Spring 1999
Henry Lieberman, Research Scientist
Ted Selker, Visiting Professor
Thursdays, 2-4 pm
Room 095, Media Lab, 20 Ames St.
TA: Brad Rhodes
Increasingly, we are realizing that to make computer systems more intelligent and responsive to users, we will have to make them more sensitive to context. Traditional hardware and software design overlooks context because it conceptualizes systems as input-output functions. Systems take input explicitly given to them by a human, act upon that input alone and produce explicit output. But this view is too restrictive. Smart computers, intelligent agent software, and digital devices of the future will also have to operate on data that they observe or gather for themselves. They may have to sense their environment, decide which aspects of a situation are really important, and infer the user's intention from concrete actions. The system's actions may be dependent on time, place, or the history of interaction. In other words, dependent upon context.
But what exactly is context? We'll look at perspectives from machine learning, sensors and embedded devices, information visualization, philosophy and psychology. We'll see how each treats the problem of context, and discuss the implications for design of context-sensitive hardware and software.
Course requirements will consist of critiques of class readings [about 3 papers/week], and a final project [paper or computer implementation project].
Tenative schedule and readings:
4 February Introduction to "the context problem" Why do computers need to take account of context? What is and isn't context? Context-sensitive vs. context-independent views of computing Combatting brittleness: adapting systems to changing context Readings: Don Norman: How People Might Interact with Agents Nicholas Negroponte: Agents: From Direct Manipulation to Delegation Henry Lieberman and David Maulsby: Software That Just Keeps Getting Better
11 February Context for software agents Determining relevance of context Implicit input and repurposing input Context overload: Dealing with too much context Readings: Jeff Bradshaw: Introduction to Software Agents Pattie Maes: Agents to Reduce Work and Information Overload Brad Rhodes and Thad Starner: The Remembrance Agent
19 February Context for embedded computing Context and communication among devices Time and physical location as context Readings: Mark Weiser: Ubiquitous Computing Pierre Wellner: Digital Desk Hiroshi Ishii: Ambient Interfaces
26 February Context and user interface design Information visualization Context-dependent presentation of information Readings: Edward Tufte: The Visual Display of Quantitative Information Ben Shneiderman: Information Visualization Muriel Cooper: Computers and Design
4 March Context for learning by example Generalizing context The "data description problem" for learning agents Readings: Allen Cypher: Watch What I Do Henry Lieberman: Mondrian: A Teachable Graphical Editor Richard Potter: Just-in-Time Programming
11 March The role of background knowledge as context Active Ontologies The "Size Matters" Approach: Cyc The Rule-Based Approach: Expert systems. The Mining Approach: Information extraction The Interactive Learning Approach: Incremental development The Reactive Approach: Just don't do representation Readings: Guha and Doug Lenat: Cyc Wendy Lehnert: Computers and Car Bombs Rod Brooks: Intelligence without Representation
19 March - Project proposals due Systems that adapt to context Cognitive Adaptive Computer Help User Modeling User-system communication through annotated examples Readings: Ted Selker: COACH: A Teaching Agent That Learns Elaine Rich: Stereotypes and User Modeling
26 March - Spring Break
1 April Philosophical and mathematical positions on context Readings: Jon Barwise and John Perry: Situations and Attitudes Lucy Suchman: Situated Systems Bonnie Nardi: Context and Consciousness
8 April Machine Learning and formal approaches Tom Mitchell, Pat Langley: Machine Learning Craig Neville-Manning and David Maulsby: Sequitur John McCarthy: Circumscription
15 April Sensing context from the environment Pattern recognition for determining context Roz Picard: Affective Computing
22 April Psychological and social perspectives on context Computers as social actors Readings: Cliff Nass and Byron Reeves: The Media Equation Joseph Bates: The Role of Emotion in Believable Agents Brenda Laurel: Metaphors with Character
29 April Final Project Reports
6 May Final Project Reports