Greg Detre
10:30 Wednesday, September 18, 2002
guy who could use cellphones to tell where you are:
skits � tell within 100 yards where a person is � what�s the value?
medical? tell heartbeat from voice
computer-supported collaborative meeting tools don�t work
video conferencing is distracting, slow, worse than telephone
meetings
Berkley � got people to take notes with palms, but nobody looked at them
Georgia tech was more successful
his idea is Mr Web, which creates a web page and nags them about things said/written down in the meeting
Colab � xerox parc, each person gets big computer, shared screen
Stanford had an ICQ-chat-like immediate keypress sharing, which worked much better than when there was a � second delay � that added a social distance, where somebody would wander off and not do what they said they would
there�s been some work on synthesising/summarising meetings
but the technology isn�t there � this is my nagging worry
it�s not about holy grail technology, it�s about what you can do at any stage
he�s looking for scenarios that are robust within broken technology
e.g. eye tracker, which doesn�t work so none of the interfaces, theirs is so much more robust using just gazes
I didn�t like the door
does the door improve the social relationship?
he�s extremely busy, and needed a vestibule
is that relevant to other people? probably, and not all of them have secretaries
can techn be useful in mediating between people?
what if people can�t stand talking to a computer?
would the door be better without a model, i.e. just with a script that flashes up on the office-occupant�s screen?
let�s say that it gets 95%
you learn to trust it, and it continually misses things (e.g. humans stop being alert because they rely on the nuclear power station checks) � are you better off without it? kayla pointed out that that tech-assisted 95% is better than the 70% you�d have actualy got on your own
good HCI is taking the tool out of the task
but, we buy things for reasons of style
goals of CA computing � discovering + creating scenarios that use the min sensors and max modelling to make as robust, believable interactions as poss, where expectations of the human are not broken
his biggest criticism of criticism:
that I didn�t look at the questions (I didn�t get them) and that it was a kneejerk attack
considers what other metrics would be good in a LAFCam
steadiness in the steadicam
annotate buttons don�t get used
voice pitch
experimental design � use any trick in the book
feel it
wizard of oz � someone pretends to be the computer
storytelling
roleplaying
list
detailed scenarios, listing possibilities
mockups � doesn�t actually work
prototypes
rigorous experiments
control + test
simplest first, figure out the important questions
dry-run quick questions
focus groups (problematic, group forces)
questionnaires (people often don�t know/say what they really feel)
quality metric
didn�t use a double-silvered mirror and expensive facility run by a professional because he would rather have a developer
important belief in experimentation + evaluation side-by-side
product guys want someone they can buy the prototype from � research people didn�t understand why the product people wouldn�t buy it
talked about the OHP with a clear screen
just took the back off it, didn�t work initially, turned off the lights, dropped chalk dust on the back to show that it didn�t need a clean production area, withstood being dropped, sent it out to a 3rd party display manufacturer, became the highest profit margin notebook ibm ever sold
cognitive adaptive computer help
designed for writing LISP
user advised not by a manual or other person, but by an adaptive on-screen mechanism
either the user or computer can initiate help, adaptive user model and records history of interactions
advisory help doesn�t intrude, but is directed
context-dependent help helps novices avoid startup problems
lisp course usability study
found to improve perceived and actual usability, as compared with the same window system
they felt more comfortable with lisp, and did not experience de-motivation with the tutorial system