Author Name
Admin_niti
Choose Report Type
Publication Date
Report Upload
Download
(1.28 MB)
vertical
Data Management & Analysis
PDF Text
Working Document:
Enforcement
Mechanisms for
Responsible
#AIforAll
DRAFT FOR DISCUSSION 2
Draft Document for discussion
The content of this draft document is solely for the purposes of discussion with
stakeholders on the proposed subject and does not necessarily reflect the views of
NITI Aayog.
The document was prepared based on expert consultations over the past year. The
information contained herein is neither exhaustive nor final and is subject to
change.
All stakeholders are requested to review the documents and provide comments on
or before 15 January 2021, preferably on email at annaroy@nic.in 3
This document (Part 2 of the series on Responsible AI) proposes a framework for enforcement of
responsible AI principles
Towards Responsible AI
DRAFT FOR DISCUSSION
2018: National Strategy for
Artificial Intelligence
Advocated responsible use of AI to
address ethical concerns
2020: Towards Responsible AI
for All (Part 1)
Proposes Principles for Responsible
Management of AI in India 4DRAFT FOR DISCUSSION
Use cases for Artificial Intelligence have emerged across sectors and the technology has shown rapid
growth over recent years
Startup investments in India for AI have happened across sectorsTime and cost to train ML system has come down
drastically in just 3 years
Source: AI Index 2019, Stanford HAI
AI -a general purpose technology showing rapid growth
Approach to manage risks cannot be isolated. Such approaches must be highly participatory and must
keep pace with technology 5
Need for a context specific approach
Risk across use cases and contexts vary and also evolve over time. One-size-fits-all approach is not
sustainable
Risk vary across use casesRisk depends on deployment
context
Enforcement depends on
regulatory environment
Use caseExample Risk
Autonomous
VehicleSafety
Credit lendingDiscrimination
Fraud detection
in healthcareInclusion
Face Recognition
Unlocking
phoneSurveillance
SectorRegulators
Health
NeHA, National
Medical
Commission,
Drug Controller
General
FinanceSEBI, PFRDA,
IRDAI
DRAFT FOR DISCUSSION
A flexible risk-based approach must be adopted. In this regard, the National Strategy for Artificial
Intelligence proposes an Oversight Body 6
Role of the oversight body
1. Manage and update
Principles for responsible AI in
India
The oversight body must play an enabling role under the following broad areas
2. Research technical, legal,
policy, societal issues of AI
3. Provide clarity on
responsible behaviour through
design structures, standards,
guidelines, etc
4. Enable access to
Responsible AI tools and
techniques
5. Education and Awareness
on Responsible AI
6. Coordinate with various
sectoral AI regulators, identify
gaps and harmonize policies
across sectors
7. Represent India (and other
emerging economies) in
International AI dialogue on
responsible AI
DRAFT FOR DISCUSSION 7
1.Manage and Update Principles for Responsible AI
The Principles should reflect the technology capabilities, risks, policy and legal environment and should
adapt accordingly
Monitor and Update
Mechanisms to translate
principles to practice
Continuously monitor and update the Principles for responsible
AI based on advances in use cases and technology
Interface with various bodies, in designing specific mechanisms
to translate principles into practice
DRAFT FOR DISCUSSION 8
2. Research into Responsible AI: Background
Research on responsible AI is vital to ‘AI for Greater Good’ and lags general AI research around the world
Research into ethics of AI is multi-disciplinary and must be aimed towards advancing the field, identifying
issues, address concerns around AI and inform policy decisions and guidelines
Research on ethics
around the world has
not kept pace with AI
research on trending
and classical topics*
*Detailed methodology on keywords used available in Prates (2018)
DRAFT FOR DISCUSSIONSource: AI Index 2019, Stanford HAI 9
2. Research into Responsible AI: Recommendations
Incentivise cross disciplinary
research
Study and monitor impact on
the ground
The Government may support research on the impact of AI in
Indian context and on fundamental research to advance
Responsible AI by prioritising funding opportunities and
fellowship programs.
