Computational Limits to Human Thinking in a Society

Report
Computational Limits to Human Thinking
in a Society With Too Much Information And Too Little Time
Raj Reddy
Carnegie Mellon University
Jan 23, 2015
Talk Presented at MSR India 10th Anniversary symposium
2
Anticipating the Next Disruptive Technology
Disruptive Past, Present and Future

1960s Mainframe



1970s Mini



One user with access to global shared resources (e. g. Browsers)
2000s Smart Phones, Tablets, and APPs



One computer, one user (e.g. Windows)
1990s Networked Computers – World Wide Web


One computer; Time sharing: tens of users; monitor instrumentation
1980s PCs and Personal Work Stations


One computer; Batch processing (e.g. JCL): thousands of users
One computer; Time sharing (e.g. TSO): hundreds of users
One User, Many Apps (e.g. IOS, Android)
Device-resident OS with APPs
2010-20s Invisible Wearables and Intention Aware Computing


One User, Thousands of Cogs (Cognition Amplifiers )
Personalized, Always On and Always Learning



Smart Phone for everyone on the planet (Language Divide, Literacy Divide)
Invisible Computing: Glass to Glassless - No Screen
Free Gigabit Connectivity
Paid by sales tax on Value Added Services
The NEXT BIG Thing will be a device that can be used by
everyone on the Planet
Human Limitations and Human Strengths
Lessons from Cognitive Science
 Human Limitations


People

Make Errors

Tend To Forget

Become Impatient

Tend To Favor Least Effort
Human Strengths

People





Learn From Experience
Tolerate Errorful, Ambiguous And Imprecise Input
Exploit Vast Amounts Of Knowledge
Communicate Using Speech And Natural Language
Wealth of Information and Scarcity of Attention
Human Limitations and Options
 To
Err is Human
to Catastrophic loss of data (“I didn’t mean to do
that”) and Unanticipated side-effects “What happened?”
 Caused by Sensory, Cognitive and Motor overload
 Lead

“Like being in a traffic jam”
 Current solutions: Partial Undo and ^C
 Future Options: Unlimited Undo and Anytime
 To
Abort
Forget is Human
non-expert occasional user can’t be expected to
remember the details
 Forgetting is caused by the loss of indexing structure
 Current solutions: Recognition vs. recall: Menus and
On-line manuals
 Future Options: Intelligent Help “How do I” and “What if”
Multi-media documentation (Why not YouTube Model?)
A
Human Limitations and Options (cont.)

To Be Impatient is Human

Waiting time to get the answer in interactive problem solving



Current solutions: Hour glasses, Wheels and Progress Bars



Long Start-up and Shutdown Times
Network operations: Unpredictable retrieval and browsing times
No idea how long it will take
Future Options: Updates in the background, Learn from
experience, Look ahead retrieval and computation
To Favor Least Effort is Human

Most people use a minimal subset of functionality in Word,
PowerPoint, etc.
 Most people avoid tasks requiring too much cognitive effort
e.g. PGP (too much work) and FTP (too complex)
 Causes: Principle of Least Effort
 Future opportunities: Autocomplete and Cognition Amplifiers
Human Limitations and Human Strengths
Lessons from Cognitive Science
 Human Limitations


People

Make Errors

Tend To Forget

Become Impatient

Tend To Favor Least Effort
Human Strengths

People





Learn From Experience
Tolerate Errorful, Ambiguous And Imprecise Input
Exploit Vast Amounts Of Knowledge
Communicate Using Speech And Natural Language
Wealth of Information and Scarcity of Attention
Video Illustrating Human Capabilities
That Computers Can’t Match, Yet
Missing Science of Human Capabilities
Lessons from Flash Forward Video

SILKy Interfaces
 Speech,
Image, Language and Knowledge
 Forgiving Interfaces: Tolerate error and
ambiguity
 “You Bet” --- “Did That Mean Yes?”
 Learning from Experience: Incremental nonintrusive knowledge acquisition
 “Darn” --- “Would you Care to Define It?”

Intelligent Help, Advice and Tutoring
“
Enthusiasm Detected --- Care to View Large
Screen?”

Intelligent Agents: Task Execution
 “Get
me Sam Gallagher, Try Every Number”
Human Limitations and Human Strengths
Lessons from Cognitive Science
 Human Limitations

People





Human Strengths

People





Make Errors
Tend To Forget
Become Impatient
Tend To Favor Least Effort
Learn From Experience
Tolerate Errorful, Ambiguous And Imprecise Input
Exploit Vast Amounts Of Knowledge
Communicate Using Speech And Natural Language
Wealth of Information and Scarcity of Attention
Too Much Information and
Too Little Time

Drowning In Too Much Information and Too Little
Time



As early as 1969, Simon discussed the problem of
“Wealth of Information and Scarcity of Attention”


Severely lacking tools to cope with the problem
Not a new problem, but the scale of the data and data
rate has changed dramatically
in Computers, Communications and Public Interest,
Martin Greenberger 1971
Spawned the Field of “Attention Economics”


White House Analogy
Need-to-Know Principle


Re-structure information: Organize, Summarize and Hyperlink
Overview, Zoom, Filter and Details on Demand
Key Principles of Information Overload

Bounded Rationality and Satisficing


When Faced with Information Overload, People Settle for
Good Enough Solutions in Preference to Optimal Solutions
When human thinking capabilities are overloaded, People
use strategies and tactics of “optimal-least-computation”


