class notes

[Jesse Cirimele]

virtual nurse summarization

video of the virtual nurse

is it better with or without the red socks question?
-study says yes
-half the class says no,
-depends on the context
-introduces the idea that it's ok to laugh a little
-when you ask engineers to design something, empathy isn't really a
part of the equation. really interesting to consider that.
-confirms the fact that you it's not a computer
-can we trust that people said that they liked it? they might have
been trying to please the experimentalist or doctor.
-we shouldn't ever ask the question "do you like this computer
interface". stupid. what does 5.5 out of 7 mean?
-want to have some kind of comparison point

agent used a lot of non-verbal cues. what did we think about the
non-verbal cues?
-like it, wanted to know more about what they were (what is a full-hand motion?)
-it's really hard to design non-verbal gestures that feel right
-if they don't look right then they look like twitching, or are distracting
-(really fast comments)...
-the same reason people don't like it is probably the same reason that
people liked it. it wasn't a person and allowed people to go slow, ask
extra questions, and it didn't come into the "uncanny valley"

it wasn't 3d, should it have been?
-no, lower fidelity allows for people to imagine more
-it would be interesting to see how much you could remove (color, turn
into line art) and still have it work?
-strawman, compare to a giant paper, html document (probably will win)
-has to be something that's anthropomorphic

touch input vs speech input
-voice input is still really hard
-but it actually works (on the recognition side)
-speech takes higher cognitive load
-it's not just speech vs touch, it's also unconstrained vs constrained
-wizard of oz the whole thing, agent uses computerized voice to reply
-privacy issues, touch means you don't have to say "i'm sick" to your
roommate too

don't ask the question: which input modalities is better?
do ask the question: _when_ is each of these modalities better?
-speech doesn't need hands (if you can't type), but you need to be
able to recover quickly
-speech user interfaces need redundancy (like human-human
conversations), does the computer understand what you said?
-speech wins: can't user your hands, want more empathy,
high-perplexity situations
-speech wins in driving (even if it's not a good idea)
-manual input wins when picking from a small number of options

activity: 3 people virtual nurse run-through.
...

speech acts paper summary

what speech interfaces have you used before?
-airlines, after a certain number of errors brings you to a person
-credit card company can say card number over the phone
-called the dmv, took 15 minutes to go through interface, "i'm sorry
but you can't use this service". a person would have been able to
figure this out sooner much much faster.
-gethuman.com - ways to talk to a person
-it's maybe not good for people who have spent a lot of time trying to
figure the problem out already. system prob won't be able to solve
complex problems.
-hierarchical systems are hard to navigate with speech, is search a
good alternate to this?
-listening to many search results might be just as annoying as
navigating a deep tree