Talk presented at the Analytics Frontiers Conference in Charlotte on March 21. The presentation evaluates opportunities and risks of AI and how consumers, businesses, society and governments can mitigate some of the risks.
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Â
Responsible AI
1. PwC AI Lab | 1
Responsible AI â Role of Consumers,
Businesses, and Governments
Dr. Anand S. Rao
Global Artificial Intelligence Lead
2. PwC AI Lab | 2
Todayâs discussion
Enterprise AI Through Four
Lenses
Enterprise AI Case Studies
Risks of AI
01
02
03
Responsible AI04
3. PwC AI Lab | 3
01
Enterprise AI Through
Four Lenses
4. PwC AI Lab | 4
AI as Sense-Think-Act
Sense
Artificial Intelligence is
becoming ubiquitous
intelligence with the ability to
see, hear, speak, smell, feel,
understand gestures, interface
with your brain, and dream
Think
AI is helping us do tasks faster,
better and cheaper â Automated
Intelligence; helping us make
better decisions â Assisted &
Augmented Intelligence, or even
taking over what we do â
Autonomous Intelligence
Act
Artificial Intelligence is
equaling or surpassing
humans in a number of other
tasks â playing games, driving
cars, recommendations
(movies, books, finance,
research), etc.
5. PwC AI Lab | 5
Statistics Econometrics Optimization
Complexity
Theory
Computer
Science
Game
Theory
FOUNDATION
LAYER
Sense Think Act
⢠Robotic process
automation
⢠Deep question &
answering
⢠Machine translation
⢠Collaborative systems
⢠Adaptive systems
⢠Knowledge &
representation
⢠Planning &
scheduling
⢠Reasoning
⢠Machine Learning
⢠Deep Learning
⢠Natural language
⢠Audio & speech
⢠Machine vision
⢠Navigation
⢠Visualization
AI that can sense⌠AI that can think⌠AI that can actâŚ
Hear
See
Speak
Feel
Understand Perceive
PlanAssist
Physical
Creative
Cognitive
Reactive
More FormallyâŚ
6. PwC AI Lab | 6
Business Lens
Metrics & Value Chain
Intelligence Lens
Automated, Assisted,
Augmented & Autonomous
Data Lens
Structured vs Unstructured
Available vs Augmented
Technology Lens
Techniques, Tools & Platforms
Four Lenses of Artificial Intelligence
7. PwC AI Lab | 7
Business Lens: Metrics & Value Chain
Operations & Development
Product
Development
Service &
Support
Operations
Outbound Logistics
Sales &
Distribution
Customers &
Marketing
Strategy &
Growth
Supply Chain &
Procurement
Finance, HR,
Planning
Inbound Logistics
How will we ensure our
product supply is meeting
demand?
VP, Supply Chain
How can we engage with our
customers to enhance their
experience?
Director, Marketing
How can we grow our market
share and which markets to
enter, exit or expand?
Director, Strategy
How do we innovate and
introduce new products and
services?
Director, Products
How do we increase customer
satisfaction and retain more
customers?
Director, Service
How can we reach more
customers and price our
products to increase sales?
Director, Sales
How can we increase
efficiency and effectiveness of
our operations?
Director, Operations
How can we get a better
return on our talent, capital,
and assets?
Director, Finance & HR
⢠Market Share
⢠Customer Experience
⢠Acquisition Rate
⢠Innovation Rate
⢠Operational Efficiency
⢠Customer Satisfaction
⢠Talent Retention
⢠Inventory Turn
Over 300+ AI Use Cases Across 8 Sectors â Sizing the Prize
8. PwC AI Lab | 8
Intelligence Lens: Four Types of Enterprise AI
No human in the loopHuman in the loop
Hardwired /
specific
systems
Adaptive
systems
Automated Intelligence
1
Assisted Intelligence
2
Augmented Intelligence
3
Autonomous Intelligence
4
+
9. PwC AI Lab | 9
Data Lens: Four Types of Data
Structured
AvailableAugmented
Unstructured
10. PwC AI Lab | 10
What is Artificial Intelligence?
