Future of ai 20180719 v9

The Future of AI:
Open Technology, Innovation,
and Service System Evolution
Jim Spohrer (IBM)
UCLA BIT Conference, Los Angeles, CA July 19, 2018
http://slideshare.net/spohrer/future-of-ai-20180619-v9
7/19/2018 IBM Code #OpenTechAI 1

Today’s talk
• AI at the peak of the hype cycle
• What’s really going on?
• Part 1: Solving AI
• Roadmap and Implications
• Open Technologies, Innovation
• Part 2: Service System Evolution
• Better Building Blocks
• Trust and Resilience

7/19/2018
© IBM UPWard 2016
3
AI (Artificial Intelligence) is popular again… you see it mentioned on billboards in SF
However, pattern recognition does not equal AI
Deep learning works if you have lots of data and compute power
We finally have lots of data and compute power – hurray!!!
So finally, deep learning for pattern recognition is working pretty well
However, AI is more than deep learning for pattern recognition…
AI requires commonsense reasoning – that will take another 5-10 years of research
How do we know this? Look at the AI leaderboards – we will get to that…

Smartphones pass entrance exams? When?
7/19/2018 (c) IBM 2017, Cognitive Opentech Group 4
… when will
your smartphone
be able to take and
pass any online
course? And then
be your coach, so
you can pass too?

IBM-MIT $240M
over 10 year AI mission

7/19/2018 Understanding Cognitive Systems 6

Questions
• What is the timeline for solving AI and IA?
• Who are the leaders driving AI progress?
• What will the biggest benefits from AI be?
• What are the biggest risks associated with AI, and are they real?
• What other technologies may have a bigger impact than AI?
• What are the implications for stakeholders?
• How should we prepare to get the benefits and avoid the risks?

Timeline: Short History
7/19/2018
© IBM Cognitive Opentech Group (COG)
8
Dota 2
“Deep Learning” for
“AI Pattern Recognition”
depends on massive
amounts of “labeled data”
and computing power
available since ~2012;
Labeled data is simply
input and output pairs,
such as a sound and word,
or image and word, or
English sentence and French
sentence, or road scene
and car control settings –
labeled data means having
both input and output data
in massive quantities.
For example, 100K images
of skin, half with skin
cancer and half without to
learn to recognize presence
of skin cancer.

Timeline: Every 20 years,
compute costs are down by 1000x
• Cost of Digital Workers
• Moore’s Law can be thought of as
lowering costs by a factor of a…
• Thousand times lower
in 20 years
• Million times lower
in 40 years
• Billion times lower
in 60 years
• Smarter Tools (Terascale)
• Terascale (2017) = $3K
• Terascale (2020) = ~$1K
• Narrow Worker (Petascale)
• Recognition (Fast)
• Petascale (2040) = ~$1K
• Broad Worker (Exascale)
• Reasoning (Slow)
• Exascale (2060) = ~$1K
97/19/2018 (c) IBM 2017, Cognitive Opentech Group
2080204020001960
$1K
$1M
$1B
$1T
206020201980
+/- 10 years
$1
Person Average
Annual Salary
(Living Income)
Super Computer
Cost
Mainframe Cost
Smartphone Cost
T
P
E
T P E
AI Progress on Open Leaderboards
Benchmark Roadmap to solve AI/IA

Timeline: GDP/Employee
(Source)
Lower compute costs translate into increasing productivity and GDP/employees for nations
Increasing productivity and GDP/employees should translate into wealthier citizens
AI Progress on Open Leaderboards
Benchmark Roadmap to solve AI/IA

Timeline: Leaderboards Framework
AI Progress on Open Leaderboards - Benchmark Roadmap
Perceive World Develop Cognition Build Relationships Fill Roles
Pattern
recognition
Video
understanding
Memory Reasoning Social
interactions
Fluent
conversation
Assistant &
Collaborator
Coach &
Mediator
Speech Actions Declarative Deduction Scripts Speech Acts Tasks Institutions
Chime Thumos SQuAD SAT ROC Story ConvAI
Images Context Episodic Induction Plans Intentions Summarizatio
n
Values
ImageNet VQA DSTC RALI General-AI
Translation Narration Dynamic Abductive Goals Cultures Debate Negotiation
WMT DeepVideo Alexa Prize ICCMA AT
Learning from Labeled Training Data and Searching (Optimization)
Learning by Watching and Reading (Education)
Learning by Doing and being Responsible (Exploration)
2015 2018 2021 2024 2027 2030 2033 2036
Which experts would be really surprised if it takes less time… and which experts really surprised if it takes longer?
Approx.
Year
Human
Level ->

Leaders and Leaderboards:
Who is winning
• How to Measure Leadership?
• Publications or Patents
• Regions China vs USA vs EU vs ROW
• Companies Microsoft vs Google vs IBM
• How to Measure Progress?
• One capability or all leaderboards?
• SQuAD – Question Answering
• EFF Measuring AI Progress

