AGENTS Heather Nodler INF 385Q - Knowledge Management Systems Dr. Don Turnbull

Heather Nodler
INF 385Q - Knowledge Management Systems
Dr. Don Turnbull
April 26, 2005
- Autonomous software processes that
interact with users and with each other
according to certain conventions
- Assist users with information gathering
and other processes
- Sometimes called “bots”
Why Agents?
In the Internet age, agents help users
cope with ever-increasing amounts of
information and work.
Some Requirements
• Processes must flow smoothly and
• Users need sense of control
• Users’ privacy must be maintained
• Users must be able to see/change
actions taken by agents
How Agents Work
Knowledge-based - Interface agent endowed
with knowledge of a specific application and a
specific user or set of users
Learning - Autonomous agent performs tasks
like a personal assistant, learning user’s
behavior through observation and
communication with other agents
Agents That Learn
• Require minimum background knowledge
• Program themselves, are more dynamic
• Increase in reliability over time, gain user
• Can get help from other agents to deal with
specific user tasks
• Assist in transfer of knowledge among users
in a group
Agents That Learn
• Use of application requires a good deal of
repetitive behavior
• Can create privacy issues, especially when
agents share info with each other
• Lack subtlety of human judgement/behavior
• Agents can exhibit imperfect behavior,
especially early on
• Users might have overblown expectations
• “Cold start” problem
Examples of Learning Agents
E-mail handling applications
Meeting scheduling agents
Electronic news filtering agents
Entertainment recommender systems
Online dating systems
Other applications
The Easy Bee
Agent Utopia
• Will serve as social equalizers, giving
full support staff to people who couldn’t
afford it otherwise
• Will make our lives easier
• Will grow smarter, more effective, and
will be entrusted with more critical
• Everything will be transparent
Agent Dystopia
• Will create conflicts/confusion
• Will invade users’ privacy
• Will allow more code-savvy users to
deceive/take advantage of others
• Will take the human element out of
processes that benefit from some error
(e.g., markets)
• Will create overstandardization
Questions for Discussion
• To what extent do you use/trust agents
now? Examples?
• What will agents of the future do?
• How can agents be applied in KM?
• Are some some things too important to
be entrusted to agents?