Artificial Intelligence can be defined as the theory and development of systems that can continuously sense its environment, think,
make decisions, and take actions that influence the environment to achieve its goals.
Technology Lens: AI Techniques
Machine
Vision
Natural
Language
Audio &
Speech
Navigation Visualization
SENSORY
LAYER
Knowledge
Representation
Reasoning
Planning &
Scheduling
Machine
Learning
Deep
Learning
COGNITIVE
LAYER
Robotic
Process
Automation
Deep
Question &
Answering
Machine
Translation
Collaborative
Systems
Adaptive
Systems
BEHAVIORAL
LAYER
Statistics Econometrics Optimization
Complexity
Theory
Computer
Science
Game
Theory
FOUNDATIONAL
LAYER
12. PwC AI Lab | 12
Case 1:
Global
Pharmaceutical
Case 2:
Construction
Company
Case 3:
Automotive
Manufacturer
Case 4:
Digital
Advisor
13. PwC AI Lab | 13
Global Pharmaceuticals
Extracting adverse drug
interaction from clinician
notes, social media, and
medical literature to
enhance productivity and
effectiveness (96%
accuracy)
14. PwC AI Lab | 14
Adverse Event Pipeline using NLP Toolkit
15. PwC AI Lab | 15
Deep Learning of Latent Relationships
Word2Vec
is able to
show the
relationship
between
Sneezing
and Anti-
histamine.
17. PwC AI Lab | 17
Digital Advisor
Gamification of Strategy
resulted in the
development of a digital
advisor that simulates
household level (128
million) financial data
into the future to enhance
financial wellness
PwC AI Lab | 17
18. PwC AI Lab | 18
$ecure is a digital advice and financial wellness toolkit, that enables a
differentiated digital advice experience for customers in a cost-efficient manner
01
02
03
Synthetic dataset of 1.28M U.S.
households with 4000+ data points
Personalized customer
experience by life stage
Agent-based model to
project household finances
01
02
03
âHouseholds Like Youâ
benchmarking for consumer
education/data augmentation
Holistic retirement planning
using advanced scenario
analysis
Intuitive planning tools and
what-if analysis that demystify
the planning process
Core Components Key Differentiators
19. PwC AI Lab | 19
Key Differentiator #1: âHouseholds Like Yoursâ matching to enable
benchmarking/data augmentation
Clientâs Name
* Illustrative
John Doe Smith
Household Zip Code 75220
Gender Male
Marital Status Married
# Dependents 2
Annual Base Income
Total Assets
Tell us a little about yourself âŚ
Weâll benchmark you against peer households âŚ
$1,650 $1,750
$765
$650
$885
$1,100
Your Household Households Like
Yours
Household Balance Sheet ($
â000)
Total Assets Liabilities Net Worth
$365 $350
$220
$165$145
$185
Your Household Households Like Yours
Household Income Statement ($
â000)
Income Expenses Surplus/Deficit
⌠and help you augment missing/incomplete data
Co-Clientâs Name Mary Jo Smith
Co-Clientâs Age 45
Age 47
Co-Clientâs Annual Base Income i
Households
Like Yours:
$175K - $195K
PwC Synthetic
Dataset
âHouseholds Like Youâ estimates increase in accuracy as more data points become available
20. PwC AI Lab | 20
Key Differentiator #2: Retirement Planning Evolved - Holistic cross-silo perspective
on current and future assets and liabilities with advanced scenario analysis
20
Rather than having to monitor
multiple metrics, users only track
fundedness, which takes stock
of current and future assets and
liabilities
Others: Incomplete retirement readiness representation
vs.