International alliances may be leveraged to facilitate exchange of
multi-disciplinary talent, data, and consolidation of research
efforts, especially in areas of social good
Top conferences on ethics of AI may be incentivised to host in
India so that challenges and approaches around the world can be
studied and motivate indegenous research
Engage with local communities, civil societies and other relevant
organisations to study and monitor impact of various AI
deployments on different communities and publish policy papers
DRAFT FOR DISCUSSION 10DRAFT FOR DISCUSSION
3. Clarify responsible behaviour: Background (1/ 2)
*Source: CIS report-AI in Healthcare
Lack of clarity on responsible behaviour has inhibited the growth of AI in India
Government
Procurement
●Procurement mechanisms
●Monitoring mechanisms
●Liability framework
Healthcare*
●Doctor-Patient Confidentiality
●Informed Consent Process
●Explainability Standards
●Liability framework
Example areas where guidance/ clarity will help 11
3. Clarify responsible behaviour: Background (2/ 2)
Standards and guidelines are being developed around the world on responsible ways of managing
technologies under specific context and may be leveraged
ISO/IEC WD TS 4213
Information technology —Artificial
Intelligence —Assessment of machine
learning classification performance
ISO/IEC WD 5059
Software engineering —Systems and
software Quality Requirements and
Evaluation (SQuaRE) —Quality Model
for AI-based systems
IEEE P7004 -Standard for
Child and Student Data
Governance
IEEE P7005 -Standard for
Transparent Employer Data
Governance
DRAFT FOR DISCUSSION 12
3. Clarify responsible behaviour: Recommendations
Standards
Leverage and ratify international standards when possible (in
consultation with relevant Ministry/ sectoral regulator)
Standards may also be created or augmented for local context
when required in consultation with BIS and relevant sectoral
regulators
Oversight body may identify design standards, guidelines and acceptable benchmarks for priority use
cases with sectoral regulators and experts. These may be made mandatory for public sector procurement
Guidelines
Develop design guidelines, and frameworks for responsible AI
through policy sandbox and controlled pilots
Create guidelines for 'Model AI procurement' RFP for various
priority use cases to guide responsible AI procurement in the
public sector. Such documents may include risk assessment,
best practices through the lifecycle, clarity on responsibility,
liability and IP considerations.
DRAFT FOR DISCUSSION 13
4. Enable access to Responsible AI tools and
techniques: Recommendations
: Recommendations
Support open technology
projects
Hackathons, workshops, open challenge mechanisms may be
used to identify and promote technology solutions for adherence
to Principles
Linguistic and NLP tools in local Indian languages may be
promoted to facilitate access to benefits of AI across the country
Promote projects for development of any tools and technologies
to enable easy access to responsible AI practices
Promote development and access to data and technology tools for responsible AI
Enable data availability and
sharing
Identify issues with data availability, sharing mechanisms and
promote a) research into data generation, identifying proxies, b)
creation and adoption of safe data sharing protocols (ex: through
model protocols, data sharing agreements)
DRAFT FOR DISCUSSION 14
5. Awareness on Responsible AI: Background
Reduce Trust issues and
apprehension of AI systems
Understand capabilities,
strengths and weakness of AI
systems
Learn about Responsible AI
Management practices, tools
and techniques
Reduce Information
asymmetry
Develop skills to identify and
think through ethical problems
Broad aim of awareness programs may be as follows,
Such programs may be entity specific (Public sector, Private sector, Academia, General Public, etc) and
may be customized to the local context
DRAFT FOR DISCUSSION 15
5. Awareness on Responsible AI: Role based training
Training needs may depend on the role
Decision MakerProcurer/
Influencer
Implementing
AgencyUserImpacted
Stakeholder
How AI/ML works
Need for ethical
thinking
Best practices in
procurement
How AI/ML works
Identify and anticipate
ethical problems
Ability to reason on
potential solutions
Ability to communicate
ways of addressing
the problems
Standards, guidelines,
best practices
Tools and techniques
for responsible AI
Grievance redressal
mechanisms, SOPs,
etc
Capabilities of a
specific AI technology
Awareness of its
limitations and safe
usage protocols
Awareness of rights
Awareness of role,
capabilities, limitations
of AI
Awareness of
grievance redressal
mechanisms
*The topics mentioned are representative only. Actual needs may depend on individual context
DRAFT FOR DISCUSSION 16
5. Awareness on Responsible AI: Recommendations
(1/2)
General Public
Local communities and regional social organisations may also
be engaged to study the impact of AI, knowledge gaps and
facilitate targeted awareness campaigns
Public Sector
Independent organisations may be leveraged for needs
assessment, and developing targeted training curriculum for
public sector officials.