“Optimal-least-computation” is the study of approximate algorithms
to find the best possible solution given limited memory capacity,
limited time, or limited bandwidth
Wealth of Information and Scarcity of Attention

When faced with too much information and too little time,
Embrace the Need-to-Know-Principle, by creating


Summarization Tools
Information Structuring Tools

Use Techniques of Journalism and Cinematography
 Overview, Zoom, Filter and Details on Demand (Hyperlink the Details)


Autocomplete Actions
Non Intrusive Notifications
Cognition Amplifiers and Guardian Angels


Cognition Amplifiers and Guardian Angels are two families of
intelligent Agents that help with scarcity of attention problem
A Cognition Amplifier is a Personal Enduring Autonomic
Intelligent Agent that anticipates what you want to do and helps
you to do it with less effort






A Guardian Angel is a Personal Enduring Autonomic Intelligent
Agent that Discovers And Warns You About Unanticipated
Events That Could Impact Your Safety, Security, and Happiness




Buying and selling: Transact with multiple providers
Email: Filter spam, understand and respond to actionable email
News: Based on topic preferences, novelty, collaborative filtering
Banking: Monitor bank account, Credit Cards, Pay Bills
Travel: Flights, hotel, schedule disruptions, cancellations
Just-in-Time Warnings: Hurricanes, Tornados, Tsunami, Earthquakes,
Flooding, Extreme Weather
Accident Alerts and Rerouting; Transport Strikes
Scarcity of Essential Resources: Food, Energy, Water etc.
13
Thousands of Cogs and Gats as Personal Assistants?
Architecture of Cogs and Gats

Cogs and Gats are Mobile Apps (like a shadow) for Each
Person on the Planet




Unlike APPs of today, Cogs are Mass-customized to Each Individual
Designed to be Non-intrusive, Autonomic, and Device Independent
Always On, Always Present and Always Working
Always Learning


Cogs and Gats Monitor, Analyze and Learn From Experience;

Learn From Own Experience And Experience of Others



Enduring (life-long)
And share knowledge with a community of Cogs and Gats
Automated Discovery of Data and Information Sources
Data, suitably anonymized, can be used to learn appropriate
responses for every possible situation by




Learning preferences by observing user choices,
Learning by task similarity and user similarity,
Learning by error correction and
Simply learning thru clarification dialog ( does that mean yes? Would14 you
care to define it?).
Conclusion

Pay Attention to Human Strengths, Limitations, Needs,
and Egos

NEVER Intrusive





Ads on the right or bottom, no animation
Abolish Dialog Boxes - Replace with Notifications!
Replace “Live Tiles” with WIWTBDIWC
Glasses and Watches To Invisible Wearables
Scarcity of Attention Implies Creation of Tools for
Minimization of Human Effort



Use Cogs Autocomplete Tasks, not Just Text
Use Gats to Help with Unpredictable Events
Avoid Redesigns that Require Re-Learning

Ribbons in Word
 Win 7 to Win 8
 $10 Rule: Are You Willing to Pay Every Customer $10 Every Time You Radically
Change Look And Feel Of Your Software Product?

One Size Fits All Software to Personalized Modular Software

Personalize and Mass-customize everything
Backup Slides
Vision: Cognition Amplifiers



Maxim is “Activity Monitoring” enables “Intention Awareness”
enables “Cognition Amplifiers” enables “80%+ Automation”
Intention Awareness can be Inferred from Prior Activity
Patterns of the User (and Others with Similar Profile)
Intention Awareness enables Creation of Personal Autonomic
Cognition Amplifiers (Cogs)




Cogs are created by continuously monitoring multimodal sensors on
modern smart phones, and discovering and interpreting user activities
and intentions thru non-intrusive learning technologies.
The Activity Monitoring System non-intrusively adapts and creates
personalized Cogs using Cog Templates from the Platform providers
and Cog developers
Cogs enable Automation of Much of the Human Work and
Daily Personal Activities Than One-size-fits-all Apps
No matter what the user is trying to do (not just search)
intention aware computing and Cogs will enable the user to do
the tasks faster, with fewer errors, and less cognitive load.
Cogs Enable the Digital Bill of Rights

Every person on the planet with a smart phone
(literate and non-literate alike) will be able to access
and benefit from non-intrusive Cogs that learn with
experience. Cogs minimize the human effort in
performing the daily tasks by

Getting the right information


To the right people


e.g. machine translation
In the right medium


e.g. just-in-time (task modeling, planning)
In the right language


e.g. categorizing, routing
At the right time


e.g. search engines
e.g. using speech for language divide and literacy divide
With the right level of detail (granularity)

e.g. summarization
18
Challenges

Privacy & Security



In Cog We Trust




How does user know Cog will do the right thing?
Undoable or minimal-cost actions  autopilot
Irreversible actions  implicit/explicit confirmation
Cog-World Interface


Guarantee privacy of individual’s data & Cogs
Allow for opt-in sharing (e.g. preferences, recs)
Too many external entities  developer’s APIs
Cog Template Library & Examples

For Cog developers, customizers, aftermarket
Possible Topics for MSR Talk Jan 23

Y3K: Collecting and Archiving Born-Digital Movies,
Music, Newspapers and Books for a Thousand Years
 Transforming Education: Mastery Education where
Time is Variable and Outcome is an A
 Digital Elections That Are Fair and Accurate
 Computational Limits to Human Thinking

Wealth of Information and Scarcity of Attention

similar documents