Greg Detre
Wednesday, October 09, 2002
people�s behaviour is non-deterministic and highly contingent, with most of the complexity arising out of social interaction
no system except the user can reliably second-guess what the user intends/needs
Dey�s component architecture may not properly equip the designer with the tools to provide the necessary visibility and control
two big issues:
intelligibility
accountability
considers problems with privacy, identity, access, storage
needs to know when to defer to the user
Design
principles and human-salient details required to realize them:
Feedforward � What will happen if I do this?
Confirmation � What am I doing and what have I done?
drew contrasts between desktop and environment
four primary methods of utilising context information:
resolving inferences
tailoring lists of options
triggering automatic behaviours
tagging information for later retrieval
ubiquitous computing
multimodal interaction
physically embodied interaction
dynamic set of devices
lack of a single focal point
multiple simultaneous users
how does the information get in there:
programmed/taught by user
learned
fixed/installed
shared
two guidelines
minimise cost of error � easy undo
minimise user confusion � avoid having multiple rules that depend on single pieces of context
ted: he
felt they avoided the interesting AI issues
when we
tell the system everything, it�s merely adaptable
Consider an untrustworthy world for context-aware session next week
Everyone is going to take a particular scenario
of their choice and present a short (5 mins) explanation of how this scenario
could be understood and augmented by context-aware technology.�� The idea is to focus on some of the papers
and ideas we discussed today by highlighting the implementation issues and
pointing out where a particular architecture (widgets, blackboards, networked,
other...) can resolve the issues.
Maybe consider an adaptive hypermedia scenario? or a file-organiser/GUI scenario?
this is a
real case of whether you trust your context-aware systems in a messy world �
high-risk
because if you�re wearing headphones, you know that you�re insulated
from the world
but if you�ve got this machine that you think you can place your trust
in, and that�s a false trust, then you�re actually worse off
it wants to listen out for horns and screeches � surely their audio
spectra should be fairly identifiable
use the doppler effects
use visual pointers to attract you to the sound localisation
what about the subtler things? or where there�s no screech, because they
just hit you
you�d probably hear the screeching
can it catch things that you wouldn�t otherwise have noticed
what if it creates sounds for things that aren�t sounds
e.g. if there�s someone coming up faster than they should
impose visual blobs to show vehicles coming up behind, or if they don�t
have lights
the problem with vision is that your vision is already busy
vibration???
maybe also connect with vibrations on the bike seat
also give you signals about your bike (either arhythmic motor sounds, or
bike clanking)
maybe have
a paranoia setting, use different modalities to signal different urgencies
if you turn off the paranoia setting, what about manufacturer liability
what does it do that military helmets don�t do?
bitchin� betty � woman�s voice that talks in your ear if you�re going to
crash, and helps wake you up if you�ve been knocked out by g-forces, plus
you�ve got a one-eye visual display
ted�s helmet: they don�t turn on blinkers when you move your head, or
display potholes (audio)
there are people in wheelchairs who can command direction using head
direction, but still waggle when they�re talking (but only in restricted
degrees of freedom)
how the hell does the helmet know about potholes?
I don�t think the helmet should learn over time � it shouldn�t have a user model, in my opinion
i.e. the difference between context-awareness and adaptive context-awareness
in contrast, the house isn�t physically dangerous to you, and the task it has to do is much more complex
having said that, the house is usually more annoying than right
the only thing that people don�t mind is its controlling the heat
I don�t think that houses are actually going to get better at this, because the complexity is so far beyond
interesting in indexing art + poetry using GIS
e.g. walking around Grantchester, where Wittgenstein used to go, and have the poems written there read to you
i.e. the world�s a museum
he�s interested in organising the information so that all the museum opening hours are easily available
ted: a www designed for moments, not information � what would we do different?
e.g. what modes of transport you have available, user preferences, city-wide problems/exhibitions etc.
Walt Whitman lived just off Brattle
the Chelsea Hotel in NY � opium smell in the halls
widgets vs broadcasting vs blackboards
how people communicate with each other
e.g. IM clients, phone conversations etc.
help you manage your resources, your privacy
the data is mine, so the architecture has to fit that
i.e. no broadcast model, for instance
although you could always build a privacy layer/protocol on top of any architecture
blackboard � big repository of events, owned by my hardware
but a blackboard is centralised, whereas the devices are distributed
in a group of people, do they all have their own? yes
when should information belong to who it�s going to benefit? e.g. when your boss wants to look at the communication patterns between employees
maybe people accept that they just don�t have privacy anymore � or alternatively, there�s a trade-off between privacy and convenience/functionality
file organiser
������� blackboard
are we
really single-channel communicators???
ted:
suspects that the MS Word spell-checker looks at the sentence structure when
suggesting words
how does the system know whether you don�t like what it�s doing now, but will like it later?
one thing you�ll notice about good hotels is that both the human and machine levels of context-awareness and responsiveness will be higher
different areas will have different densities and inter-connectedness of sensors
Kirsh & Agre � philosophers/sociologists
add a paper on architecture
philosophical analyses of contextual situations that exist
me:
adaptive hypermedia???
Minsky
Napier:
databases
paper on how recall spatial iconography
book on AI and geography