Instructable Agents:

1
IT/CS 803, Spring 2002
Instructable Agents Introduction
George Tecuci tecuci_at_cs.gmu.eduhttp//lalab.gmu.
edu/
Learning Agents LaboratoryDepartment of
Computer Science George Mason University
2
Overview
Course objectives and Class introduction
Overview of the Disciple approach to agent
development
Modeling an experts problem solving process
How to choose a PhD advisor Brainstorm of the
modeling process
Assignment
3
Course objectives
Hands-on experience with designing and developing
an end-to-end instructable agent.
Writing a publishable research paper, by each
student, focusing on one aspect of the agent
development process.

• In-depth evaluation of the Disciple
  agent-development process
• Strengths and weaknesses
• Lessons learned
• Future research directions.

Development of a comprehensive evaluation
methodology.
Writing a detailed research report (by the entire
class) on how to choose a PhD advisor (both
general advice and description of the developed
agent). Making the agent publicly available.
Publishing a joint paper (by the entire class) on
how to choose a PhD advisor.
4
Overview of the agent development
Course objectives and Class introduction
Overview of the Disciple approach to agent
development
Modeling an experts problem solving process
How to choose a PhD advisor Brainstorm of the
modeling process
Assignment
5
(No Transcript)
6
Modeling-based ontology specification
Identify the strategic COG candidates for the
Okinawa_1945 scenario
strategic COG candidate
Which is an opposing force in the Okinawa_1945
scenario?
scenario
US_1945
Identify the strategic COG candidates for US_1945
Okinawa 1945
Is US_1945 a single member force or a
multi-member force?
Ontology specification
US_1945 is a single-member force
force
Identify the strategic COG candidates for
US_1945which is a single-member force
What types of strategic COG candidates should I
consider for this single_member force?
opposing force
single member force
multi member force
US 1945
Subject Matter Expert Knowledge Engineer
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
Modeling an experts problem solving process
Course objectives and Class introduction
Overview of the Disciple approach to agent
development
Modeling an experts problem solving process
How to choose a PhD advisor Brainstorm of the
modeling process
Assignment
19
Modeling based on the task reduction paradigm
Illustration of modeling identification and
testing
General guidelines for the modeling process
Specific guidelines for the modeling process
20
The general task reduction paradigm

• A complex problem solving task is performed by
• successively reducing it to simpler tasks
• finding the solutionsof the simplest tasks
• successively composing these solutions until the
  solution to the initial task is obtained.

S1
T1

S11
S1n
T1n
T11

S111
T111
S11m
T11m
21
Question-answering based task reduction
Let T1 be the problem solving task to be
performed. Finding a solution is an iterative
process where, at each step, we consider some
relevant information that leads us to reduce the
current task to a simpler task or to several
simpler tasks. The question Q associated with
the current task identifies the type of
information to be considered. The answer A
identifies that piece of information and leads us
to the reduction of the current task.

T1
S1
Q1

S11
A1n
A11
S1n
T1n
S11a
T11a
S11b
T11b

S11b
Q11b

S11bm
S11b1
A11bm
A11b1

T11bm
T11b1
22
Modeling the planning process

T1 is a general action/taskthat accomplishes the
goal. Ask a question about the current situation
to determine alternative ways of performing this
action. If the answer of Q1 is A11, then to
perform T1 one could perform T11. When the action
to perform (e.g. A11) is completely defined,
break it down into sub-actions (e.g. T11a and
T11b). In this case the question and the answer
summarize the solution, or could simply be
absent. Continue this reduction process until the
you obtain elementary actions.
T1
S1
Q1

A1n
A11
S1n
S11
T1n
T11
Q11a

S11
A11a
S11a
T11a
S11b
T11b
S11b
Q11b

S11bm
S11b1
A11bm
A11b1

T11bm
T11b1
23
Modeling the planning process (cont.)

