Sin ttulo de diapositiva

1
SPAIN
Agents to Simulate Social Human Behaviour in a
Work Team
Arantza Aldea
aaldea_at_etse.urv.es
Barcelona, February 2003.
2
Research Activities

• Applications of AI to industry
• Multi-Agent Systems
• Ontology-Based Search Information Extraction
• Model-Based Reasoning

3
Multi-Agent Systems

• Information Society
• AgentCities
• Simulation of work teams

4
Agents to simulate social behaviour

• Introduction Team Configuration
• The Model
• General Architecture
• Internal Architecture of the Agents.
• Tasks Representation
• Organisation Level
• The Prototype
• Future Work
• Conclusions

5
Introduction
Brief Description

• In any complex project that requires several
  people working together, the configuration of the
  best possible team within the working
  constraints, is a difficult task (Biegler et al.
  1997).
• Intelligent Multi-Agent Systems can be used to
  simulate the human behaviour
• Emotions and Personality
• Sociability
• Groups Organisation
• Case Study selection of people to integrate a
  team in charge of the conceptual design of a
  chemical process.

6
The Model General Architecture
The Team Simulation Process.
7
The Model Agents Architecture
Perception
Sensor
Social Status
Cognition
Emotion
Physics
Behaviour
Actor
Basic Architecture of PECS Agents. (Urban 2000)
8
The Model Agents Architecture
Perception
Social Status
Cognition
Emotion / Personality
Behaviour
Actor
PECS Modified Architecture.
9
Internal Parameters of the Agents

10
Personality Trends
Amiable. Priority Relationships Speciality
Support Were all in this together so lets
work as a team
Expressive. Priority Relationships Speciali
ty Socialising Let me tell what happened to
me...
Driver. Priority The task Speciality
Being in control I want it done right and I
want it done now.
Analytical. Priority The task Speciality
Processes systems Can you provide
documentation for your claims?
11
The Model Tasks Representation

• Number of Participants
• Duration
• Sequence
• Difficulty of the Task
• Type of Task

• Deadline
• Priority
• Cost
• Quality

12
The Model Organisation Level
Differents organisation types will be tested
Centralised Hierarchy
Tree Hierarchy
Without Hierarchies
13
The Model Agent Behaviour
The agent behaviour emerges by evaluating values
of its internal properties, randomly modified
around the initial values of each of the
properties.
Normal distribution with mean m (internal value
of the agent, e.g. stress, interest, etc.) and
standard deviation s
These random variations are due to the
non-deterministic nature of human behaviour
14
The Model Team Evaluation
Three Dimensions to Evaluate a Team

• Number of Members.
• Agents with Different Characteristics.
• Type of Organisations.

15
The Prototype Main Components
Configuration Component

• Team Configuration
• Tasks Configuration

Results Component

• Show Team Behaviour
• Graphical Statistics (future work)

16
The Prototype User Interfaces
Use of JADE 2.5
Selection of Weight Agent Internal Parameters
Tasks Configuration
17
The Prototype Initial Results
Example Configuration
Team Members
Project

• 1 Project Manager
• 3 Engineers
• 3 Technicians
• 3 Assistants

• 12 Different tasks

Organisation

• Centralised Hierarchy

Number of simulations

• 100

18
The Prototype Initial Results
We observe that

• Agents with high stress and in charge of a
  specialised task were the agents with more
  failures
• Decreasing the stress parameter and executing
  the same number of simulations, the number of
  failures also decreased
• Agents with an amiable personality and working
  in a task by themselves had more failures than
  successes
• The most successful agents contains a high
  experience and analytical personality.
• The creativity parameter only increases the
  number of successes when the agent is in charge
  of specialised tasks.

19
Future Work
Things to do...

• We are evaluating Fuzzy Logic to improve the
  agents behaviour.
• More studies at organisation level social
  characteristics, co-ordination, re-organisation,
  etc.
• Implementation of other types of team
  co-ordination tree hierarchical organisation,
  and no hierarchy
• Validate the experimental results by comparing
  them with human teams
• Unexpected events how external factors affect
  the team behaviour

20
Conclusions
At the end ...
... help the project manager to select the most
suitable team members for a specific project.