Notetaker:	Marcia Crosland

Adaptive User Interfaces 
Presentation October 7,1994

Schlimmer, J. C., & Hermens, L. A. (1993). Software agents: Completing
patterns and constructing user interfaces, Journal of Artificial
Intelligence Research

Questions posed by the presenter:
(1). Is this a useful system?
(2). Does the user train the system or vice versa?
(3). Is always adhering to the paper paradigm a good thing?

Open Discussion Time: 1:37

Question: If you misspell into a BOX, is it permanent?
 Yes, even if you spell it right it might come out wrong in the BOX.

Comment: The main problem with this system is that it is not
scalable. The algorithms are not appropriate for all fields, such as a
census taker.

Comment: Perhaps the system would useful if it was a front into a data
base, what you present to a user.  .......Well, maybe if you
extrapolate the data base, change the repository of the data base.

Question: Is this system adaptable? Is it useful for all tasks, such
as taking notes for the next poem I want to write? How could it be
useful if "it" could not predict what I was going to write next.  I
don't believe it is adaptable.  Entry into this system would need to
be predictable, reliable, and high frequency items for it to be
useful. Perhaps it is better suited for non creative entries, special
groups, and inventory lists but not for all professions and tasks.

Question: If this system is pen based what happens if you just want to
doodle?

Comment: The New York Stock Exchange is a pen based system. They enter
new information into an existing portfolio. A new field could be added
when a company undergoes some transformation such as bankruptcy or a
merger.

Professor comment: Maybe we are asking the wrong questions. AI
research generated this system. The designers had knowledge. The task
is simple. You write something and out pops something. The problem
really is, as stated before, one of scalability. It works by adapting
to the task and how the user adapts to the task. It doesn't adapt to
the user.  This system arose from theory, not application. Watch the
user and create a better interface for the user.
If you understand the 2 algorithms driving the system, it is a simple machine.

Comment: Change the domain to that of an insurance agent. There are
forms that have names and data. Something new can be written and the
interface will adapt to the new information. The system infers what
comes out is a new category. I could see that this would useful.

Question: Is the system adaptable then?  No, the system is not really
adaptable because the user has little opportunity to change things.

Question: Is it adaptive?  Yes, by the definition established in an
earlier class, the system is adaptive. You enter something, then it
will adapt automatically.

Question: But what about the speed of use and the ease?  Speed
decreases but what do we know of the errors? If you have a long list
of notes the speed certainly decreases. Systems like Newton make many
errors.

Question: Does it keep pausing?  Yes. The system is just too
simple. It presents too many choices at one time. The user can see
only so much at one time.  Human Factors has shown us that a user can
only pay attention to a certain amount of information at any given
time.

Take the minivan example, the options could go on forever. Someone
could order a red minivan with many options but never again. The
system would need to drop categories that are not used after a fixed
time.  A list frequency algorithm would need to be added.

Other serious problems: If you meet Bob, and then another Bob, and
again, the system just goes through a long list of Bobs. The FMS is
too simple to express complex processes.

Comment: There is a lot of time spent trying to pick up the desired
categories. The processing overhead is great. The system search should
be for your specific needs.

Professor Comment: The zip codes have changed in Germany. There are
many codes that are different even for each street. On my computer I
can narrow the field of what I see on the screen. There is an ongoing
streamlining, as I type in an address the numbers in the field
continue to narrow.  This is one way to improve the system performance
I think.

Last Questions posed by the Presenter: In the paper, it says we don't
want to make the user do anything he doesn't already do. Is this what
we are striving for? Don't we want to move away from pen based systems
instead of supporting our old ways of doing things?

Time: 2:00 PM