d esigning future environments david kirsh dept of cognitive science ucsd
TRANSCRIPT
Designing Future Environments
David Kirsh
Dept of Cognitive Science
UCSD
Question
• How can we design environments that are:
– Cognitively congenial• Cognitively more efficient• Less stressful• Reduced cognitive overload
– More fulfilling – provide a better experience• More aesthetic• More fun• Let us be more creative
– ?? what else??
Agenda
• Background
– Technology
– Changing conception of agent-environment coupling
• Cognitive Principles of Interactivity
• A Science of Design?
• Coordination at Starbucks
Background of Inquiry
Background of Inquiry: Technology
– Walls are data walls
– Internet everywhere
– Wireless everything
– Near field haptics
– Easy telepresence
– Effective digitization of paper– Sensors make it easy to cross
from physical to digital
– Rooms are context aware
Context aware Ubiquitous computing Peripheral Robotic
Architecture
Gehry’s Disney Auditorium: Los Angeles
Agent-Environment Coupling
Classical Model of Activity
D CA
Project Structure (Meaning, Interpret environment )
AGENT ENVIRONMENT
DI’ve got to make
DCA
Formalizing problem solving
Move D onto A
Move C onto AMove A onto D
Move C onto DMove A onto C
Move D onto C
What are we abstracting from?
Activity Space
Activity Space
Activity Space
Activity Space
Activity SpacePure Structure of Task
task environment (state space) is abstraction
Classical formulation
• Humans adapt to structured environments – Develop efficient routines
• Problem is to describe the environment of activity– Classical approach environment is
collection of task environments
– Formally each task environment is a connected graph of choice points
A single task environment
Interim Summary
• Agent lives in many task environments• Switches tasks as necessary• Projects lots of structure
Physical space andStuff occupying it
Theorist postulates as
many task environments
as tasks agent performs
in an environment
So how wrong is this view?
Throw it out completely - almost
• Natural tasks are ill defined
– We do lots of extra-task actions
• Action set not well defined
– Our goals are more flexible
– Other considerations
• Multi tasking causes interference and must be managed
• Consequence function is not well defined or hard to predict
• No well defined set of choice points
• Metric of closeness to goal not usually well defined
X
Two Types of Environment
E state changes autonomously
Griddle
E state changes onlywhen agent intervenes
Tower of Hanoi
Dynamical Environment
Autonomous changes
•Gravity
•Heat
•Syrup
•Air Temp
manageJob is to control a process
regulate
Continuous action, states
Some Very Dynamical Problems
• Coordination problems
World acts back continuously
Part of a system
Some state space problems
Forced Choice
Most of life has a little of both
Starbucks
Real environments are hectic
• Given:– We do lots of things in them– We change tasks a lot– We negotiate our tasks with others– We ‘negotiate’ our tasks with ourselves– Plenty of interruption– We get distracted
Kitchen Environment
Familiar collaborative environment
What extra is going on in these environments?
• Representation rich• Dialogue• Gesture• Situated thinking
– Pointing to representation and talking about what something means in that context
• Negotiating what the goal is and when you have made done well enough
• Managing space• Managing attention• Coordinating your activity so you know what to do next
and what you have already done
How do we put them together?
• Better science of interactivity
• Develop principles of design
Cognitive Principles of Interactivity
Externalization
We externalize to increase our power
Externalization
Externalization
• Externalizing let’s us interact using principles of visual processing that are different than internal visualization and thinking
• We are more data driven – more coupled – than we think
• Externalizing can break cognitive set
• We learn norms of reasoning,tricks of manipulation
Internal necker cube does not oscillate
Examples of Externalization
• Verbalize our thoughts
• Sketch
• Gesture
• Write on paper
• Point
• Set our ingredients before cooking
External Representations that help us think
Conceptual Mathematics. Visual Proofs.
Prove: the sum of the odd numbers,
1 + 3 + 5 + … + 2n – 1 = n2
Baigrie, Brian S., Ed. Picturing Knowledge. Buffalo: University of Toronto Press, 1996.
Conceptual. Mathematics. Visual Proofs.
Baigrie, Brian S., Ed. Picturing Knowledge. Buffalo: University of Toronto Press, 1996.
We coordinate our actions - inside and outside to increase our power
36
Scrabble shows Dynamic Couplingof Projecting & Creating Structure
ecesrrruutt
ee cs rrr uu tt
ece s rrr uu tt crust strut e
restructure
Project Structure = Mentally Represent
Create structure
Mentally reorder
Physically reorder
37
Interaction at high speed
• Manipulate physical objects to save mental manipulation
• Tetris examples:
– physical rotation saves mental computation & is faster
• Piece recognition
• Placement decision
– physical translation improves certainty
• Both interactive strategies involvemillisecond coordination
Work done with Paul Maglio
We develop systems that encode information
39
Short Order Cook: Hamburgers
• Activity is coordinated by reference to state
• Environment is prepared to make state explicitAlex Kirlik’s example
Preparing by demarcating regions
• reduce clutter, reduce combinatorics of problem, track state better
Encoding Assembly Order
• Converts combinatoric nightmare to ‘simple’ hill climbing
Preparing the workspace
• Topological constraints are more natural
Why Re-arrange Cards?
As dealt
A Science of Design?
Interface
Interface - redesigned
Why is one better than another?
Why??
Why is one better than another?