AI Benefits
• Access to expertise
• “Insanely great” labor productivity for trusted service providers
• Digital workers for healthcare, education, finance, etc.
• Better choices
• ”Insanely great” collaborations with others on what matters most
• AI for IA = Augmented Intelligence and higher value co-creation interactions

AI Risks
• Job Loss
• Shorter term
bigger risk
= de-skilling
• Super-intelligence
• Shorter term
bigger risk
= bad actors
https://maliciousaireport.com/
https://www.wsj.com/articles/automation-makes-us-dumb-1416589342

Huang and Rust, JSR
• test

Robots by Country
• Industrial robots per
10,000 people by country
7/19/2018 IBM #OpenTechAI 16

Leaderboard - rankings
• Korea leads

Learderboard – rankings 2
• China is below world average

Brian Arthur - Economist
• The term “technological unemployment” is from John Maynard Keynes’s 1930 lecture,
“Economic possibilities for our grandchildren,” where he predicted that in the future, around
2030, the production problem would be solved and there would be enough for everyone, but
machines (robots, he thought) would cause “technological unemployment.” There would be
plenty to go around, but the means of getting a share in it, jobs, might be scarce. We are not quite
at 2030, but I believe we have reached the “Keynes point,” where indeed enough is produced by
the economy, both physical and virtual, for all of us. (If total US household income of $8.495
trillion were shared by America’s 116 million households, each would earn $73,000, enough for
a decent middle-class life.) And we have reached a point where technological unemployment is
becoming a reality. The problem in this new phase we’ve entered is not quite jobs, it is access to
what’s produced. Jobs have been the main means of access for only 200 or 300 years. Before
that, farm labor, small craft workshops, voluntary piecework, or inherited wealth provided access.
Now access needs to change again. However this happens, we have entered a different phase for
the economy, a new era where production matters less and what matters more is access to that
production: distribution, in other words—who gets what and how they get it. We have entered
the distributive era.

Other Technologies: Bigger impact? Yes.
• Augmented Reality (AR)/
Virtual Reality (VR)
• Game worlds
grow-up
• Blockchain/
Security Systems
• Trust and security
immutable
• Advanced Materials/
Energy Systems
• Manufacturing as cheap,
local recycling service
(utility fog, artificial leaf, etc.)

Artificial Leaf
• Daniel Nocera, a professor of energy
science at Harvard who pioneered the
use of artificial photosynthesis, says that
he and his colleague Pamela Silver have
devised a system that completes the
process of making liquid fuel from
sunlight, carbon dioxide, and water. And
they’ve done it at an efficiency of 10
percent, using pure carbon dioxide—in
other words, one-tenth of the energy in
sunlight is captured and turned into fuel.
That is much higher than natural
photosynthesis, which converts about 1
percent of solar energy into the
carbohydrates used by plants, and it
could be a milestone in the shift away
from fossil fuels. The new system is
described in a new paper in Science.

Food from Air
• Although the technology is in its infancy,
researchers hope the "protein reactor"
could become a household item.
• Juha-Pekka Pitkänen, a scientist at VTT,
said: "In practice, all the raw materials
are available from the air. In the future,
the technology can be transported to,
for instance, deserts and other areas
facing famine.
• "One possible alternative is a home
reactor, a type of domestic appliance
that the consumer can use to produce
the needed protein."
• According to the researchers, the
process of creating food from electricity
can be nearly 10 times as energy
efficient as photosynthesis, the process
used by plants.

Exoskeletons for Elderly
• A walker is a “very cost-effective”
solution for people with limited
mobility, but “it completely
disempowers, removes dignity,
removes freedom, and causes a
whole host of other psychological
problems,” SRI Ventures president
Manish Kothari says. “Superflex’s
goal is to remove all of those areas
that cause psychological-type
encumbrances and, ultimately,
redignify the individual."

Stakeholders
• Individuals
• Families
• Businesses and
other Organizations
• Industry Groups
• Regional
Governments:
• Cities
• States
• Nations

“The best way to predict the future is to inspire the
next generation of students to build it better”
Digital Natives Transportation Water Manufacturing
Energy Construction ICT Retail
Finance Healthcare Education Government

Be Prepared
• Understand open AI code + data +
models + stacks + communities
• Leaderboards
• Ethical conduct
• Learn 3 R’s of IBM’s Cognitive
Opentech Group (COG)
• Read arXiv
• Redo with Github
• Report with Jupyter notebooks on
DSX and/or leaderboards
• Improve your team’s skills of rapidly
rebuilding from scratch
• Build your open code eminence
• Understand open innovation
• Communities + Leaderboards
1972 used
Punch cards
2016 used
IBM Watson
Open APIs to win…

