This presentation was given at the OECD Network of AI Specialists (ONE) held in Paris on February 26 and 27. It was presented at the joint session of ONE and the Digital Council. It covers some of the key trends and developments in AI including operationalizing AI, responsible AI and National AI Strategies
1. AI Developments
& Trends
How AI is automating
transforming, and
disrupting business
Dr. Anand S. Rao, Global AI Lead, PwC
2. PwC Experience Center
THE METHOD
2
TWO PATHS TO AI
Enterprises are
realizing the value from
digitization to AI along
two distinct but related
paths—to enhance
productivity, increase
profits and enhance
experience.
Digitization
Productivit
y
Experienc
e
Profit
s
Revenue
s
Data (Volume, Velocity, Variety, Veracity, Value)
Artificial Intelligence
Simplification
Standardization
Automation
Cognification
Analytics
Productivity Experience ProfitsRevenues
Automation Path Analytics Path
Personalization
3. PwC Experience Center
THE METHOD
3
WHAT IS AI?
AI is the theory and
development of
systems that sense the
environment, make
decisions, and act that
would normally require
human intelligence.
Hear
See
Speak
Feel
AI that can act…
• Robotic process automation
• Deep question & answering
• Machine translation
• Collaborative systems
• Adaptive systems
AI that can sense…
• Natural language
• Audio & speech
• Machine vision
• Navigation
• Visualization
AI that can think…
• Knowledge & representation
• Planning & scheduling
• Reasoning
• Machine Learning
• Deep Learning
Statistics Econometrics Optimization
Complexity
Theory
Computer
Science
Game
Theory
FOUNDATION LAYER
Understand
Plan
Assist
Learn
Digital
Reactive
Physical
Creative
4. PwC 4
Adoption>
AI is automating, transforming,
and disrupting all industries
Value>
AI is projected to contribute
significant value to global GDP
Operationalize AI >
AI is moving from proof-of-
concepts to operational systems
Responsible AI>
Responsible use of AI is gaining
increased attention
AI as a National Competitive Advantage>
Countries are developing AI
strategies to gain national
competitive advantage
Investments>
Dramatic rise in deal value &
volume is reshaping AI
5. PwC
“Just as electricity
transformed almost
everything 100 years ago,
today I actually have a hard
time thinking of an industry
that I don’t think AI will
transform in the next several
years”
Value>
AI is projected to contribute
significant value to global
GDP
Source: Sizing the Prize, PwC Report, 2017;
Andrew Ng – Google Brain, Baidu, Stanford
GlobalGDPupliftduetoAI
($intrillions)
Global GDP Impact of AI through 2030
2030 IMPACT: $15.7T
60%
40%
6. PwC
Adoption>
AI is automating,
transforming, and disrupting
all industries
“Many CEOs feel they need to bring AI
into their organisation. There’s this
fear factor that if you’re not on the AI
bandwagon, then you’re going to lose
out to competitors that are going to be
eating your market, because they’re
using technologies to make decisions
faster and better than you.
Daniel Hume – CEO, Satalia, AI Solution provider
7. PwC 7
AUTOMATE TRANSFORM DISRUPT
Large Global Pharmaceutical
company using NLP and
Machine Learning to extract
adverse drug interaction from
multiple unstructured data
sources
Global airline built predictive
aircraft maintenance system
that reduced maintenance
related costs from Delays &
Cancellations
Global auto manufacturer
gamified their strategy to
evaluate go-to-market
scenarios for a new rideshare
and autonomous vehicle
business unit
96%
accuracy
$14-18M
Savings
200K
GTM scenarios
$1+ Bn
Acquisition
15%
reduction in delays
$25M
Cost reduction
8. PwC
“In fact, the business plans of
the next 10,000 startups are
easy to forecast: Take X and
add AI. Find something that
can be made better by adding
online smartness to it.”
Investments>
Dramatic rise in deal value &
volume is reshaping AI
Source: CapIQ data for Artificial Intelligence Software transactions. Includes M&A, private placements and IPOs
Note: Some transactions included did not have transaction value data available
Kevin Kelly – Founding Editor of Wired
9. PwC
Operationalize AI >
AI is moving from proof-of-
concepts to operational
systems
Operationalizing AI “Software developers are
from Mars, data scientists
are from Venus, and you
need Machine Learning Ops
on Earth to operationalize
Artificial Intelligence”
10. PwC
Responsible AI>
Responsible use of AI is
gaining increased attention
“We need to be vigilant
about how we design and
train these machine-learning
systems, or we will see
ingrained forms of bias built
into the artificial intelligence
of the future.”