Academic Institutions, Private sector, and relevant experts may
also be involved for training on use cases, and best practices.
States, departments and bodies with experience in responsible
deployment may host others and create sister-city agreements
for knowledge transfer
DRAFT FOR DISCUSSION 17
5. Awareness on Responsible AI: Recommendations
(2/2)
Private Sector
Private sector may be encouraged to create open knowledge
resources on risks, case studies and best practices on
responsible AI, in collaboration with academic institutes.
Ethics-by-design standards for Responsible AI may mandate
training for all stakeholders
Academic Institutes
Courses to be introduced at the earliest appropriate level to
develop the skills to think through ethical issues early and learn
to identify effective ways of addressing them
Model curriculum may be created for Universities to leverage-
(developed in collaboration with the Ministry of Education).
In universities where multi-disciplinary faculty is not available,
cross-university collaboration and guest lectures may be
considered.
DRAFT FOR DISCUSSION 18
6. Coordinate with sectoral regulators
Multiple regulators across sectors are regulating data and AI. This requires coordination to prevent
inconsistent policies and ambiguity, especially for cross sectoral AI use cases
Coordinate ApproachesCoordinate approaches across various regulators to avoid
duplication of efforts and inconsistent policies
Identify risksAssist regulators in identifying risks w.r.t AI use cases and
design policies, benchmarks, or ratify standards as applicable
Monitor policies
Work with various civil societies, research institutions, industry
bodies and other relevant agencies to monitor existing policies
and regulations gaps, inconsistencies, and other issues and
provide recommendations
Publish policy papers and promote any such activities that
contribute to realising benefits of AI while managing the risks
DRAFT FOR DISCUSSION 19
7. Represent India in International dialogue on AI
International collaboration on
Responsible AI
Identify avenues for International collaboration on Responsible AI
Provide India's (and other emerging economies) perspective on
responsible AI in International forum
Policies to enable International
collaboration
Assist relevant ministries (MeitY, MEA in development of cross
border data sharing protocols to facilitate collaborative research
Assist in facilitating International University collaborations on
Responsible AI
DRAFT FOR DISCUSSION Design of Oversight Body 21
Global Practices
United KingdomSingapore
Centre for Data Ethics
and Innovation
Advisory Council on
Ethical Use of AI and
Data
Under Department for Digital, Culture,
Media & Sport
Independent Board comprising
expert and influential individuals
from a range of fields relevant to its
mandate
Under Infocomm Media Development
Authority (IMDA)
Eleven council members include
international leaders in AI; advocates
of social and consumer interests;
and leaders of local companies
DRAFT FOR DISCUSSION 22
India’s approach: Highly participatory advisory body
is proposed
Considerations behind design of the Oversight mechanism
●Existing regulatory instruments are best placed to enforce rules, standards and guidelines. The
oversight mechanism may serve in advisory capacity
●It must interface with existing regulators across sectors
●Have dedicated resources to drive each mandate
●Technology easily blends across other technologies and must not be viewed in silo. Ethics should
be seen as not just limited to AI but also other emerging technologies such as ARVR, etc
DRAFT FOR DISCUSSION 23
Proposed composition of the advisory body
For effective functioning, the body must include,
●Computer Science and AI experts,
●Legal experts,
●Relevant sectoral experts,
●Civil societies,
●Humanities and Social Science experts
●Industry representatives
●Representatives from Standard setting bodies
●Government support for interfacing across Ministries and Departments
Additional experts may be opted in by the body depending on the requirement
DRAFT FOR DISCUSSION
A Council for Ethics and Technologyis proposed with a multi-disciplinary composition Institution-wise
Structures for
Enforcement 25
Procurement in Public Sector
Constitute Ethical Committee
An ethics committee may be constituted for the procurement,
development, operations phase of AI systems and be made
accountable for adherence to the Responsible AI principles
Composition depends on use
case
Composition of the committee will depend on the use case. A
model terms of reference and composition of such a committee
is proposed in the following slides
Procurement of AI systems may include a review by an ‘Ethical Committee’
DRAFT FOR DISCUSSION 26
Model Terms of Reference of Ethical Committees (1/2)
Ethical Committees are accountable for enforcement of principles in the AI system’s lifecycle
●EC should assess the “potential of harm” and potential benefits, evaluate plan for mitigating risks
and provide recommendations on whether the AI solution should be approved.