7/19/2018 27
1955 1975 1995 2015 2035 2055
Better Building Blocks

Courses
• 2015
• “How to build a cognitive system for Q&A task.”
• 9 months to 40% question answering accuracy
• 1-2 years for 90% accuracy, which questions to reject
• 2025
• “How to use a cognitive system to be a better professional X.”
• Tools to build a student level Q&A from textbook in 1 week
• 2035
• “How to use your cognitive mediator to build a startup.”
• Tools to build faculty level Q&A for textbook in one day
• Cognitive mediator knows a person better than they know themselves
• 2055
• “How to manage your workforce of digital workers.”
• Most people have 100 digital workers.
7/19/2018 28
Take free online cognitive classes today at cognitiveclass.ai

Prepare for AI Future
• Do you have a GitHub account? Get it.
• Yes: proceed
• No: sign up
• Do you program? Either OK, partnering is best.
• Yes: Learn and do 3 R’s (read, redo, report)
• Github master: Code, Content (Data), Community (IBM Code can help)
• No: Learn to read and execute code with partner (T2T)
• Do you have favorite AI leaderboards?
• Yes: Learn and do 3 R’s (read, redo, report advances)
• Kaggle master: Combine top decorrelated solution, new solution
• No: Find a mentor with favorites, do together
• Are you AI prepared? Do you know/do data, models, solutions?
• Yes: Find favorite leaderboards you can do 3 R’s for today
• Figure-Eight master: Labeled data that matters most
• No: Wait until one model: one model that can do them all
• Then rapidly rebuild in least time, energy (“zorch”), data, code
7/19/2018
© IBM Cognitive Opentech Group 2018
29
1. Where do we get labeled data?
We create it: Figure Eight,
Mechanical Turk, etc.
2. External/internal challenge?
10M minutes from birth to adult
2M minutes from novice to expert
Not just external states, but
internal states are data as well…
The challenge of data for AI models
3. AI models as ”data” instruction set
Computer’s have instruction sets
Arithmetic, Logic, etc.
Models are becoming instructions
Models are data/experience

Step Comment
GitHub Get an account and read the guide
Learn 3 R's - Read, Redo, Report Read (Medium/arXiv), Redo (GitHub), Report (Jupyter Notebook)
Kaggle Compete in a Kaggle competition
Leaderboards Compete to advance AI progress
Figure Eight Generate a set of labeled data (also Mechanical Turk)
Design New Challenges build an AI system that can take and pass any online course, then
switch to tutor-mode and help you pass
Open Source Guide Establish open source culture in your organization

7/19/2018
© IBM 2015, IBM Upward University Programs Worldwide
accelerating regional development
34
I have…
Have you noticed how the building blocks just
keep getting better?

Learning to program:
My first program
7/19/2018
35
Early Computer Science Class:
Watson Center at Columbia 1945
Jim Spohrer’s
First Program 1972

7/19/2018
© IBM UPWard 2016
36
Fast Forward 2016:
Consider this…

Cupertino Teens
• IBM Watson on Bluemix
AI for NLP
entity identification

Build: 10 million minutes of experience

Build: 2 million minutes of experience

Hardware < Software < Data < Experience < Transformation
Value migrates
Pine & Gilmore (1999)
Transformation
Roy et al (2006)
Data
Osati (2014)
Experience
Life Log

Types: Progression of models
Models = instruction set of future
Task & World Model/
Planning & Decisions
Self Model/
Capacity & Limits
User Model/
Episodic Memory
Institutions Model/
Trust & Social Acts
Tool + - - -
Assistant ++ + - -
Collaborator +++ ++ + -
Coach ++++ +++ ++ +
Mediator +++++ ++++ +++ ++
Cognitive
Tool
Cognitive
Assistant
Cognitive
Collaborator
Cognitive
Coach
Cognitive
Mediator

7/19/2018
46
Cognitive Mediators
for all people in all roles

Occupations = Many Tasks
7/19/2018
47

Watson Discovery Advisor
7/19/2018
48
Simonite, T. 2014. Software Mines Science Papers to Make New Discoveries. MIT. November 25, 2014.
URL: http://m.technologyreview.com/news/520461/software-mines-science-papers-to-make-new-discoveries/

Trust: Two Communities
Service
Science
OpenTech
AI
Trust:
Value Co-Creation,
Transdisciplinary
Trust:
Ethical, Safe, Explainable,
Open Communities
Special Issue
AI Magazine?
Handbook of
OpenTech AI?

Resilience:
Rapidly Rebuilding From Scratch
• Dartnell L (2012) The Knowledge: How to
Rebuild Civilization in the Aftermath of a
Cataclysm. Westminster London: Penguin
Books.

Future of ai 20180719 v9

Recomendados

Recomendados

Mais conteúdo relacionado

Mais procurados

Mais procurados (20)

Semelhante a Future of ai 20180719 v9

Semelhante a Future of ai 20180719 v9 (20)

Mais de ISSIP

Mais de ISSIP (20)

Último

Último (20)

Future of ai 20180719 v9

Notas do Editor