Picture source: Betterment.com
Limited guidance on how much
to save, due to absence of the
liabilities side of the equation
Basic scenario analysis focused
primarily on asset growth across
multiple economic environments
* Illustrative
0%
20%
40%
60%
80%
100%
120%
140%
47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65
Fundedness(%) Age â Head Of Household (J. Smith)
Projected Fundedness To Retirement
Pessimistic Expected
Emergency
Healthcare
(Client)
College Tuition
(Elder Child)
Constraine
d
OverfundedUnderfunded
Long-Term Care
(Spouse)
In addition to macroeconomic factors, $ecure features sophisticated
scenario analysis that captures significant life events as well
$ecure: Holistic retirement readiness monitoring
21. PwC AI Lab | 21
PwCâs Digital Services
Six success factors to derive maximum benefits from
artificial intelligence
Start from business
decisions
01
Demonstrate value
through pilots before
scaling
02
Blend intuition and
data-driven insights
03
Address âbig dataâ â
donât forget âleanâ data
04
Fail forward â
test and learn culture
05
Focus on Responsible
AI from the start
06
23. PwC AI Lab | 23
Risks of AI
âThe automation of factories has
already decimated jobs in
traditional manufacturing, and the
rise of artificial intelligence is likely
to extend this job destruction deep
into the middle classes, with only
the most caring, creative or
supervisory roles remaining.â â
Stephen Hawking
âIâm increasingly inclined to think
that there should be some
regulatory oversight, maybe
at the national and international
level, just to make sure that we
donât do something
very foolish.â â Elon Musk
24. PwC AI Lab | 24
Recent Fatality from a autonomous vehicle
It happened at 10 p.m. in Tempe, Arizona, where ride-hailing company Uber had been picking up
passengers in autonomous vehicles for more than a year. Elaine Herzberg, 49, was walking her bicycle
down a four-lane road and was starting to cross when the gray Volvo, operated by Uber, hit her at
about 40 mph, according to local police.
25. PwC AI Lab | 25
Control
ď Risk of AI going ârogueâ
ď Inability to control
malevolent AI
ď Swarm drones
Performance
ď Risk of Errors
ď Risk of Bias
ď Risk of Opaqueness
ď Risk of stability of
performance
ď Lack of feedback process
Security
ď Cyber intrusion risks
ď Privacy risks
ď Open source software risks
ď Digital, Physical, Political
security
Robust AI: Performance, security and control risks
26. PwC AI Lab | 26
Software Risks: Bias Risk â How can we avoid data bias in
recommendations?
COMPAS, a system
used by US Judges
to forecast which
criminals are likely
to reoffend was
biased. It concluded
that almost âblacks
are almost twice as
likely as whites to be
labeled a higher risk
but not actually re-
offend.â
COMPAS
27. PwC AI Lab | 27
Security Risks: Cyber Intrusion riskâ How can we prevent
âcyberâ intrusion of automated or electronic vehicles?
After hackers Charlie
Miller and Chris Valasek
hacked the Jeep Cherokee
and stopped the car off
the highway, Chrysler
issued a 1.4 million
vehicle recall and mailed
USB drives with software
updates to affected
drivers.
Simulated âCyber Intrusionâ
28. PwC AI Lab | 28
Control Risks: âRogueâ riskâ How can we ensure that an AI
designed with benevolent intent does not go ârogueâ?
Tay, a Microsoft
chatbot, released to
interact with the
public began
tweeting racist and
inflammatory
remarks in under 24
hours and had to be
decommissioned.
Tay Chatbot
29. PwC AI Lab | 29
Societal
ď Risk of Autonomous
Weapons proliferation
ď Risk of âintelligence divideâ
Ethical
ď âLack of Valuesâ risk
ď Value Alignment risk
ď Goal Alignment risk
Economic
ď Job displacement risks
ď âWinner-takes-allâ
concentration of power risk
ď Liability risk
Beneficial AI: Ethical, economic, and societal risks
30. PwC AI Lab | 30
Ethical Risks â How can a autonomous vehicle learn the
âvalueâ of human life?
