2. Agenda
Ø AI & ML – Relationship
Ø AI/ML on fast track in 21st century – WHY ?
Ø Human Learning vs. Machine Learning
Ø High Level Understanding of ML Learning Methods & Applications
Ø Neural Network – Human Brain Mimic
Ø Image Classification, Recognition, Object Detection
Ø Major Steps to ML Model Design
Ø AI Scope & Application – Unlimited
Ø Strength & Weakness – Human vs Artificial Intelligence
Ø Collaboration – Human + Artificial Intelligence
Ø Collaborative Intelligence – Complementary, not Competitive
Ø Ethical AI – Challenges & Strategy
3. AI & ML – Relationship
Ø Artificial Intelligence - Artificial intelligence leverages computers and machines to mimic
the problem-solving and decision-making capabilities of the human mind
Ø Rule based AI – Reasoning, NLP & Planning
Ø Pattern based AI – Machine Learning
4. AI/ML on fast track in 21st century – WHY ?
Ø AI/ML journey started in 1950-60s, but lack of Infrastructure for Model training
Ø Availability of High-end computing infrastructure like Cloud Computing, needed
for model training
Ø Rapid progress in associated Digital ecosystem along with Cloud computing - IoT,
BigData, Industry 4.0
Cloud
Computing
Industry
4.0
BigDat
a
IoT
AI
8. Deep Learning – Neural Network (Human
vs. AI)
Artificial Neurons (Processing Node) consists of
Ø Many Input connection(s) along with Weights
(Chemical Signals received by Synapse and
chemical signals are decoded into electrical
responses by Synaptic Receptors in Human
Brain)
Ø Computation Unit (Nucleus) consists of
o A Linear function (Wx + B)
o An Activation function ( e.g. Sigmoid ,
Softmax)
Ø Output
9. Image Classification, Localization, Object Detection,
Image Recognition
Ø YOLO V3 (You Only Look Once) is a
real time Object Detection
algorithm to identify tiny objects in
videos, live feeds or images
Ø Object Detection – When multi class
objects reside in the same picture
Ø YOLO v3 Object Detection does
Localization & Labelling
(Classification) together
10. Machine Learning - Model Designing Major
Steps
Ø Data Type – Numerical, Categorical, Time-series,
Structured/Un-Structured Text, Image
Ø Data Preparation
• Data Cleaning & Transformation - Image
Noise removal, Data Outlier removal, Null
Values removal
• Encode Categorical data (Hot-encoding)
• Select & Split Training/Test/Validation Data
• Feature Scaling, Dimensionality Reduction
(Feature Selection, PCA, LDA etc.)
Ø Train, Test & Validate
Ø Hyperparameter Tuning
Ø Prediction
12. Strength & Weakness – Human vs Artificial
Intelligence
Attribute Human
Intelligence
AI Impact
Thinking Ability &
Imagination à Creativity
Strength Weakness Artist, Poet, Novelist, Scientist
Removal Data Bias
(e.g. – Gender/Ethnicity
biased recruitment)
Far behind Strength Machine Learning BIAS can be eliminated by right data
selection and changing Model. But difficult to remove
ingrained human bias
Logic + Emotion Both Only Logic • Have Ethical, Legal & Societal benefit
• Leadership trait demands the balancing of these two
attributes
Decision Making Strength Far behind • AI Decision making is completely data driven, no
emotion
• AI can enhance human analytic & decision-making
by providing right data at right time
• So, combination of AI & Human Intelligence can give
much better result than by any of this in silo
Computational ability Linear Exponential • AI computational
13. Attribute Human
Intelligence
AI Impact
Repetitive Task &
Scalability
Far behind Expert • Human can utilize more time on strategic, productive
and creative thinking
General prediction
ability
Weakness Data based
prediction
(Strength)
• AI (ML) based data driven prediction output is much
better in quality and faster than Human Intelligence
Intuition/Instinct/Reflex/
Common
Sense/Correlation of
events or incidents/Self
Learning &
improvement/Cause
Effect Analysis
Strength Weakness • Will be very difficult for AI to become at per like
Human.