T1
Follow the tree from bottom to top to compose the
elementary actions into plans, as illustrated
bellow S11b is the union of S11b1 S11bm
This leads to alternative plans. S11 is the set
of plans obtained by composing the sub-plans from
S11a and the sub-plans from S11b.
S1
Q1

A1n
A11
S1n
S11
T1n
T11
Q11a

S11
A11a
S11a
T11a
S11b
T11b
S11b
Q11b

S11bm
S11b1
A11bm
A11b1

T11bm
T11b1
24
Modeling the critiquing process
To assess a course of action with respect to a
specific principle or tenet one needs a certain
amount of information about that course of
action, information related to that principle or
tenet. This information is obtained by asking a
series of questions. The answer to each
question allows one to reduce the current
assessment task to a more specific and simpler
one. This process continues until one has
enough information to recognize a weakness or a
strength.

T1
S1
Q1

A1n
A11
S1n
S11
T1n
T11a
Q11a

A11am

A11a1
S11am
S11a1
Each leaf is a solution (a weakness or a
strength). The solution corresponding to an
intermediate node is the union of the solutions
of its immediate children.
25
Modeling the identification process

To identify a center of gravity candidate for a
given scenario (e.g. Sicily_1943) one needs a
certain amount of information which is obtained
by asking a series of questions. The answer to
each question allows one to reduce the current
identification task to a more specific and
simpler one. This process continues until one
has enough information about an entity in the
scenario to identify it as a center of gravity
candidate.
T1
S1
Q1

A1n
A11
S1n
T1n
S11
T11a
Q11a

A11am
A11a1

S11am
S11a1
Each leaf of the tree is a solution (a COG
candidate). The solution corresponding to an
intermediate node is the union of the solutions
of its immediate children.
26
Modeling based on the task reduction paradigm
Illustration of modeling identification and
testing
General guidelines for the modeling process
Specific guidelines for the modeling process
27
I need to
Identify and test a strategic COG candidate for
the Okinawa_1945 scenario
What kind of scenario is Okinawa_1945?
Okinawa_1945 is a major theater of war scenario
Therefore I need to
Identify and test a strategic COG candidate for
Okinawa_1945 which is a major theater of war
scenario
Which is an opposing force in the Okinawa_1945
scenario?
Japan_1945
Therefore I need to
Identify and test a strategic COG candidate for
Japan_1945
Is Japan_1945 a single-member force or a
multi-member force?
Japan_1945 is a single-member force
Therefore I need to
Identify and test a strategic COG candidate for
Japan_1945 which is a single-member force
continues
US_1945
28
I need to
Identify and test a strategic COG candidate for
Japan_1945 which is a single-member force
What type of strategic COG candidate should I
consider for a single-member force?
I consider a strategic COG candidate with respect
to the government of Japan_1945
Therefore I need to
Identify and test a strategic COG candidate with
respect to the government of Japan_1945
continues
I consider a strategic COG candidate with respect
to the controlling elements outside of the
government of Japan_1945
I consider a strategic COG candidate with respect
to the civilization of Japan_1945
I consider a strategic COG candidate with respect
to the armed forces of Japan_1945
I consider a strategic COG candidate with respect
to other sourcesof moral or physical strength,
power and resistance of Japan_1945
29
I need to
Identify and test a strategic COG candidate with
respect to the government of Japan_1945
What type of governing body controls Japan_1945?
Japan_1945 has a feudal god-king government
Therefore I need to
Identify and test a strategic COG candidate with
respect to the feudal king-god government of
Japan_1945
Who is the feudal god-king of Japan_1945?
Emperor_Hirohito
Therefore I need to
Identify Emperor_Hirohito as a strategic COG
candidate with respect to the government of
Japan_1945
Test whether Emperor_Hirohito is a viable
strategic COG candidate with respect to the
government of Japan_1945
continues
30
I need to
Test whether Emperor_Hirohito is a viable
strategic COG candidate with respect to the
government of Japan_1945
What is the strategic goal of US_1945?
Unconditional_surrender_of_Japan
Therefore I need to
Test whether Emperor_Hirohito can impose
Japan_1945 to accept the unconditional_surrender_o
f_Japan
Does Emperor_Hirohito have the power to cause the
people_of_Japan_1945 to accept unconditional_surre
nder_of_Japan?
Yes, because he is seen as divine by the
people_of_Japan_1945 and his will is actually
their will
Therefore I need to
Test whether Emperor_Hirohito, who can impose his
will on the people_of_Japan_1945, can impose
Japan_1945 to accept the unconditional_surrender_o
f_Japan
continues
31
I need to
Test whether Emperor_Hirohito, who can impose his
will on the people_of_Japan_1945, can impose
Japan_1945 to accept the unconditional_surrender_o
f_Japan
Does Emperor_Hirohito have the power to cause the
military_of_Japan_1945 to accept
unconditional_surrender_of_Japan?
Yes, because Emperor_Hirohito is the god-king of
Japan_1945 and the commander in chief of the
military_of_Japan_1945
Therefore I need to
Test whether Emperor_Hirohito, who can impose his
will on the people_of_Japan_1945 and on the
military_of_Japan_1945, can impose Japan_1945 to
accept the unconditional_surrender_of_Japan
Does Emperor_Hirohito have the power to cause the
government_of_Japan_1945 to accept
unconditional_surrender_of_Japan?
Yes, because Emperor_Hirohito is the head of the
government_of_Japan_1945
Therefore
Emperor_Hirohito is a strategic COG candidate
that cannot be eliminated
32
Sample modeling for US_1945

