecture
representations that may only either assert or deny allowing such a knowledge modeling are laid out.
statements, with no measure of partial belief. One [6] .Entropy Reduction Engine uses
such system is first order predicate calculus. Heterogeneous Knowledge Representation .
Because all statements may be either true or false, 3.12 No Explicit Representation
it may be possible to store only part of the Usually the representation of knowledge
statements , from which all true statements (or all takes some form. Examples include: first-order
predicate logic, frames, networks, scripts, etc. database (c.f. Prodigy). Unfortunately, this
However, knowledge does not have to be flexibility has limitations. First-order predicate
represented explicitly. The subsumption and logic is composed of statements that are assumed
ATLANTIS agents rely completely on the world to be true. The statements are composed of atoms
for information. There is no explicit knowledge ,predicates, two substatements joined by a
representation -- no global representation of state conjunction, disjunction, or implication, a negated
nor any underlying symbol level. "The world is its substatement, and a statement with an existential
own best model"[7]. or universal quantifier (in this case, atoms in the
statements can be replaced by variables in the
4 SPECIFIC EXAMPLES OF quantifier). [1]
REPRESENTATIONS AND MEMORY
STRUCTURES 4.4 Strips-like Operator Representation
STRIPS, or the Stanford Research Institute
4.1 Associative Memory Problem Solver, was proposed by Fikes and
Associative memory refers to a memory Nilsson in 1971[1] and included a representation
for operators that was intended to solve the frame
organization in which the memory is accessed by
problem. STRIPS uses well-formed formulas of
its content . Thus, reference clues are "associated"
the first-order predicate calculus and specifies
with actual memory contents until a desirable
operators by a precondition list, an add-list and a
match is found. Production systems are obvious
examples of systems that employ such a memory. delete-list. The preconditions must be satisfied by
Associative memory stands as the most likely the current state before an operator is applied. The
effects of the operator are given by the add and
model for cognitive memories, as well. Humans
delete lists. The add-list adds new, instantiated
retrieve information best when it can be linked to
well-formed formulas (or wffs, logical descriptions
other related information. This linking is fast,
direct and labyrinthian in the sense that the of the world) to the current state. The delete-list
memory map is many-to-many and homomorphic removes wffs from the current state.
Although STRIPS did resolve some of the
[8].
issues related to the frame problem, it (and all
Associative Memory is a primary property of the
systems that use a STRIPS-like representation)
Soar .
suffers from a requirement for explicitness -- all
4.2 Episodic Knowledge
Two particular types of knowledge -- actions (including secondary effects) must be
procedural and declarative -- have been used included in the model of the operator. In complex
worlds, this is often impossible [10].
extensively in the design and development of
4.5 Frame-Like Representation
cognitive architectures. However, neither of these
Marvin Minsky in the book on computer
types of knowledge characterize the knowledge
vision proposed frames as means of representing
that humans use to remember events. Such
remembrances are called episodic knowledge. common-sense knowledge. Here knowledge is
Since this knowledge is, by definition, organized in to small packets called frames. The
contents of the frame are certain slots which have
experiential, it must learned by the agent rather
values.
than pre-encoded (of course, it is possible to
To example the concept of frames, try to
conceive of episodic knowledge being pre-
recall the structure of our computer centre.
encoded but, once running, additions to the
episodic memory would then be learned). This Various equipments like computer system, dump
terminals, printers, air-conditions, un-interrupted
knowledge can confer the capability to perform
power supplies etc. are placed at some position in
protracted tasks or to answer queries about
the computer centre. By this time we have formed
temporal relationships and to utilize temporal
an image of computer centre in mind and it is
relationships.
Episodic knowledge is a key feature of the Homer easy for us to identify without going there where
architecture. things are kept. Suppose we visit a new computer
center. Our mind now considers a series of
4.3 First-order Logic Representation
permutation and combinations . The frame of the
Many of the architectures analyzed build
computer centre in mind is oriented towards the
upon a substrate of First Order Predicate Calculus.
present situation with some slots being added (for
This is a very descriptive declarative
representation with a well founded method of new objects) and some slots being removed. For
deriving new knowledge from a database. Its every equipment /object, we form a frame.
A frame can be defined as a data structure that
flexibility makes it a good choice when more than
has slots for various objects and a collection of
one module may add to or utilize a common
frames consists of various objects and a collection Chaining
of frames consists of expectations for a give
situation. A frame structure provides facilities for Impasse-
describing objects, facts about situations, driven
procedures in what to do when a situation is Control Y Y
encountered. Because of there facilities a fames Serial
provides, frames are used to represent the two Processing Y Y
types of knowledge, namely declarative factual Parrellel
and procedural. Like a semantic network, one of Processing Y Y
the chief properties of frames is that they provide Asynchrono
a natural structure for inheritance. ISA Links us
connect classes to larger parent classes and Processing Y Y
properties of the subclasses may be determined at Interruptible
both the level of the class itself and from parent Processing Y Y
classes.
Open-Loop
In addition, the values of a particular attribute
processing Y
need not necessarily be filled with a value but may
Closed -
also indicate a procedure to run to obtain a value.
Loop
This is known as an attached procedure. Attached
Processing Y
procedures are especially useful when there is a
Hierarchical
high cost associated with computing a particular
Organizatio
value, when the value changes with time or when
n Y Y Y Y
the expected access frequency is low. Instead of
computing the value for each instance, the values Modular
are computed only when needed. However, this Organizatio
computation is run during execution (rather than n Y Y Y Y
during the establishment of the frame network) Symbolic
and may be costly. World
THEO , Homer , Meta Reasoning Model Y Y Y Y Y
Architecture extensively use frames for Size of the
representation . Knowledge -
Base Y Y
In the Table1A and Table 1 B , Rows indicate the Glass Box
Memory , Knowledge and knowledge approach Y Y
representation and the column indicates the Black Box
particular architecture . Y in the cell means that Approach Y Y
the architecture corresponding to column has the Declarative
type of memory or knowledge and knowledge Representat
representation . ion Y Y Y
Procedural
Table 1A Representat
ion
Global
Representat
ion Y Y Y Y
ADAPTIVE INTELLIGENT