Kate Crawford – Co-founder of AI Now Institute, NYU
63%
Had no formal measure or
policy for addressing ethical
issues arising from AI
18%
Believed that developer or
person who approved the AI
solution is accountable for the
actions of the AI
83.7%
Not very confident in their or
their organization's ability to
detect, then shut down a
malfunctioning AI before any
serious problems were caused
13.2%
Could clearly articulate the
cause if their organization’s AI
made an incorrect decision or
did something unexpected
11. PwC 11
Performance
• Risk of errors
• Risk of bias
• Risk of opaqueness
• Risk of performance instability
Security
• Adversarial attacks
• Cyber intrusion risks
• Privacy risks
• Open source software risks
Ethical
• Lack of values risk
• Value alignment risk
Societal
• Reputational risk
• Autonomous weapons
proliferation
• Risk of intelligence divide
Economic
• Job displacement
• Liability risk
• Risk of “winner takes all”
concentration of power
BUSINESS-LEVEL RISKS
NATIONAL-LEVEL RISKS
Risks
• Lack of human agency in AI
supported processes
• Inability to detect/control
rogue AI
Control
12. PwC 12
ETHICS & LEGAL
Ensure AI development is in line with major local and global regulations, both enacted and emerging; and allow the business to evaluate the ethics
of an AI system and how to operationalize ethics in the organization
INTERPRETABILITY
Enable human users to understand, appropriately
trust, and effectively manage the emerging
generation of AI
BIAS & FAIRNESS
Uncover bias in the underlying data and model
development process and enable the business to
understand what process may lead to unfairness
ROBUSTNESS & SECURITY
Assess the performance of AI over time to
identify potential disruptions or challenges to
long term performance
GOVERNANCE
Introduce enterprise-wide and end-to-end accountability for AI applications and consistency of operations to minimize risk and maximize ROI
Ethical & Societal
Performance & Security
Control
PwC’s Responsible AI toolkit helps companies develop AI that
is fair, explainable, safe, robust, transparent, and accountable
13. PwC
Evaluating different fairness definitions, detecting biases, and bias
interventions need to be part of model design & development
From how data is sourced, how models are built, to how models are used, biases can be
embedded into a machine learning process.
Bias
Detection
Bias Intervention
Fairness
Definition
What are the different fairness
definitions? Which one should we use
for what purpose?
How do we detect bias with respect to
different decisions, protected attributes,
definitions, and datasets?
How do we correct algorithms for biases and
measure tradeoff between accuracy and
fairness?
14. PwC
Fairness Definitions: There are over 30 mathematical definitions of
fairness. Which do you choose?
Fairness is not a ‘fuzzy’ concept – it is a social construct that can be
defined mathematically in multiple ways
Bias
Detection
Fairness
Definition
Bias
Interven
tion
Interpre
ting
results
Bias
Detection
Fairness
Definition
Bias
Interven
tion
Interpre
ting
results
Bias
Detection
Bias
Intervention
Fairness
Definition
Important considerations:
It is required to define fairness in order to
measure to bias in the system. Bias is a function
of:
● The fairness definition selected
● The decision being made
● The protected attributes such as
Race, Gender, Age, and the combination
thereof
● The dataset and model output being
measured
Statistical Measures
True Positives
False Positives
False Negatives
True Negatives
Positive Predictive Value
False Discovery Rate
False Omission Rate
Negative Predictive Value
True Positive Rate
False Positive Rate
False Negative Rate
True Negative Rate
Predicted & Actual Outcomes
Predictive Parity
False Positive Error Rate Balance
False Negative Error Rate Balance
Equal Opportunity
Conditional Accuracy
Overall Accuracy
Treatment Equality
Predicted Outcomes
Statistical Parity
Conditional Statistical Parity
Predicted Probabilities
Test Fairness
Well-Calibration
Balance for Positive Class
Balance for Negative Class
Similarity-based Measures
Causal Discrimination
Fairness Through Unawareness
Fairness Through Awareness
Causal Definitions
Counterfactual Fairness
No Unresolved Discrimination
No Proxy Discrimination
Fair Inference
Continuous (Regression)
Mean Residual Difference
Net Compensation
Fair Covariance
15. PwC
Fairness is a spectrum; an organization must decide how fair a system
will be, and to with respect to whom
Accuracy Fairness Trade-Off
There are many ways an AI can develop biases; for an organization to adopt AI, it must be certain
the model is fair. However, there are many definitions of fairness, and it is not possible to be fair
to both the individual and the group.
Population
Fairness
E.g.: People of a
state getting loan
approvals without
discrimination
Individual
Fairness
E.g.: Similar
individuals in
terms of multiple
demographic and
behavioral
attributes getting
approvals without
discrimination
Large group
fairness
E.g.: People of the
state belonging to a
specific race getting
approvals without
discrimination
Small group
Fairness
E.g.: Females in the
state belonging to a
specific race getting
approvals without
discrimination
Fairness Scale
16. PwC
AI as a National Competitive Advantage>
Countries are developing AI
strategies to gain national
competitive advantage
“Artificial intelligence is the future, not
only for Russia, but for all humankind.
It comes with colossal opportunities,
but also threats that are difficult to
predict. Whoever becomes the leader
in this sphere will become the ruler of
the world.”
Vladimir Putin – Russian President