Agents%20that%20Reduce%20Work%20and%20Information%20Overload

1
Agents that Reduce Work and Information Overload
by Pattie Maes Published in Communications of the
ACM, July 1994

• Lauri Sumari, 2002

2
Agenda

• walkthrough with comments
• feel free to interrupt
• general critique
• conclusion
• discussion

3
Earlier Agents (pre-1994)

• programmed by the end-user (1988)
• knowledge-based (Greek oracle) (1991)
• problems
• fixed knowledge
• too much non-reusable effort required

4
Solution Machine Learning

• everything gets better gradually
• agent's competence
• user's model of the agent
• trust enjoyed by the agent
• number of automated tasks (efficiency)

5
How to Learn

• analyzing user actions
• search for patterns
• user feedback
• reaction to suggestions
• user gives explicit examples
• asking other agents
• trustworthiness is learnable

6
Example Agents

• four example agents
• email handling
• meeting scheduling
• news filtering
• recommendation of entertainment (killer app?)
• choice of domains arbitrary

7
Critique

• "virtual users" and over-relying users?
• legal responsibility of agents black white?
• sponsored by Apple?

8
Conclusion

• Pros
• understandable
• good examples of agents at work
• various learning methods

• Cons
• some opinions
• "sponsored by Apple"?