Uniform
Memory , Knowledge &

Access to
Knowledge Y Y
Knowledge
Representation

Consistency Y Y
Subsumption

Homogenou
ATLANTIS

SYSTEMS
PRODIGY

s ( Uniform)
ICARUS

Knowledge
THEO

Representat
ion Y Y Y Y
Forward &
Back-ward Y Y
Heterogene Asynchronous
ous Processing Y
Knowledge Interruptible
Representat Processing Y Y
ion Open-Loop
No-Explicit processing Y
Representat
Closed -Loop
ion Y
Processing
Associate
Hierarchical
Memory
Organization Y Y
Episodic
Modular
Knowledge
Organization Y Y Y Y
Meta-
Knowledge Symbolic
Y
World Model Y Y Y Y Y
First-Order
Logic Size of the
Knowledge -
Representat
Base Y
ion Y
Glass Box
STRIPS-
approach Y Y
LIKE
OPERATO Black Box
R Approach Y Y Y
REPRESEN Declarative
TATION Y Representatio
Frame - n Y Y Y
Like Procedural
Representat Representatio
ion Y n Y Y
Network Global
representati Representatio
on Y Y n Y Y Y Y Y Y
Uniform
Access to
Knowledge Y Y
Table 1B Knowledge
Consistency Y
Homogenous
ENTROPY REDUCTION