I need to
Identify and test a strategic COG candidate with
respect to the civilization of US_1945
At what level is the civilization of US_1945
organized?
US_1945 is an industrial civilization
Therefore I need to
Identify and test a strategic COG candidates with
respect to the industrial civilization of US_1945
Who or what is a strategically critical
industrial civilization element in US_1945?
Industrial_capacity_of_US_1945 because it is a
major generator of war_materiel_and_transports_of_
US_1945
Therefore I need to
Identify industrial_capacity_of_US_1945 is a
strategic COG candidate with respect to the
civilization of US_1945
Test whether industrial_capacity_of_US_1945 is a
viable strategic COG candidate with respect to
the civilization of US_1945
continues
33
I need to
Test whether industrial_capacity_of_US_1945 is a
viable strategic COG candidate with respect to
the civilization of US_1945
What is the strategic goal of Japan_1945?
US_giving_honorable_end_of_hostilities_to_Japan
Therefore I need to
Test whether reducing industrial_capacity_of_US_19
45 would cause US_1945 to accept
US_giving_honorable_end_of_hostilities_to_Japan
What is the main effect of the reduction of the
industrial_capacity_of_US_1945?
The reduction in the quantity of newly produced
war_materiel_and_transports_of_US_1945
Therefore I need to
Test whether the reduction in the quantity of
newly produced war_materiel_and_transports_of_US_1
945 by the industrial_capacity_of_US_1945 would
cause US_1945 to accept US_giving_honorable_end_of
_hostilities_to_Japan
Would the reduction in the quantity of newly
produced war_materiel_and_transports_of_US_1945
have a deteriorating effect on US_1945?
No, because US_1945 has a surplus of
war_materiel_and_transports_of_US_1945
Therefore
Industrial_capacity_of_US_1945 is a strategic COG
candidate that can be eliminated
34
Modeling based on the task reduction paradigm
Illustration of modeling identification and
testing
General guidelines for the modeling process
Specific guidelines for the modeling process
35
General guidelines

• Partition the domain into classes of problems.
• Select representative problems for each class.
• Model one class at a time.
• Model one example solution at a time.
• Organize the top level part of the problem
  solving tree to identify the class of the problem.

