This document discusses developing an ethical framework for machine learning. It proposes combining intellectual quotient with emotional and spiritual quotients when designing machine learning algorithms. This would create four types of algorithms: mechanical learner, cognitive learner, ethical learner, and ethical master. The framework is aimed at minimizing bias in algorithms and making ethical future decisions. It provides implementation strategies at the national, organizational, and design levels, and suggests collaborations, guidelines, and monitoring to help ensure quality of values in machine learning.
1 of 17
Download to read offline
More Related Content
S6.2 ethical framework for machine malhotra
1. Ethical Framework for
Machine Learning
26-28 November
Santa Fe, Argentina
Charru Malhotra
Associate Professor (e-Governance
and ICT)
Indian Institute of Public Administration
New Delhi INDIA
charrumalhotra[at]gmail.com
2. 26-28 November
Santa Fe, Argentina
Way Forward
05
● Ethics’ Guidelines, Collaborations,
Authorities : to check compliance
of EF in ML
Implementation
Strategy
04
● Changes at 3 levels- National,
Organisational , Design Approach
Proposed Ethical
Framework (EF)
03
Combine EQ & SQ with IQ to creates ML
Algos: ‘Mechanical Learner,’ ‘Cognitive
Learner ‘, ‘Ethical Learner’, ‘Ethical
Master/ Leader’
Ethical Issues in ML
02
• M/C replace Humans- so who rules?
• Unethical Hacks !
• Biased Databases …
Introduction
01
● Ethics : Right V/s Wrong
● ML : Learning through Algos
Agenda
4. 26-28 November
Santa Fe, Argentina
Rules of Conduct
Kantianism
Emanuel Kant (1724- 1804)
- duty central to morality
-not what ‘we do’ but what
we ‘ought to do’
2.Social Contract Theory
Thomas Hobbs et. al (1588-1679)
Moral rules to govern
relationships among citizens
Aristotle (384-322 B.C.E)
virtuous ways benefit
possessor & society
1. Virtue ethics
4. Utilitarianism
Jeremy Bentham (1748-1832)
John Mill (1806- 1873)
Examine right & wrong based
on consequences of act / rule.
Evolution
of Ethics
3.Kantianism
02
ETHICS ‘Ethos’
Values
Morals
what is right and wrong ….how to act on it …to have happiness for… self
and society
5. 26-28 November
Santa Fe, Argentina
SOME
ETHICAL
ISSUES
o Decisions left to m/cs esp i
complex, huge data
o Whose judgment is better
o Replace human existence?
o internal frauds ?
o remote hacking ?
o perception-modifications?
Machines recording, analysing preferences
Machines learning from ‘unclean’ sources
o Non-inclusive policies?
03
7. 26-28 November
Santa Fe, Argentina
2017-FACEBOOK has revealed how its artificial intelligence created its own language and began nattering
in private ….
Google Photos ‘goof’ up in facial recognition…
MS Chatbot - “ Hail Nazi” tweets !!!
2016-Online Feud of Wiki-Bots
2016-Self Driven Uber Cars jump red lights;
Smart Home catches fire in Blackberg-Penn
2016-Semi-automated drones ( LETHAL AUTONOMOUS WEAPON SYSTEMS)deployed between demilitarize
zone of South and North Korea
2017-A Russian Robot,Promobot R77 wants to go Free; Sophie wants to kill the
human race
Reality itself is an AI Simulation that is malfunctioning
05
8. 26-28 November
Santa Fe, Argentina
06
INTRODUCTION
The study is an interplay of ML ( the hard science)
and Ethics ( the soft science)
AIM
How to minimize /eliminate bias in ML
algorithms for making predictions and
future decision making
9. 26-28 November
Santa Fe, Argentina
Secondary Sources
Academic Literature of ML & Ethics
20XX
Social Media
Netflix, Youtube, Blogs,Fiction too
Proposed Ethical Framework
DESIGNING ML ALGOs- Consider Emotional Quotient (EQ) &
Spiritual Quotient (SQ) too apart from the Intellectual Quotient (IQ)
Methodology
Descriptive, Exploratory & Analytical
+
07
10. 26-28 November
Santa Fe, Argentina
1942
-Isaac Asimov (“Runaround” ,
1942)
-3 laws of robotics”,
-human-like robots to act
servants & need programming
rules to prevent harm
2018
2009
Wallach & Allen (2009) suggest top-down, bottom-up &
hybrid approach for machine morality.
● Top-down- moral frameworks of philosophers,
● bottom-up- machine learn through manipulation
like a child
2017
2016
Satya Nadella (2016) (CEO,
Microsoft), 10 essential rules for
an ideal AI
Sundar Pichai (2018)
(CEO,Google) -seven principles
that guide Google
(i) socially beneficial; (ii) avoid
unfair bias; (iii) built & test for
safety; (iv) accountable etc.
