Table of Contents
Intelligent Interface Agents
Outline
Careful: the word “agent” is used differently in different communities
All share ascription of human-like characteristics to software
Whaddya mean by...
Interface = Set of Tools?
Complexity of direct-manipulation interfaces grows quickly
Only a limited set of tools can be available at once
Growth of direct-manipulation interfaces is not sustainable
How do we cope with increasing complexity of applications?
How do we cope with increasing complexity of applications?
Non-solutions
Non-solutions
Interface = Set of Agents?
Both AI and human interface perspectives are essential
PPT Slide
Human Interface: Try to make the best possible interface for engaging the user's intelligence
PPT Slide
Make the system that includes both the user and the system as intelligent as possible
Visions of future interface agents
Issues for future agent interface scenarios
Some people don’t want agents
Agent Personalities
Instructible Agents
Architecture for Agent Interfaces
Architecture for conventional interactive interfaces
Architecture for integrated agent interfaces
“Marionette strings”
The agent must maintain a “mental model” of the interface
Machine Learning Methods for Intelligent Agents
Machine Learning Methods for Intelligent Agents
Characterizing Domains
Generalization
Version Spaces
Clustering
Discrimination Net
Machine Learning References
Agent Infrastructure
Object issues
Mobile Programs
But what about mobile objects?
Actor model of computation
An “HTML” for objects?
An “HTML” for objects?
Apiary (Lieberman, 1983)
What now?
No Agents without Multi-Agents?
Multi-Agent issues arise in the user interface
Multi-Agent issues arise in the user interface
Multi-Agent issues arise in the user interface
Multi-Agent issues arise in the user interface
Multi-Agent issues arise in the user interface
Interactive user interfaces are heterogenous agent systems
Agents on the Internet
Examples of agents in products
Examples of agents in products
Letizia: An Interface Agent for Assisting Web Browsing
Letizia is a user interface agent for a Web browser
Letizia Recommendation Page
Letizia’s Screen Layout
Why the name Letizia?
Traditional Web search agents
Browsing is a cooperative search between the user and Letizia
The most important thing about Letizia is its control structure
User’s Search [Depth-First]
User’s Search & Letizia’s Search
Persistence of Interest
Inferring user intent from browsing behavior
Inferring user intent from browsing behavior
Each heuristic doesn't have to be all that great
Topic pages Personal page
Finds Agent-related papers of home page’s author
Letizia is in the tradition of behavior-based interface agents
Letizia’s main value to the user lies in...
Agent Personalities
Instructible Agents
Current graphical interfaces don’t learn
How should we improve teacher/student communication?
The role of examples in teaching
Programming by Example [or “Programming by Demonstration”]
Agent operates in interactive graphic domain
Mondrian: An Instructible Graphical Editor
Mondrian: An Instructible Graphical Editor
Dominoes: Icons representing visual examples
Storyboards: Visual history of recorded procedures
Speech output
Speech Input
Generalization notices “significant” relations
Generalization prefers designated example objects
Generalization records dependencies
Application: Training for technical procedures
Video as source for learning procedures
User selects subsets of image as visual representation of objects
Graphical annotation of example video frames
Graphical annotation of example video frames
Visual representation of conceptual structure
Initial frame / Disassemble / Final frame
Machine Learning Methods for Intelligent Agents
Machine Learning Methods for Intelligent Agents
Characterizing Domains
Generalization
Version Spaces
Clustering
Discrimination Net
Machine Learning References
Object issues
Agent Infrastructure
Mobile Programs
Sharing knowledge among agents
But what about mobile objects?
Actor model of computation
An “HTML” for objects?
An “HTML” for objects?
No Agents without Multi-Agents?
Multi-Agent issues arise in the user interface
Multi-Agent issues arise in the user interface
Interactive user interfaces are heterogenous agent systems
The agent must maintain a “mental model” of the interface
Wizard of Oz Experiment
Turvy - Maulsby
Wizard of Oz experiment task
Wizard of Oz experiment method
“Short order programming” exercise
Things to think about
Examples of Interface Agents
Eager - Cypher
Eager - Cypher
Dynaform, Roboformat - Schlimmer
Dynaform, Roboformat - Schlimmer
Maxims - Metral, Lashkari
Maxims - Metral, Lashkari
Remembrance Agent
Remembrance Agent - Rhodes, Starner (Emacs)
Kasbah
Kasbah - Chavez, Guttman
Firefly/HOMR/Ringo
Firefly - Metral, Lashkari (spinoff)
Yenta
Yenta
Friend-of-a-Friend Finder - Minar, Moukas
Hardcopy books for learning about Agents
Interface Agent Resources: HCI
Interface Agent Resources: HCI
Interface Agent Resources: AI
Interface Agent Resources: AI
Web resources
Web resources
PPT Slide
|