●Ethical Committees (EC) must ensure the AI system is developed, deployed, operated and
maintained in accordance with the Principles
●EC should determine the extent of review needed for an AI system depending on inherent risks and
benefits including but not limited to external audit.
●EC should ensure accessible and affordable grievance redressal mechanisms for decisions made by
the AI system.
DRAFT FOR DISCUSSION 27
Model Terms of Reference of Ethical Committees (2/2)
●EC should ensure creation of structures within the entity for protection of ‘whistleblowers’ reporting
unethical practices
●Every EC should have a documented Standard Operating Protocol (SOP) on functioning. The SOP
may be reviewed and updated periodically to reflect changing requirements
●Every EC review must be documented, including the risks identified, mitigation strategy, and
comments from the committee members
DRAFT FOR DISCUSSION 28
Model Composition of Ethical Committee (1/2)
Ethical Committees should have multi-disciplinary composition without Conflict of Interest
MemberDefinition
Chairperson●Nodal point of contact, accountable for independent and efficient
functioning of the committee
●Must be able to ensure active participation of all members in
discussions and deliberations
●Ratify minutes of EC meetings
Member Secretary●Must be a member of the organization or institute and should be
able to dedicate time for EC reviews
●Ensure effective procedures and protocols for EC review
Data Science and/or AI expert
(one or more depending on
requirement)
●Must be a qualified data scientist
●Must identify procedural or technical risks during development and
deployment including, data collection, annotation, management,
storage, processing, training, maintenance, and monitoring.
DRAFT FOR DISCUSSION 29
Model Composition of Ethical Committee (2/2)
MemberDefinition
Sector expert●Must have expertise in the sector and wide ranging deployment
scenarios
●Must evaluate safety, reliability, access and affordability of grievance
redressal mechanism
Legal expert●Must have expertise in relevant rules and regulations relevant to the
AI system
●Must evaluate legal considerations for the AI system
Social scientist/ ethicist (one or
more depending on requirement)
●Must have background in social or behavioural science or relevant
expertise. Must be sensitive to local cultural and moral values.
●Must assess impact on community, socio-cultural, religious,
philosophical context
Representative of Stakeholder
community (one or more,
depending on requirement)
Must be a stakeholder of the AI solution. Serve as a representative of the
user community
DRAFT FOR DISCUSSION 30
Private Sector (1/ 3)
Encourage Self-Regulation in
general
India currently lags in terms of private sector investment in AI
and risks under local context are yet to be fully understood (AI
Index, 2019).
