Computational Modeling of Cognitive Processes in Plan ... · Computational Modeling of Cognitive...
Transcript of Computational Modeling of Cognitive Processes in Plan ... · Computational Modeling of Cognitive...
Computational Modeling of Cognitive Processes in Plan Authoring
J. William Murdock & David W. AhaIntelligent Decision Aids Group
Navy Center for Applied Research in Artificial IntelligenceNaval Research Laboratory, Code 5515
Washington, DC 20375{murdock,aha}@aic.nrl.navy.mil
http://www.aic.nrl.navy.mil/~aha/ida/
TC3 Workshop: Cognitive Elements Of Effective Collaboration(January 15-17 2002, San Diego)
Report Documentation Page Form ApprovedOMB No. 0704-0188
Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering andmaintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, ArlingtonVA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if itdoes not display a currently valid OMB control number.
1. REPORT DATE JAN 2002 2. REPORT TYPE
3. DATES COVERED 00-00-2002 to 00-00-2002
4. TITLE AND SUBTITLE Computational Modeling of Cognitive Processes in Plan Authoring
5a. CONTRACT NUMBER
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S) 5d. PROJECT NUMBER
5e. TASK NUMBER
5f. WORK UNIT NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) Naval Research Laboratory,Navy Center for Applied Research inArtificial Intelligence,Code 5515,Washington,DC,20375
8. PERFORMING ORGANIZATIONREPORT NUMBER
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)
11. SPONSOR/MONITOR’S REPORT NUMBER(S)
12. DISTRIBUTION/AVAILABILITY STATEMENT Approved for public release; distribution unlimited
13. SUPPLEMENTARY NOTES TC3 Workshop: Cognitive Elements of Effective Collaboration, 15-17 Jan 2002, San Diego, CA.
14. ABSTRACT
15. SUBJECT TERMS
16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF ABSTRACT Same as
Report (SAR)
18. NUMBEROF PAGES
16
19a. NAME OFRESPONSIBLE PERSON
a. REPORT unclassified
b. ABSTRACT unclassified
c. THIS PAGE unclassified
Standard Form 298 (Rev. 8-98) Prescribed by ANSI Std Z39-18
Murdock & Aha, NRL 2/16
Context
Task: Plan authoring• Intelligent system collaborates with human user• Human is in charge; system makes recommendations
Task: Plan authoring• Intelligent system collaborates with human user• Human is in charge; system makes recommendations
Key Assumption:• Human has a better “big picture” understanding of the
situation than does the intelligent system
Key Assumption:• Human has a better “big picture” understanding of the
situation than does the intelligent system
Benefit:• When the system makes a good recommendation, the
user can accept rather than having to manually enter a new step in the plan.
• Faster plan authoring
Benefit:• When the system makes a good recommendation, the
user can accept rather than having to manually enter a new step in the plan.
• Faster plan authoring
Murdock & Aha, NRL 3/16
Challenge
• An intelligent system can recommend actions for a plan based on its limited knowledge.
• We are considering situations in which the human has a better general understanding of the problem.
• The recommendation the system makes will, at first, be worse than what the human can do.
• How can we get the system to improve during the course of the planning process?
• An intelligent system can recommend actions for a plan based on its limited knowledge.
• We are considering situations in which the human has a better general understanding of the problem.
• The recommendation the system makes will, at first, be worse than what the human can do.
• How can we get the system to improve during the course of the planning process?
Murdock & Aha, NRL 4/16
Our Main Theme
1. Determine:a. Goals: what the user is trying to dob. Approach: how the user is trying to do it
2. Generate suggestions that are compatible with both – i.e., Because the user is the expert, understand the user
and then do things the user’s way.
1. Determine:a. Goals: what the user is trying to dob. Approach: how the user is trying to do it
2. Generate suggestions that are compatible with both – i.e., Because the user is the expert, understand the user
and then do things the user’s way.
System requirement: A cognitive model of the userSystem requirement: A cognitive model of the user
Goal: Travel from NRL to USD
Murdock & Aha, NRL 5/16
Cognitive Models
Task-Method-Knowledge (TMK):– Tasks: What a part of a process does.– Methods: How a part of a process works.– Knowledge: What the process uses and alters.
Task-Method-Knowledge (TMK):– Tasks: What a part of a process does.– Methods: How a part of a process works.– Knowledge: What the process uses and alters.