Should the AV continue and
(definitely) kill one pedestrian
who is disobeying the law?
Or should the AV swerve and
(potentially) kill two pedestrians
who are obeying the law?
MITâs Moral Machine
MITâs Moral
Machine allows
users to select
scenarios to
understand
human ethics to
determine what
the âmachine
ethicsâ should be
31. PwC AI Lab | 31
Economic Risks â How can we manage job losses due to
automation from becoming a major economic issue?
Automation Job Losses
A number of studies
are predicting job
losses, up to 50% or
more, from automation
in different sectors in
different geographies.
32. PwC AI Lab | 32
Societal Risks â How can we ban the proliferation of
autonomous weapons designed to âkillâ?
Autonom0us Weapons Proliferation
Source: Why we should really ban Autonomous Weapons,
Stuart Russell, Max Tegmark, and Toby Walsh, August 3,
IEEE Spectrum, 2015
34. PwC AI Lab | 34
Responsible Artificial Intelligence
We define Responsible Artificial Intelligence, as the combination of building Robust AI systems that
will engender âtrustâ in todayâs AI system as well as work towards the development of AI that will be beneficial
to society today and in the future.
Robust Artificial Intelligence, is
concerned with the verification,
validation, security and control of AI
systems
Beneficial Artificial Intelligence,
is concerned with maximizing the
social benefit of AI
⢠Reduce or eliminate software risks
⢠Reduce or eliminate security risks
⢠Reduce or eliminate control risks
⢠Reduce or eliminate economic risks
⢠Reduce or eliminate societal risks
⢠Reduce or eliminate ethical risks
35. PwC New Services | 35
Robust Artificial Intelligence
⢠Verification: Modular agent-based architectures; verifiable
substrates of operating systems and platforms; adaptive
control theory; and deep learning theory
⢠Validation: Computational models of ethical reasoning; goal
stability; reasoning under uncertainty; and bounded
rationality
⢠Security: Software, hardware, and psychological
containment; tripwires â detection and response; detecting
intent to deceive.
⢠Control: Corrigibility and domesticity; safe and unsafe agent
architectures.
Research Priorities
Robust Artificial Intelligence, is concerned with the verification, validation, security and control of AI
systems
1. Define business use case criticality and vulnerability
2. Select interpretability requirements in terms of
explainability, transparency, and provability
3. Design and build models while performing business,
performance, and acceptance trade-offs
4. Monitor ongoing model performance and governance
Business Implications
Verification
Did I build the system right?
⢠How to prove that a system
satisfies certain desired formal
properties?
Validation
Did I build the right system?
⢠How to ensure that a system that
meets its formal requirements does
not have unwanted behaviors and
consequences?
Security
How do I secure the system?
⢠How to prevent intentional
manipulation by unauthorized
parties?
Control
How do I control the system?
⢠How to enable meaningful human
control over an AI system after it
begins to operate?