Existent Ethical Frameworks
Comest of UNESCO:
proposes technology-based
ethical framework on
robotics
08
11. 26-28 November
Santa Fe, Argentina
IQ
EQ
SPIRITUAL
QUOTIENT
• Intelligence
• Rational Clarity
• Scientific Decision
Making (dm)
• Empathy
• Sympathy
• Respecting
existence of
others
PURPOSE
PEACE
Existing ML algos
are based on IQ
Learn & recognise
patterns without
many rules
09
12. 26-28 November
Santa Fe, Argentina
LOW Spiritual Quotient (SQ) High
LOW
High
Emotional
Quotient
(EQ)
COGNITIVE LEARNER ALGO
medium complexity jobs
MECHANICAL LEARNER ALGO
repetitive jobs
ETHICAL MASTER ALGO
best-case for global decisions
ETHICAL LEARNER ALGO
higher complexity jobs
IQ
Proposed Ethical Framework
01
02
03
04
LOW EQ( neutral responses)
LOW SQ ( immature values)
HIGH SQ
( mature values about actions taken,
Awareness about implications for
ALL)
High EQ
(empathetic responses,
context-sensitivity, learning
process is inclusive)
MOST UTOPIAN /IDEAL TYPE
(inclusive, rationale, ‘welfare-
for-all’ for all life-forms
DHARMA principles- no
shades of black or white)
10
13. 26-28 November
Santa Fe, Argentina
LOW Spiritual Quotient (SQ) High
LOW
High
Emotional
Quotient
(EQ)
COGNITIVE LEARNER ALGO
medium complexity jobs
empathetic, inclusive, contextual
MECHANICAL LEARNER ALGO
repetitive jobs neutral responses & immature values
ETHICAL MASTER ALGO
V. complex decisions by country leaders
ETHICAL LEARNER ALGO
higher complexity jobs- mature values,
awareness of implications for all; community dm
IQ
Proposed Ethical Framework
01
02
03
04
‘welfare-for-all’
Assembly Line Production
Restaurant Robots
TeleMedicine
Industry 4.0
Gia Sensitive Global Policies
C
R
M
Electronic Assistants (Alexa)
NLP
Predicting User Behaviour
Humanoids-Sophia/ Rashmi
Network mgmt, Intrusion /Fault
Detection, Clustering
Weather Forecasting
Dynamic DM for resource-mgmt. –
Traffic Mgmt; SMART HOMES
Military
Drones ; AV
Facial Recognition
DSS/ ES for Global Leaders &
Policy Makers
11
14. 26-28 November
Santa Fe, Argentina
National Level COE
AUTHORITY;
REGULATIONS; ETHICAL
INDICES/STDs/
COLLABORATIONS
● Work with multi experts
● National Ethical
Framework
● Support Policy-
enhancements
● Standards for big data
● Encourage research in
emerging areas by
global / local alliances
Organizational
Level
‘ETHICS are a MUST’
● Develop, Follow &
Monitor Ethical F/w
Templates
● Changes in hiring,
CB- training,
promotion &
feedback
● Incorporation of
ethical aspects at
design stage
● Assigning an
Ethics engineer
Design & Authority
“Humans stay the
Master”
● Design using EF-
Citizen-Centric
approach
● Machine should
be obedient to Owner,
Design & Authority
● Provide some
autonomy to
machine but basic
program should be
locked so that data
cannot be erased o
hacked.
Implementation Strategies
12
15. 26-28 November
Santa Fe, Argentina
Formulate
Based on the
proposed EF,
formulation of
International
guidelines for
QoV ( just like we
have QoServices ,
Qo Experiences)
Intervene
Interventions at
National level like
● Regulatory
Framework
● Standards
/Authority
● EF Templates for
ML programming
Collaborate
Associations of
leaders, academia,
ethics- advocates ,
practitioners at
global level
Multi-disciplinary
diverse design
teams at local level
Monitor &
Control
National authority
to check ethical
indices of ML
innovations for
QoV ;
Organisations to
check compliance
Way Forward : QoV (Quality of Values) in ML
13
16. 26-28 November
Santa Fe, Argentina
WTO, WCO, GENEVA (2018)
UNCTAD (United Nations Conference on
Trade and Development), GENEVA ( 2018)
The Swiss Embassy-India ( 2017)
The World Bank (2000 – till Now)
The Economist Intelligence Unit , HK-2017)
IIMD- International Institute for
Management Development (Lausanne,
Switzerland) (2016)
ADB - Asian Development Bank (2014)
Panthéon Sorbonne, PARIS (2016,2014)
Wollongong University, AUSTRALIA (
2011)
The Winrock International (1999)
Central Universities - JNU,DU,
BHU,IIT-D, Rajasthan University
Central ministries- meity ü, DoPT, MHA,
MEA,DST, CPWD, DARP&G
STATE GOVTS – KARNATAKA,
HARAYANA
INDUSTRY ASSOCIATIONS - NASSCOM
MEDIA/ Civil Society – The LokSabha, DEF
IT Consultancies with International
agencies
National Agencies
28 years of
IT Journey
-From MIS/IT/ICT realm (MCA ,MCSD )
to the ‘real world’ (IIT-Delhi) of 600
villages of India
(2005-2009)
-NE (Sikkim, Shillong, Mizoram)
-South (TN, Kerala, K’tka)
-North (Punjab, Haryana, Rajasthan)
-Centre (MP, 36garh,UP)
Almost 70.5
millions for IIPA
in last 5 years
20 Mphils/PhD
- 30 Papers/
chapters/ articles
- 6 handbooks
- 5 reports
Generated
Guided
About the Speaker
14
17. 26-28 November
Santa Fe, Argentina
‘Gracias’
charrumalhotra@gmail.com vinod.kotwal@nic.in intern5.icm@mea.gov.in
Special Thanks to- Ms.Ishita Malhotra and Mr. Gaurav Khandelwal; Presentation Courtesy- Shilpa, Nishtha, Heeba, Rashmi
Charru Malhotra Vinod Kotwal Surabhi Dalal
“For the longevity of natural life on earth it is imperative we design
algorithms for machines that become aware of what harmony, empathy,
conservation of resources is… that knows that WE ARE ALL ONE”
15