270 likes | 285 Views
A Multi-Agent Design of a Peer-Help Environment Julita Vassileva Computer Science Department University of Saskatchewan. Jim Greer Gord McCalla John Cooke Julita Vassileva Ralph Deters. Chhaya Mudgal Vive Kumar Diego Zapata Mike Winter Lori Kettel Kamal Elbashir Shawn Grant
E N D
A Multi-Agent Design of a Peer-Help EnvironmentJulita VassilevaComputer Science DepartmentUniversity of Saskatchewan
Jim Greer Gord McCalla John Cooke Julita Vassileva Ralph Deters Chhaya Mudgal Vive Kumar Diego Zapata Mike Winter Lori Kettel Kamal Elbashir Shawn Grant Kevin Kostuik ARIES Lab at Department of Comp Science at the U of S
Adaptation in a Distributed Environment Finding an Appropriate Resource ? Key Problems: • location of resources (human or electronic) • matching resources to need • motivation to participate
I-Help: a Peer Help Environment in University Class Why? • Peer help is inexpensive • Encourages sense of community • The helper also learns
I-Help: a Peer Help Environment in University Class How? • Asynchronous (discussion forum): CPR • On-line resources created by students • Synchronous help: PHelpS
User Model: a Resource Repository for the Personal Agent User Resources Data Base Data Base • Knowledge • Social Capabilities • Relationships • Time • Currency Matchmaking ... Data Base
The Issue of Motivation • Why would busy people offer help? • to build relationships / obligations • if it’s part of organizational culture • if there is a reward • class participation points ... • performance review • Adding extrinsic motivation • currency in exchange for help??? • Money reflects real social costs and benefits
A Global View: Agent Economy Questions: • What economy type: barter or market? • What real world equivalent for virtual currency? • Zero-sum game or cumulative w.r.t. some resource? • Social control and protection mechanisms?
An experiment (fall 99) • Student peer help for sale!! • 65 students in 3rd year Comp.Sci. class • the I-Help architecture engineered for this class • students use I-Help throughout the class • helpers get paid, helpees pay • rate of pay negotiable (by agents)
User-Agent Relationship • What to do if the user doesn’t keep the contracts made by her agent? • Agent negotiates with or persuades the user to adopt some goal • User has to pay a penalty: I.e. Agent punishes the user
Human-Agent Interaction • Anthropomorphic Agents: • How much autonomy? • What type of relationship? • Agent “Persona” ?
Supporting Communication • Finding peers who are ready, willing and able to help • Supporting effective interaction and collaboration • Motivating users to help • A new dimension of communication: between human world and agent world
Possible Business Applications In workplace environments: • Team formation: locating competent & available employees • Supporting training on the job, just in time • Establishing peer-help networks, reinforcing the “weak ties” in the company • Capturing and developing repositories of organizational knowledge
More Business Applications… Enhancing customer support and helpdesk services: • Automatic dispatching of help-requests to the competent helpdesk employees who are available and managing the waiting lists • Offloading helpdesk employees by shifting most of the requests to the customers
En Example Session: • Domain: Helping secretaries to work with MS Office • Personal Agent (IKE) provides help to the user by: • Finding the related MS Help entry • Finding related materials on the web • Finding a peer helper who is competent and currently on-line to help with the problem
The users enter their competence in their user profiles when they use the system first
The agent has identified the topics from the question and searches for related materials on the Web
The user wants peer help: IKE finds a list of people who are on-line and know about the topic of the question; the user has to select one
The agent of the selected helper alerts him that someone needs his help
The agent of the person asking for help tells her that the selected helper has agreed to help
The chat has finished, each partner has to evaluate the session and the other one; The results of the evaluation are used to update the user profiles
IKE has done his job and thanks the user for agreeing to help