In many countries around the world, public awareness and
market forces have incentivised the private sector to self
regulate
Voluntary self-regulation may be a good starting point for India
as well. This may evolve as the risks become clear
Private sector may be encouraged to use ethics-by-design structures (defined by standards bodies) in the
organisation. Adherence may further be incentivised through a carrot-and-stick approach
DRAFT FOR DISCUSSION 31
Private Sector (2/ 3)
Mandate adherence for high
risk use cases through
guidelines, standards and
other instruments
High risk use cases include all such use cases that have the potential to
cause significant harm to individuals
Such use cases, guidelines and adherence mechanisms may be defined
by the ‘Council for Ethics and Technology’ in consultation with sectoral
regulators and experts. Adherence may be through self-declaration or
through an independent third party audit, depending on the level of risk
International standards may not always to relevant, exhaustive or
available for Indian context. Hence it is critical that the Government
plays a role in ensuring the definition of 'acceptable behaviour' is clear
For high risk use cases*, adherence mechanisms may be mandated
DRAFT FOR DISCUSSION
*We invite comments as a part of public consultation on a framework to identify high risk applications and practical means toensure adherence 32
Private Sector (3/ 3)
Compliance to standards and guidelines has sometimes raised concerns in terms of creating a barrier to
entry for the startups. However, startups around the world have found unique ways to manage the costs,
some of them include,
a assigning accountability for ethics to their existing leadership team;
b) leveraging online courses, workshops, open materials so the entire team is aware of the risks and
develop the skill to ask the right questions;
c) leverage open tools and techniques;
Investment firms around the world are also being sensitized about the economic cost of non-adherence
Cost of compliance for ethical structures
DRAFT FOR DISCUSSION 33
Research Institutions
For research, a model ToR, composition and review mechanism for AI research may be developed by the
‘Council for Ethics and Technology’ in collaboration with the Ministry of Education
The existing Institute Review Board and peer review mechanism may be augmented with necessary
experts and cross-disciplinary skills. Cross-University collaboration may be considered in case the relevant
skills are not available
Journal and Conference may be recommended to include of ‘Statement of Ethical consideration’ in all
submissions
Government funding and fellowship opportunities on AI offered by various Ministries and Departments may
mandate institutional adherence to responsible AI structures
Existing Institute Review Board and Peer-review mechanism may be augmented
DRAFT FOR DISCUSSION 34
Acknowledgement
Prof Mayank Vatsa (IIT Jodhpur)
Arghya Sengupta and Ameen Jauhar (Vidhi Centre For Legal Policy)
Google Team
John C Havens (IEEE)
We acknowledge the contribution of the following experts for reviewing this document and providing
comments Thank You
Enforcement
Mechanisms for
Responsible
#AIforAll
DRAFT FOR DISCUSSION 2
Draft Document for discussion
The content of this draft document is solely for the purposes of discussion with
stakeholders on the proposed subject and does not necessarily reflect the views of
NITI Aayog.
The document was prepared based on expert consultations over the past year. The
information contained herein is neither exhaustive nor final and is subject to
change.
All stakeholders are requested to review the documents and provide comments on
or before 15 January 2021, preferably on email at annaroy@nic.in 3
This document (Part 2 of the series on Responsible AI) proposes a framework for enforcement of
responsible AI principles
Towards Responsible AI
DRAFT FOR DISCUSSION
2018: National Strategy for
Artificial Intelligence
Advocated responsible use of AI to
address ethical concerns
2020: Towards Responsible AI
for All (Part 1)
Proposes Principles for Responsible
Management of AI in India 4DRAFT FOR DISCUSSION
Use cases for Artificial Intelligence have emerged across sectors and the technology has shown rapid
growth over recent years
Startup investments in India for AI have happened across sectorsTime and cost to train ML system has come down
drastically in just 3 years
Source: AI Index 2019, Stanford HAI
AI -a general purpose technology showing rapid growth
Approach to manage risks cannot be isolated. Such approaches must be highly participatory and must
keep pace with technology 5
Need for a context specific approach
Risk across use cases and contexts vary and also evolve over time. One-size-fits-all approach is not
sustainable
Risk vary across use casesRisk depends on deployment
context
Enforcement depends on
regulatory environment
Use caseExample Risk
Autonomous
VehicleSafety
Credit lendingDiscrimination
Fraud detection
in healthcareInclusion
Face Recognition
Unlocking
phoneSurveillance
SectorRegulators
Health
NeHA, National
Medical
Commission,
Drug Controller
General
FinanceSEBI, PFRDA,
IRDAI
DRAFT FOR DISCUSSION
A flexible risk-based approach must be adopted. In this regard, the National Strategy for Artificial
Intelligence proposes an Oversight Body 6
Role of the oversight body
1. Manage and update
Principles for responsible AI in
India
The oversight body must play an enabling role under the following broad areas
2. Research technical, legal,
policy, societal issues of AI
3. Provide clarity on
responsible behaviour through
design structures, standards,
guidelines, etc
4. Enable access to
Responsible AI tools and
techniques
5. Education and Awareness
on Responsible AI
6. Coordinate with various
sectoral AI regulators, identify
gaps and harmonize policies
across sectors
7. Represent India (and other
emerging economies) in
International AI dialogue on
responsible AI
DRAFT FOR DISCUSSION 7
1.Manage and Update Principles for Responsible AI
The Principles should reflect the technology capabilities, risks, policy and legal environment and should
adapt accordingly
Monitor and Update
Mechanisms to translate
principles to practice
Continuously monitor and update the Principles for responsible
AI based on advances in use cases and technology
Interface with various bodies, in designing specific mechanisms
to translate principles into practice
DRAFT FOR DISCUSSION 8
2. Research into Responsible AI: Background
Research on responsible AI is vital to ‘AI for Greater Good’ and lags general AI research around the world
Research into ethics of AI is multi-disciplinary and must be aimed towards advancing the field, identifying
issues, address concerns around AI and inform policy decisions and guidelines
Research on ethics
around the world has
not kept pace with AI
research on trending
and classical topics*
*Detailed methodology on keywords used available in Prates (2018)
DRAFT FOR DISCUSSIONSource: AI Index 2019, Stanford HAI 9
2. Research into Responsible AI: Recommendations
Incentivise cross disciplinary
research
Study and monitor impact on
the ground
The Government may support research on the impact of AI in
Indian context and on fundamental research to advance
Responsible AI by prioritising funding opportunities and
fellowship programs.