⢠Determine critical and vulnerable sectors (e.g.,
autonomous vehicles, healthcare systems, safety critical
infrastructure, airspace) that require explicit regulations
⢠Facilitate industry, research, and government discussions
on Robust AI
Regulatory Implications
36. PwC AI Lab | 36
Our Robust AI framework helps businesses design, build, and
deploy AI systems that can be âtrustedâ
PwCâs Robust AI Framework
AI
Tradeoffs
Monitoring of data for model training to
ensure data does not skew model
performance
Determining artificial intelligence
algorithm accuracy as required by
business use case
Ensuring algorithm decisions are
explainable to end user in such a way the
user trusts the predictions for the given
use case
Determining the appropriate scope and
system requirements for an artificial
intelligence application
Identification of potential threats that
may undermine or shift algorithm
decision making
Requiring artificial intelligence
algorithms to function reliably and
predictably
37. PwC AI Lab | 37
Explainable AI to improve customer experience
Source: Gunning, DARPA I/2O, 2017
38. PwC AI Lab | 38
National AI Strategies
USA
ď Unmanned Aircraft Systems (UAS) (Oct 2017)
ď Big Data: A Report on Algorithmic Systems,
Opportunity, and Civil Rights (May 2016)
ď AI, Automation, and the Economy (Dec 2016)
ď Preparing for the Future of Artificial Intelligence (Oct
2016)
China
ď Next generation AI Development Plan (July 2017)
with key focus areas and key guarantee measures
addressing the Science & Technology as well as
regulations and competitive policies
United Kingdom
ď§ Growing the Artificial Intelligence Industry in the
UK (October 2017): Recommendations to
o Improve access to data
o Maximize UK AI Research
o Improve supply of skills
o Support uptake of AI
Germany
ď§ Ethics Commission: Automated and Connected
Driving (June 2017)
Japan
ď Artificial Intelligence Technology Strategy (March
2017)
ď New Robot Strategy (February 2015)
39. PwC New Services | 39
Beneficial Artificial Intelligence
⢠Economic Modeling of AI Adoption: Automation and AI
impact - whom, when, and by how much; valuing knowledge
and insights
⢠Ethics research: Value alignment, AI rights, autonomous
weapon systems ban and/or control
⢠Wealth redistribution: Universal Basic Income and
alternative policy assessment and experimentation
Research Priorities
Beneficial Artificial Intelligence, is concerned with maximizing the social benefit of Artificial
Intelligence
Business Implications
Economic Issues
How do we estimate benefits?
⢠How do we calculate the economic
impact of automation and AI?
Social Issues
How do we share benefits?
⢠What social policies (e.g., universal
basic income) to distribute the
wealth generated by automation
and AI?
Legal Issues
What rules & regulations do we
need?
⢠What laws do we need to pass to
protect people, life, and property?
Ethical Issues
How do we ensure the AI is
used for social good?
⢠What values should autonomous
systems have and who decides the
values?
⢠Liabilities and Laws for Autonomous systems:
Autonomous car liability; drone air space regulations;
road traffic rules
⢠Policy Formulation: Taxation, education, social
security, energy and transportation, competition, privacy,
cyber , autonomous weapons etc.
Regulatory Implications
⢠Future of Work: Impact assessment of automation and AI;
change management; training and re-skilling workforce;
creation of new roles; community participation
⢠Non-Profit Groups: Making the case for policy changes at
the national (e.g., drone rules) and international levels (e.g.,
autonomous weapons ban)
40. PwC AI Lab | 40
Reskilling
⢠Workforce reskilling
⢠Digital fitness
⢠University education
Key Elements of AI Strategy
Basic AI R&D
⢠Moonshot projects
⢠University funding
⢠Business incentives
Business Protection
⢠Local companies
⢠Specific industry sectors
⢠Algorithmic governance
Specialized AI Tech.
⢠Drones
⢠Autonomous vehicles
⢠Service robots
Consumer Protection
⢠Data security
⢠Income security
⢠Digital anonymity
Ethics
⢠Citizen monitoring
⢠Autonomous weapons
⢠Beneficial use of AI
42. PwC AI Lab | 42
PwCâs Digital Services
Thank you.
Š 2018 PwC. All rights reserved. Not for further distribution without the permission of PwC. âPwCâ refers to the network of
member firms of PricewaterhouseCoopers International Limited (PwCIL), or, as the context requires, individual member firms of
the PwC network. Each member firm is a separate legal entity and does not act as agent of PwCIL or any other member firm.
PwCIL does not provide any services to clients. PwCIL is not responsible or liable for the acts or omissions of any of its member
firms nor can it control the exercise of their professional judgment or bind them in any way. No member firm is responsible or
liable for the acts or omissions of any other member firm nor can it control the exercise of another member firmâs professional
judgment or bind another member firm or PwCIL in any way.
Dr. Anand S. Rao
Global AI Lead
anand.s.rao@pwc.com
@AnandSRao