by Car by Train
…
Maps, Routes,Vehicles, etc. Travel
Start Car Drive Stop Car
Existing work on TMK has involved:– Intelligent systems that automatically adapt– Executable cognitive models of recorded protocols– Many other topics– But never cognitive models of users
Existing work on TMK has involved:– Intelligent systems that automatically adapt– Executable cognitive models of recorded protocols– Many other topics– But never cognitive models of users
Murdock & Aha, NRL 6/16
TMK AnalysisInterpret the user’s reasoning and goals
TMK AnalysisInterpret the user’s reasoning and goals
Overview of Decision Making Architecture
TMK PredictionInfer what the user intends to do next
TMK PredictionInfer what the user intends to do next
Automated PlanningSelect actions that accomplish the goals
Automated PlanningSelect actions that accomplish the goals
Recommendation FilteringEliminate actions that are inconsistent with
the predictions of a user’s intentions
Recommendation FilteringEliminate actions that are inconsistent with
the predictions of a user’s intentions
Murdock & Aha, NRL 7/16
User
Plan Authoring Tool
Recommendation
RecommendationFiltering
ExpectedBehavior
TMK Prediction
Knowledge Task
Method Method Method
Task
Task Task Task …
TMK Cognitive Model of the User
Interpretationof user’s
reasoning
Inferred Goal
TMKAnalysis
Action …Action
Fact Operator
Fact Operator…
Potential Actions
Automated Planner
Murdock & Aha, NRL 8/16
An Illustrative Logistics Example
• User has some boxes to ship and two available trucks.• One truck is faster, so a planner will recommend it.• The user wants to use the slower truck.• Challenge: Recognize that the user has chosen the
slower truck and make recommendations that abide by that choice.
• User has some boxes to ship and two available trucks.• One truck is faster, so a planner will recommend it.• The user wants to use the slower truck.• Challenge: Recognize that the user has chosen the
slower truck and make recommendations that abide by that choice.
Murdock & Aha, NRL 9/16
Slow Truck
Boxes
Slow Truck
What the user wants
Slow Truck
Fast Truck
Slow Truck Slow Truck Slow Truck Slow TruckSlow TruckSlow Truck
Boxes
Murdock & Aha, NRL 10/16
What the automatedplanner would do
Slow Truck
Fast Truck Fast Truck Fast Truck Fast TruckFast Truck
Boxes
Fast TruckFast Truck Fast Truck
Boxes
Fast Truck
Murdock & Aha, NRL 11/16
Recommendations withoutcognitive modeling
Slow Truck
Fast Truck
Slow TruckSlow Truck
Boxes
Slow Truck
Boxes
Slow Truck Slow Truck Slow Truck Slow Truck Slow Truck
Murdock & Aha, NRL 12/16
Recommendations withcognitive modeling
Slow Truck
Fast Truck
Slow TruckSlow Truck
Boxes
Slow Truck
Boxes
Slow Truck Slow TruckSlow Truck Slow Truck Slow Truck
Murdock & Aha, NRL 13/16
Plan Authoring Tool
TMK Prediction
Input DriveAction
User
RecommendationFiltering
Drive(SlowTruck)
Load(Box1,SlowTruck)
…Load(Box2,SlowTruck)
TMK Cognitive Model of the User
ConstructShipping Plan
LinearDeliberation …
Decide onVehicle … Input Load
Action
…
Interpretation:Decided on SlowTruck
Goal: Move boxes
TMKAnalysis
Input DriveAction
Trucks,Boxes,
Locations
Unload(Box1,SlowTruck)
Drive(SlowTruck)
Boxes Unload
Trucks Drive…
Automated Planner
Murdock & Aha, NRL 14/16
Domain Applicability
• Data-intensive environments– System can make a significant contribution
• Decisions require additional background knowledge and/or subjective judgments– System must collaborate with the user
• Expert human users– System’s approach should match user’s preference
• Data-intensive environments– System can make a significant contribution
• Decisions require additional background knowledge and/or subjective judgments– System must collaborate with the user
• Expert human users– System’s approach should match user’s preference
• Example: Noncombatant Evacuation Operations (NEOs)– Tasks: Deliberative and execution-time planning– Data: Doctrine, SOPs, and records of past NEOs are available– Requires that human have final authority over decisions
• Example: Noncombatant Evacuation Operations (NEOs)– Tasks: Deliberative and execution-time planning– Data: Doctrine, SOPs, and records of past NEOs are available– Requires that human have final authority over decisions
Murdock & Aha, NRL 15/16
Open Issues
• Is TMK adequate for modeling users?– If not, what augmentations are needed?
• Where do TMK user models come from?1. Encoded by system designers from cognitive studies?2. Extracted automatically through experience?3. A combination of (1) and (2)?
• Is information from user actions adequate for judging user intentions?
– If not, what other information can enable coordination between the user and the system?
• Is TMK adequate for modeling users?– If not, what augmentations are needed?
• Where do TMK user models come from?1. Encoded by system designers from cognitive studies?2. Extracted automatically through experience?3. A combination of (1) and (2)?
• Is information from user actions adequate for judging user intentions?
– If not, what other information can enable coordination between the user and the system?
Murdock & Aha, NRL 16/16
Summary
• Intelligent system collaborates with a human user; the system provides recommendations during plan authoring.
• Cognitive model is used to infer what the user wants to do and how the user wants to do it.
• User’s cognitive model is composed of Tasks, Methods, and Knowledge (TMK)
• Proposed tool recommends actions for doing what the user wants in the way that the user wants to do it.
• Intelligent system collaborates with a human user; the system provides recommendations during plan authoring.
• Cognitive model is used to infer what the user wants to do and how the user wants to do it.
• User’s cognitive model is composed of Tasks, Methods, and Knowledge (TMK)
• Proposed tool recommends actions for doing what the user wants in the way that the user wants to do it.