• Cleaner, more white space• Better visual layout• Modularizes activity – helps to plan, review, compare
Design Principle One
• What is semantically related is visually related
• What goes together semantically should of together visually
• What is semantically associated should be visually associated
Discovering Organizing principle
• People try to impose order– We need to figure out what order they will project
so that their behavior becomes predictable
• Pattern discovery 1 4 9 16 25 …
• Gestalting
• _ r _ a _ b _ e -- fragment completion
Enough for now
Design Principle Two
• When you operate on a to effect A – Use intuitive correspondences so that
• Actions on a have a natural interpretation in A• Design so that there are intuitive correspondences
• When there are multiple widgets that affect multiple targets – Use intuitive correspondences so that
• Widgets in action domain have a natural interpretation in target domain
• Design so that there are intuitive correspondences
Collection of symptoms in patient
Diagnosis = organizing principle
Understand Organizing Principles
Some people think of this as a mental model that allows one to predicthow they will interpret other symptoms, or predict what they will expect
Discovering Organizing principle
• People try to impose order–
• 1 4 9 16 25 …
•
• p r e a m b l e -- fragment completion
Cow
Cow with outline
Predictability
• Once we know how someone organizes a set of elements we can predict their behavior better
Design Principle Three
• Facilitate construction of patterns or mental model or organizing principle so that we can increase the probability that people will make the right correspondence between a and A
Inconsistent organizing principles
Design Principle Four
• If subjects have more than one organizing principle ensure that they are consistent otherwise you cannot predict whether a means A or whether a means B
Design Principle Five
• Recognition is better than recall
• Give users visual choice rather than conceptual choice
Recognition vs. Recall
Better still
Design Principle Six
• Find effective ways of coordinating individuals
Putting it together in an environment• When is one environment better than another?
• Relative to a set of tasks
Question
How does redesigning an environment reshape routines?
•Artifacts
•Technology
•Cue structure •Spatial layout
How do we know they are better?
• Performance measures– Faster– Fewer Errors– Agents can do more complex things with them– Fewer serious errors – less variance
Speed Accuracy
Probabilityoferror
Time
1
0
Better
Speed Accuracy of Routine
Ri is better than Rj if it can be performed (dominates)
• More quickly without increase in expected error
• More error-free without decrease in speed
Complexity of Routines
Probabilityoferror
Time
1
0
Complexity of Routines
AcceptableError
AcceptableTime
Acceptable
C2
C1
C3
C4
Ri is better than Rj if
• Ri tasks are more complex
• can be performed in acceptable time and error rate
Variance of Routines
Variance
Time
1
0
Variance of Routines
AcceptableVariance
AcceptableTime
Acceptable
V2
V1
V3
V4
• Reduce the variance in output
• For each error rate in the speed accuracy curve the output will be
more standardized
• Narrowing the distribution of error size not the number of errors
Seriousness of error
Distribution of Errors
0
Recovery of Routines
Design Challenge:
• redesign the environment to lower recovery time
• redesign to facilitate vigilance and error detection
Learnability
100%
Degree of Mastery
Incentive
Late adopter
Average adopter
Early Adopter
$$$
0
More usable
Cost to Learn a New Technology
Starbucks
Five major Steps in espresso cafés
1. interact with client to specify order
3. take cashmake change offer receipt
2. communicate order
4. prepare the order
5. announce completion of orderqueue for client to collect
Espresso facts
• Called espresso because made for a specific customer and served immediately.
• A double espresso is – 1.5 - 2 ounce liquid extract – prepared from 14-17 grams of (medium) ground coffee – purified water of 88-95°C has been forced through– at 9-10 atmospheres of pressure – for a brew time of 22-28 seconds. – Crema should make up 10-30% of the beverage
• Cappuccino – A shot of espresso topped with equal parts of steamed and foamed milk
(wet cappuccino) – a shot of espresso topped with all foamed milk (dry cappuccino).
• Frothed milk should be 150°• Steamed milk should be 150° to 170°
Design Challenge
• Increase robustness of process– Reduce error
– Reduce variance of error
– Eliminate disastrous errors
• Process more drinks per hour – Routines and tech support
higher throughput
• Increase quality of service– Better interaction with customer
• Increase drink complexity• Routines are easier to master
• Error is always lurking– Noisy– Distractions– Surprises
• Interruptions, intrusions
• Multi-tasking, Task
Switching
• Multiple tasks in same
physical space
• High staff turnover
Costs to Minimize Problem Areas
Starbucks Revolutionary Technology
• Changes cognitive efficiency of whole system
• Minimizes costs in most areas
Technology of coordination
Form on cup
Why is it so remarkable?• Reduces errors
– Losing the order – Confusing one order with another
• Robust to interruption
– If barista forgets order just look it up
– Supports recoverability – increase state
• Tolerates breakdown
– If barista burned another picks it up off the floor
• Supports multi-tasking
– Locks info to object so more modular– Move along as in production like process
• Order complexity can go up
– Cup allows linear process
• Read, execute, read execute …
• Lowers cognitive demands
The End
Epistemic: Hard to reach states
TC WA E R S
Task: Call out all the words you can think of that can be made with some or all of these letters.
Hard to get more than 20 words in 5 min
Scrabble
Epistemic Action: Self-Cueing
TC
W
A
E R
S
Re-arrangement is allowed.
Close Spatial Coupling
• Humans are closely coupled in space and time to their environment
• This can be exploited:– re-arrange environmental resources to
• simplify judgment • computation
Preparing a Hand for Gin Rummy
• organization encodes current strategy
Player 1 Player 2
Segment E in a more congenial way
Count the dots.
which dot is the starting dot?
Adapt the world to our perceptual system
First and last
now stand out
Actions that compensate for Attention limitations
How many dots?
Requires coordination of rhythmic inner counting with hand movement
Re-arrange Distractors
reduce descriptive complexityreduce visual complexity