( Uniform)
Memory , Knowledge &

Knowledge
META REASONING

Representatio
ARCHITECTURE

n Y
Representation

RALPH-MEA

Hetrogenous
Knowledge
ENGINE
HOMER

Representatio
TETON
SOAR

n Y Y Y
No-Explicit
Representatio
Forward & n
Back-ward Associate
Chaining Y Memory Y
Impass Episodic
e-driven Knowledge Y
Control Y Y Y
Meta-
Serial Knowledge Y Y Y Y
Processing Y Y Y Y First-Order
Parallel Logic
Processing Y Y Y Representation
STRIPS-LIKE expected situation , decides
OPERATOR whether to and how to
REPRESENT respond .
ATION Y Y --- focus attention on the
Frame - Like most critical aspects of
Representatio current situation .
n Y Y --- operating continuously
without rebooting .
Network
--- able to coordinate with
representation
external agent .
Table 2
(more or less similar to
human being )
Sl Architect Features
7 Meta Many ideas in MAX may
N ure
Reasonin traced to Prodigy.
o
g (MAX) --- rule-based forward –
1 Subsumpt Complicated Intelligent chaining engine that
ion behavior operates on productions .
Into --- is designed to support to
Simple behavior modules modular agents.
Organized into --- they are used to
layers respond to a dynamic
environment in a timely
2 THEO Plan-Then-Complie manner .
By this --- modules are categorized
Integrating learning , in to Behavior and monitor
planning and knowledge .
representation
3 ICARUS Specific representation of --- Some of the modules
long term memory .It uses are:
3 independent 1.attention focusing
asynchronous modules 2.multiple problem solving
responsible for strategies
1.Perception 3.execution monitoring
2.Planning 4.goal-directed exploration
3.Effecting 5.explantion-based learning
4 PRODIG Storing the knowledge in a 6.process interruption
Y form of first order predicate 7.intelligent resumption
logic (FOPL) called 8 HOMER --- Is not designed for
Prodigy Description general intelligence .
language (PDL) . --- underlying philosophy is
Has a modular architecture to synthesize several key
that stores the knowledge areas of AI to form one
symbolically. complete system . (like
5 ATLANT Integrating planning and planning, learning , natural
IS reacting in a heterogeneous language understanding ,
asynchronous architecture robotic navigation ) .
for mobile agents . HOMER answers questions
It consists of 3 layers: posed by users and carries
1.Control layer out instructions given by
2.sequencing layer users .
3.deliberative layer --- is a modular structure.
6 Adaptive To reason about and It consists of :
Intelligen interact with other dynamic 1.memory
t System entities in real time . 2.planner
(AIS) --- problem solving 3.natural language
techniques interpreter and generator .
--- when encountering un- 4.reflective processes
5.plan executer 12 Entropy --- focuses on problems
Reductio that require planning ,
9 SOAR --- originally known as n Engine scheduling and control
STATE OPERATOR AND (ERE) --- uses many different
RESULT . problem solving methods
--- main goal is that full such as :
range of capabilities to be 1.problem reduction
handled by an intelligent 2.temporal projection
agent from highly routines 3.rule-based execution
to extremely difficult open-
problems
--- the underlying SOAR 5 CONCLUSION
architecture is the view that
symbolic system is
necessary and sufficient The problem of AI is to describe and build agents
condition for general that receive percepts from the environment and
intelligence . This is known perform actions, and each such agent is
as Physical Symbolic implemented by a function that maps percepts to
system Hypothesis (PSSH) actions. It explains the role of learning as
--- ultimate aim is to get extending the reach of the designer into unknown
general intelligent agent environments, and shows how it constrains agent
--- is based on a production design, favoring explicit knowledge representation
system ie. It uses explicit and reasoning . It analyzes basic techniques for
production rules to govern addressing complexity . The idea is to Integrate
its behaviors . state-of-the art AI techniques into intelligent agent
designs, using examples from twelve agents to
full knowledge-based agents with natural
10 Teton --- is a problem solver language capabilities and so on . This leads to the
--- uses two memory areas study of Multi-Agent systems and its applications
1.Short-Term . In depth analysis of various Agent architectures
memory and their capabilities is to build a Multi Agent
2. Long-Term System that will be suitable for our future work on
memory Supply Chain Management
--- like human beings ,
interruption are allowed . REFERENCE
--- it has a feature called [1] Elavine Rich , Kevin Knight and Shivshankar
Execution Cycle which , Artificial Intelligence by The Mcgraw Hill
always look for what to do Publishing Company limited .
next . [2] Kurt Vanlehn , Rule-Learning Events in the
11 RALPH- --- is a multiple execution acquisition of a complex skill ,
MEA architecture Journal article by, Journal of the Learning Science
--- like human being , ,vol 8 , 1999
selecting best one from the [3] Philip E. Agre , Hierarchichy and History in
environment Simon’s “Architecture of Complexity “ Journal of
--- RALPH – MEA uses Learning Science 12(3) , 2003
Execution Architecture [4] Newell a , Unified theories of Cognition ,
(EA) to select from one Harvad University Press , Cambridge ,
state to best one . Massachusetts , 1990
--- it uses the following : [5] Levesque H , Brachman R , A fundamental
1.Condition action trade off in Knowledge representation and
2.Action utility reasoning , 1998
3.Goal – based [6] Vincent Rialle , Annick Vila , Yves Besnard ,
4.Decision Theoretic Artificial Intelligence in Medicine 1991
[7] Vincent C.Muller , Is there a future for AI
without representation , springer Verlag , 2007
[8] Pat Langley , Dongkyu Choi , Seth Rogers ,
Interleaving Learning Laboratory Centre for the
study and languages and information , Stanford
University , Stanford .
[9] A.S Maida , A Uniform architecture for rule-
based meta-reasoning and representation :- case
study . , IEEE Computer Society Press.
[10] Artificial Intelligence: Modern Approach by
Stuart J. Russell , Peter Norvig Prentice Hall
Series in Artificial Intelligence.

You might also like