Discussion notes from 10/13/94 8113d

Alfred Kobsa: User Modeling: Recent Work, Prospects and Hazards.

Susan Farrell

1. None of the UM shells (except BGP-MS) help make
assumptions, they only store and process them. Can systems
be developed that help make assumptions or are all user
assumptions domain (in)dependent?

- generic gain (simple) vs. specific assumptions (useful),
more-specific assump- tions may be application dependent. -
may be dependent on abstraction levels.

how much is domain (in)dependent? - dependent on
relationship of tasks to subtasks

- inference mechanisms that make assumptions about users
depending on where they are in the heirarchy of the UM


2. Should users be allowed to edit their modeling data? Are
users' assessments of their abilities always right?

-make assessment of their expertise?

-could users negatively impact their performance...would
that be self-limiting?

-possibility of creating contradictory data in UM

-providing user preferences may create user
happiness...choices add flexibility

-user should be able to override system

-trigger between what you're looking at and what your goals
are could cause stress? Impairing efficiency?

-UM editing is a metadialog that increases complication of
system. short-term data, e.g., task model, may not make
sense to modify,; whereas long-term modification might make
more sense, e.g., if UM is wrong about an assumption.

-can't do UM in natural language because machine doesn't
represent knowledge that way; can't do UM in system's
terms, because user won't understand it

-multiple levels of abstraction, like groupings,or
stereotypes that could translate into assumptions?

-what about data entry method preferences, such as mouse,
keyboard, menu, buttons, etc.?

-task or perference-level alterations?...some things span
those domains. Task-dependent changes make more sense.

-user preferences change for each task, but system tries to
make assumptions about what user wants to do next. More
complex, but heuristics might be used.

-UM and task modeling changes made to interface would
confuse (multiple) users.

-UM changeability based on user expertise?

-system could question you at first for expertise level and
level could be changed by user or system.


3. Is Alfred Kobsa paranoid? Compare and contrast part 9 to
"guns don't kill people, people kill people."

-could system mislead users?

-is the paranoia about marketing data-escape?

-could UM be used to assess worker productivity?

-could system take advantage of user?

-privacy and legal issues

(No. We're all paranoid and for good reasons)

-but it may be unenforceable/impossible to prevent spying

-see comp.privacy: dangers of Social Security Numbers


4. User plan modeling: feasable, scalable to real-size
systems?

5. Representation: user wants to edit model-->computer
stores it efficiently either store so user can read it (AI
approaches?) or make some kind of translator?
-- 
Susan Farrell
Georgia Institute of Technology
susan.farrell@gtri.gatech.edu