Intelligent UIs

1
Intelligent UIs

• Scott Klemmer
• 08 November 2005

2
Midquarter Evaluation

• 8 responses thanks!
• Responders generally enthusiastic about readings
  and format one dissenter basic literature
  should not be reviewed
• Three areas for improvement
• not enough time to do all the readings, write
  the critiques and get enough sleep to go to class
  and participate
• Some way to know how we're doing in the class.
  (especially with projects)
• I think the student presentations should be more
  focused on interaction than lecturing
• Overall Excellent / Very Good / Good / Very Good
  / Poor / Fair / Very Good / Excellent

3
The Direct Manipulation Ideology

• Display as much information as possible
• Predictable
• Rapid, reversable interactions
• User initiates all actions

4
The goal high information density
5
Command Line Low density and indirect
manipulation
6
guis have improved density and more direct
manipulation
7
but still have a ways to go
8
Ben Shneiderman on design methods

• 30 years of planning work in AI is essentially
  down the tubes because of lack of attention to
  the user interface. The designers deliver a
  system and the first thing that the users say is,
  This is great but what we really want to do is
  change these parameters. The designers say,
  Well, you know, we didnt put that in the
  interface. They just havent thought adequately
  about the interface, nor done testing early
  enough.

9
The Intelligent Interfaces Ideology

• Agents know habits, preferences, interests
• Mixed initiative computer is sometimes proactive
• prompt-based telephone interfaces are an example
  of complete computer initiative

10
Some recent successes

• Spam Filtering
• Toyota Prius braking system

11
How Spam Filtering Works

• Uses a Bayesian network
• Begin with a set of ham (good) and spam messages
• Look at tokens (email addresses, words) and their
  relative frequencies in ham and spam
• e.g., mortgage occurs in 400 of 3,000 spam
  mails and 5 out of 300 legitimate emails. Its
  spam probability would be 0.8889(400/3000
  divided by 5/300 400/3000).

12
Understanding Intelligent UIs

• q Why was this message classified as spam?

13
(No Transcript)
14
(No Transcript)
15
Collaborative Filtering

• aka recommender systems
• Introduced in 1992, roughly simultaneously by
• David Goldberg, Xerox parc (email)
• Joe Konstan, Berkeley -gtumn (NetNews)
• and explored soon after by many, including
• Pattie Maes, mit media lab (music)

16
Traditional DM v. Collaborative Filtering
17
How do they work?An Example Algorithm

• Yezdezard Lashkari, Feature Guided Automated
  Collaborative Filtering, Masters Thesis, MIT
  Media Laboratory, 1995.
• Webhound
• Firefly

18
Webhound, Lashkari, 1995
19
Webhound, Lashkari, 1995
20
Webhound, Lashkari, 1995
21
Webhound, Lashkari, 1995
22
Attentional Interfaces

• Chris Schmandt (MIT Media Lab)
• James Fogarty Scott Hudson (CMU)
• Eric Horvitz (MSR)

23
Everywhere Messaging

• C. Schmandt, N. Marmasse, S. Marti, N. Sawhney,
  S. Wheeler, IBM Systems Journal, 2000
• Several systems
• Clues Finds time-critical emails
• Active Messenger Delivers these to one of many
  devices
• Nomadic Radio Wearable audio
• comMotion Location-aware

24
Clues
25
Active Messenger
26
Nomadic Radio
27
comMotion