36
Partition the domain into classes of problems
Workaround Domain
Workaround damaged tunnels
Workaround damaged bridges with fording
Workaround damaged bridges with fixed bridges
Workaround damage
Workaround damaged bridges
Workaround damaged bridges with floating bridges
Workaround damaged bridges with rafts
Workaround damaged roads
37
Partition the domain into classes of problems
COA Domain
Each principle and tenet leads to a different
class of critiquing task.
38
Partition the domain into classes of problems
CoG Domain
Major theater war
Counter insurgency
Type of scenario
Counter terrorism
Peace keeping
Drug/law enforcement
39
Modeling based on the task reduction paradigm
Illustration of modeling identification and
testing
General guidelines for the modeling process
Specific guidelines for the modeling process
40
Specific guidelines for the modeling process
1. Identify the problem to be solved, then form
a task name by writing a clear, thorough, natural
language sentence describing that
problem. 2. Follow each task or sub-task with a
single, concise, question relevant to solving the
named task. - Ask small, incremental questions
that are likely to have a single category of
answer (but not necessarily a single answer).
This usually means ask who, or what, or
where, or what kind of, or is this or that
etc., not complex questions such as who and
what, or what and where, 3. Follow each
question with one or more answers to that
question. - Express answers as complete
sentences, restating key elements of the question
in the answer. - Even well formed, simple
questions are likely to generate multiple
answers. Select the answer that corresponds to
the example solution being modeled and continue
down that branch. Go back and explore possible
branches in a solution tree when you are ready to
model a new example solution.
41
Specific guidelines for the modeling process
(cont.)
4. Evaluate the complexity of each question and
its answers. When a question leads to apparently
overly complex answers, especially answers that
contain an and condition, rephrase the question
in a simpler, more incremental manner leading to
simpler answers. 5. For each answer, form a new
sub-task, or several sub-tasks, or a solution
corresponding to that answer, by writing a clear,
thorough, natural language sentence describing
the new sub-tasks or solution. - To the extent
that it is practical, incorporate key relevant
phrases and elements of preceding task names in
sub-task names to portray the experts chain of
problem solving thought and the accumulation of
relevant knowledge. - If the answer has led to
several sub-tasks, then model their solutions in
a depth-first order.
42
Specific guidelines for the modeling process
(cont.)

• 6. After completing a solution tree for an
  example solution, revisit the potential branches
  of that tree to model additional example
  solutions within that category of solutions,
  reusing existing model components to the greatest
  extent possible.
• Utilize the tools and learning ability of
  Disciple to the greatest extent possible to
  minimize the amount of modeling required.
• Only completely model solutions that are unique
  in their entirety. Entirely unique solutions
  will be rare.

43
How to choose a PhD advisor
Course objectives and Class introduction
Overview of the Disciple approach to agent
development
Modeling an experts problem solving process
How to choose a PhD advisor Brainstorm of the
modeling process
Assignment
44
Preliminary modeling
I need to
Identify a potential PhD advisor for Bob
What is a research area of interest for Bob?
Artificial Intelligence
Therefore I need to
Identify a potential PhD advisor for Bob in the
area of Artificial Intelligence
Who is a professor who is an expert in Artificial
Intelligence?
John Doe
Therefore I need to
Determine whether John Doe is an appropriate PhD
advisor for Bob
continues
Dan Smith

Information security

45
Preliminary modeling
Determine whether John Doe is an appropriate PhD
advisor for Bob
Is John Doe likely to stay on the faculty for the
duration of Bobs dissertation?
Yes because John Doe is a tenured faculty
Other possible answers
Yes because X has a very good record is very
likely to get tenure
No because X is likely to retire in the near
future
No because X is a visiting faculty that will
leave next year

Therefore I need to
Identify John Doe as a potential PhD Advisor for
Bob
Assess whether John Doe is an appropriate PhD
Advisor for Bob
46
I need to
Assess whether John Doe is an appropriate PhD
Advisor for Bob
Which are the main criteria to consider in
judging John Doe?
Professional reputation (1,11,14,15,24,25),
general personality and work compatibility
(21,22 8, 10, 23), learning experience
(12,14,19,20,24), responsiveness to the student
(2, 3, 5, 6, 10,26), support (15, 16, 17, 18,
19), quality of results (9, 12, 13, 20,24,25)
Therefore I need to
Assess John Doe with respect to his professional
reputation
Assess John Doe with respect to his general
personality and work compatibility
Assess John Doe with respect to the expected
learning experience
Assess John Doe with respect to his
responsiveness to students
Assess John Doe with respect to the support
offered to students
Assess John Doe with respect to the expected
quality of students work
47
Assignment Understanding the expertise domain
Perform literature search on the subject of
choosing a PhD advisor and prepare a short
presentation for the class. Due date February
4th, 2002