ibm research june 14, 2007 an ip continuum for adaptive interface design jeff pierce...
TRANSCRIPT
IBM Research
June 14, 2007
An IP Continuum forAdaptive Interface Design
Jeff [email protected]
© 2007 IBM Corporation
2
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Two preliminary comments
Unindicted co-conspirator: Charles Isbell– Assistant Professor in College of Computing @ Georgia Tech
– AI researcher, primarily machine learning
– Co-director of APIE lab
– Adaptive Personalized Information Environments
Questions, few answers– Primarily work done at Georgia Tech
3
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Design of Adaptive (“Intelligent”) Interfaces
Common goal: “right” information / action at the “right” time in the “right” way
Traditional approaches:– Provide people with access to information / commands
– Use AI to present information / act automatically
Challenges:– People
– AI
4
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
People are…
Forgetful Easily distracted Idiosyncratic Inconsistent Short-sighted
… but still surprisingly good at many tasks
5
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Artificial Intelligence is…
Short on intelligence Easily mistaken Slow to learn Limited by available information Dependent on feedback
…but surprisingly good at finding patterns given sufficient data
6
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Challenge
Draw on the strengths of people and AI while mitigating their weaknesses
– Combine HCI + Machine Learning
Obvious questions: how?
7
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
One answer
Mixed-initiative user interfaces– No Action – Dialog – Action
Horvitz ’99
8
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Example Case: Helping users make better decisions in the morning
Hunt ‘02Wesveen et al ‘01
Approach: next-generation alarm clock No Action – Dialog – Action doesn’t work particularly
well here
9
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
An Alternative: The IP Continuum
10
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Night Before
Decisions: alarm time, what to wear Displays unusual: schedule, weather
Unusual Usual
11
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
During Alarm
Decisions: when to get up Displays unusual: schedule, weather, traffic
Unusual Usual
12
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
After Alarm
Decisions: what to wear, when to leave Displays unusual: schedule, weather, traffic
Unusual Usual
13
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
IP Continuum
Who is responsible for accomplishing a particular task?
HCI AIAdaptive / Intelligent User Interfaces
14
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
IP Continuum
Particular labels (content and position) not important; relative responsibility is
Tool for inspiring and comparing alternative designs
Not really intended to be prescriptive (at least not yet)
15
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
IP Continuum
Associated metrics– Benefit to user (when right)
– Cost to user (when wrong)
– Attention requirement
– Likelihood of success
– Frequency of action
– Quantity and quality of data required
– Quantity and quality of feedback required
– Cost of gathering data
– Resource costs (e.g. storage, processing)
16
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
IP Continuum
Benefit to user (when right)vs.
Cost to user (when wrong)
No benefit,no cost
Much benefit,much cost
17
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
IP Continuum
– Attention requirement
– Likelihood of success
– Frequency of (user) action
Higher Lower
18
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
IP Continuum
– Quantity and quality of data required
– Quantity and quality of feedback required
– Cost of gathering data
– Resource costs (e.g. storage, processing)
Lower Higher
19
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Deployed examples: simple assistance
20
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Deployed examples: independent action
21
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Deployed example: act with recovery
22
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Research project examples
Alarm ClockNext-generation calendarEmail attachments reminderNext-generation remoteSWIMM: personalized “radio stations”
23
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Next-Generation Calendar
Draw attention to infrequent, uncommon meetings
24
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Attachments Reminder
Help user remember to add attachments
Problem: 85% accuracy
Small study: 19 participants
Most popular:
Highlight attach button (D)
Two-step send (E)
25
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Next-Generation Remote
Allow a single remoteto control multiple devices
26
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Next-Generation Remote
27
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
SWIMM
Provides, in essence, personal radio stations
28
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Some open questions
What determines where an application fits?– Where a particular feature of an application fits?
How much variation is there between users?– Spam filters
How much variation is there over time?– In most cases probably shift from left to right, but how much?
Can we be more rigorous about the relative positions of potential interface designs?
29
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Comment: care and feeding of learning systems
30
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Explicit vs. implicit feedback
Explicit– Intentional, high quality
– Costly (to user), low quantity
Implicit– Side effect, low(er) quality
– Free, high quantity
31
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Designing interfaces for feedback
Alternate focus for interface design– Instead of “usability”, “efficiency”
Maximize the quality and quantity of feedback you extract
– Use feedback to drive improvements in usability, efficiency
32
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
SWIMM
Explicit– Never
– Not now
– Not here
– Rating
Implicit– Song completion
– Song reordering
– Change volume
33
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Current work: Personal Information Environments
34
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Personal Information Environments
35
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Examples Services
36
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Information, commands, activities
How can your devices collaborate to improve your interaction experience?
37
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Conclusions & Future Work
The IP continuum is a tool to inform / guide the design of adaptive / intelligent UIs
Possible next steps:– Amass experience from a variety of application domains
– Distill design principles
– How to guide choice of region for particular application– …and possibly user(s)
– Quantify trade-offs along continuum (?)
– Explore approaches to designing for feedback
38
IBM Research
An IP Continuum for Adaptive Interface Design | Jeff Pierce © 2007 IBM Corporation
Questions?