Adoptability to New
environment
Comes by
nature
Hard to adopt • Human plays the role of AI trainer to mimic human
like personality. E.g. Chatbot
• AI plays the role only for FAQ they have trained for,
unlikely using common sense in case of Human
Strength & Weakness – Human vs Artificial
Intelligence
14. Collaboration - Human + Artificial Intelligence
Speed
Flexibility
Process
Charact
eristics
Scale
Decision
Making
Personaliz
ation
Principles – Human + AI Collaboration to improve Process
Characteristics
ü Reimagine Business Processes
ü Embrace
experimentation/employee/stakeholder
involvement
ü Actively direct AI strategy
ü Responsibly collect data
ü Redesign work to incorporate AI and
cultivate related employee skills
Human AI
Training – Train machine to mimic like human Amplifying - AI can enhance human analytic &
decision-making by providing right data at right
time
Explaining – Explain the behavior of AI based
outcome
Interacting – Human-machine interface
Sustaining – Invigilator to examine whether AI
systems are functioning properly, safely and
responsibly
Embodying – AI is embodied in robot that what
augments human worker
Human + AI Collaboration - Role of Human & AI
15. Collaborative Intelligence – Complementary, not
Competitive
Functional
Aspect
Process
Characteristics
Application (Collaborative
Intelligence)
Role - Complementary
Decision
Making
• Flexibility
• Speed
• Decision Making
• Design Thinking + AI
led Analytics
(Optimize GoTo Market
Strategy, Process
Optimization, Fraud
Analysis)
• Rule based AI
(Product Design)
• Role of Human – Input considering stakeholder
empathy, SME (Process Decomposition)
• Role of AI – Text Analytics (NLP) and ML based
scoring
Repetitive Task
& Scalability
• Speed
• Scale
• Microsoft’s AI Assistant,
Cortana
• Apple’s SIRI
• Amazon’s Alexa
(Customer Service)
• Human plays the role pf Trainer on the basis
FAQ and personality trait development
• AI plays the role in Interaction
Augmentation
of Human
Worker
• Speed
• Scale
• Flexibility
• Cobot Arm
• Chairless Exoskeleton
(Human Robots to avoid
injury but need human
efficiency in the Assembly
Line in Automobile Industry
(Repetitive but not
Automated)
• Human Intelligence responsible for visual
inspection and Assembly line. Robot is not as
efficient as Human in such repetitive task
• Cobot/Exoskeleton extends the mobility limit
of human worker and eliminate them from
repetitive task like lifting heavy weight,
eliminate injury to worker, reduce fatigueness
16. Collaborative Intelligence – Complementary, not
Competitive
Functional
Aspect
Process
Characteristics
Application (Collaborative
Intelligence)
Role - Complementary
General
Prediction
• Speed
• Scale
• Personalization
• Predictive Maintenance
• Clustering
• Instant Fraud Detection
• Weather Forecasting
• Persona specific
Recommendation
(Historical or Real time
Data Analytics)
• Human utilize AI’s general prediction ability for
better decision making
17. Ethical AI – Challenges & Strategy
Challenges Strategy Example
• Risk/Benefit Analysis –
Human Life/Societal Damage
• Public/Private partnership is necessary to assess
Risk/Benefit of the product/Project objective
• Steps to mitigate/minimize human/societal
damage or consequence
• Assess the real need of AI Implementation
objective & outcome
Self Driving Car
• Create huge unemployment
• Hard to accept any car accident because
of unaccountable entity (Machine), not
person
• Reduce Pollution
• Human Values - AI doesn’t
have “common sense” what
human has. Therefore,
consideration of Human
Values is essential.
Ethical design should be adopted considering human
values through collaboration between private & public
institutions, concerned stakeholders and Government
British Standards Institute BS 8611 standard
on the "ethical design and application of
robots and robotic systems" provides some
useful guidance: "Robots should not be
designed solely or primarily to kill or harm
humans; humans, not robots, are the
responsible agents; it should be possible to
find out who is responsible for any robot
and its behaviour."
• Decision making & liability -
Challenge in apportion
responsibility of decisions.
Risk Ownership ?
Define the guideline of identifying Risk ownership Accountability - Insurance claim on
accidental damage for Self driving car will
be a concern if clear accountability is not
defined
18. Ethical AI – Challenges & Strategy
Challenges Strategy Example
• AI product must be
ü Transparent
(Explainable AI, Robust AI,
Fair AI, Private AI)
ü Un-Biased
• Data Protection & IP
• Cybersecurity
• Global or generic guideline
for Ethical AI
Ethical AI - Policy Guideline
• Evaluate Biasness and eliminate, if any
• Laws must be enforced to protect sensitive data
and IP right
• More reliance on AI, more importance on
cybersecurity is needed
• Transparency of AI Product/Solution must be
measured against
ü Explainability - must for customer
reassurance and regulators
ü Robustness - must resist risk,
unpredictability, volatility in real world
scenario
ü Fairness - must not discriminate based on
race/gender/religion or other similar
ü Data Privacy - Since AI Training requires lots
of data , Defining Privacy standard and no
theft guarantee is a must
• Every Use Case evaluation is needed
independently, not generic. This is also a challenge
• Biased AI - One of the renowned
eCommerce giant revealed Gender
biasness in their AI based Talent
Acquisition system
• Data Protection on Biometric based
Authentication
• Cybersecurity – Hacking to AI
Algorithm/solution or data could be a
danger
• Transparency - FACIAL RECOGNITION
SOFTWARE HAS COME UNDER
SCRUTINY FOR ISSUES WITH RACIAL
BIAS AND PRIVACY CONCERNS. IBM
has abandoned from offering,
developing or researching on Facial
Recognition Technology.
19. Summary
Ø AI does not mean only Machine Learning
Ø Similarity between Human Learning & Machine Learning
Ø Different Machine Learning Methods
Ø Machine Learning Model Design Steps & it’s Importance
Ø Scope of AI Application - Unlimited
Ø Human & Artificial Intelligence – Both has it’s own Strength & Weakness
Ø (Human & Artificial) Intelligence Collaboration – Why & How ?
Ø Ethical AI – It’s a Need, not an Option
Ø Ethical AI - Challenges & Strategy