Intelligent Agents

• Introduction (Mark Bentley)
  • What are agents and agencies?
  • Why use them?
  • Types of agents
• Building agents (Angela Thomson)
  • Building Interface Agents
    • Criteria for Building Interface Agents
    • Approaches to Building Interface Agents
    • Rule-Based Approach
    • Knowledge-Based Approach
    • Machine Learning Approach
    • Machine Learning Example
    • Using Intelligent Agents
    • Interoperation of Agents
    • Group Reasoning of Agents
    • Distributed Communicating Agents
    • Agent Issues
• Agents Today (Rob Parkhill)
  • Agents Today
    • Search Agents
    • Personalized Agents
  • Agents in 5 years
  • Agents in 10 years
  • Agents in 20 years

What are intelligent agents?

Agent

a system that can perform a task with a certain amount of "intelligence" due to some specialized skill (learning systems) or knowledge (knowledge bases).
Basically, intelligent agents don't just blindly do what they are told. They can be set up to evaluate choices and help the user do things.

Agency

where many agents come together, and their combined actions allow a greater range of action. The selection of agents allows the agency to pick the right one for the job. Much like you would not send an plumber to fix a computer, and agency tries to use the agent best suited to the job. Many systems referred to as agents are really agencies but for most uses people don't really need to know which they are using.

Why use intelligent agents?

• hide complexity of tasks
  the computer can do the hard work for the user
• perform tasks on the user's behalf
  the computer can act for the user and either notify them when it is done, or notify them when it notices something they might need to know. This can also be used for intelligent mail and news readers.
• can train or teach users
  training systems can look at the user's skill level and how they use the system and provide case sensitive suggestions. This means beginner users don't have to read man pages, but can still probe the system for information
• help different users collaborate
  acts as a go between for processes. Users can let their agents negotiate for things like meeting times.
• monitor events and processes
  can be set to let know users know what is happening in events, or signal if something happens

Types

Agents can initiate communication, monitor events, and perform tasks.

Agents are being used in:

Mail readers

they can be set to filter out, prioritize and forward messages. Can get problems of mail being forwarded through several users and then deleted without ever being read. Also important information may be missed because that person usually has nothing important to say.

News readers

can be set to flag certain information, filter out unwanted things (an intelligent KILL file). Can run into the same problems as mail of accidentally missing information (even happens with human assistants).

Meeting schedulers and other negotiators

autonomous agents bargaining for meeting times. Theses can cause problems if they learn to outwit the other agents (the "smartest" agent can sometimes manipulate the others)

Information retrieval

agents and agencies can be designed to search for information from a wide variety of sources. Instead of just using one search engine, an agency could look at the possible methods and select several methods which should produce the best results. The U.S. Navy currently has a proposal for an intelligent information system for naval commanders.

Skip to Rob's talk