International alliances may be leveraged to facilitate exchange of
multi-disciplinary talent, data, and consolidation of research
efforts, especially in areas of social good
Top conferences on ethics of AI may be incentivised to host in
India so that challenges and approaches around the world can be
studied and motivate indegenous research
Engage with local communities, civil societies and other relevant
organisations to study and monitor impact of various AI
deployments on different communities and publish policy papers
DRAFT FOR DISCUSSION 10DRAFT FOR DISCUSSION
3. Clarify responsible behaviour: Background (1/ 2)
*Source: CIS report-AI in Healthcare
Lack of clarity on responsible behaviour has inhibited the growth of AI in India
Government
Procurement
●Procurement mechanisms
●Monitoring mechanisms
●Liability framework
Healthcare*
●Doctor-Patient Confidentiality
●Informed Consent Process
●Explainability Standards
●Liability framework
Example areas where guidance/ clarity will help 11
3. Clarify responsible behaviour: Background (2/ 2)
Standards and guidelines are being developed around the world on responsible ways of managing
technologies under specific context and may be leveraged
ISO/IEC WD TS 4213
Information technology —Artificial
Intelligence —Assessment of machine
learning classification performance
ISO/IEC WD 5059
Software engineering —Systems and
software Quality Requirements and
Evaluation (SQuaRE) —Quality Model
for AI-based systems
IEEE P7004 -Standard for
Child and Student Data
Governance
IEEE P7005 -Standard for
Transparent Employer Data
Governance
DRAFT FOR DISCUSSION 12
3. Clarify responsible behaviour: Recommendations
Standards
Leverage and ratify international standards when possible (in
consultation with relevant Ministry/ sectoral regulator)
Standards may also be created or augmented for local context
when required in consultation with BIS and relevant sectoral
regulators
Oversight body may identify design standards, guidelines and acceptable benchmarks for priority use
cases with sectoral regulators and experts. These may be made mandatory for public sector procurement
Guidelines
Develop design guidelines, and frameworks for responsible AI
through policy sandbox and controlled pilots
Create guidelines for 'Model AI procurement' RFP for various
priority use cases to guide responsible AI procurement in the
public sector. Such documents may include risk assessment,
best practices through the lifecycle, clarity on responsibility,
liability and IP considerations.
DRAFT FOR DISCUSSION 13
4. Enable access to Responsible AI tools and
techniques: Recommendations
: Recommendations
Support open technology
projects
Hackathons, workshops, open challenge mechanisms may be
used to identify and promote technology solutions for adherence
to Principles
Linguistic and NLP tools in local Indian languages may be
promoted to facilitate access to benefits of AI across the country
Promote projects for development of any tools and technologies
to enable easy access to responsible AI practices
Promote development and access to data and technology tools for responsible AI
Enable data availability and
sharing
Identify issues with data availability, sharing mechanisms and
promote a) research into data generation, identifying proxies, b)
creation and adoption of safe data sharing protocols (ex: through
model protocols, data sharing agreements)
DRAFT FOR DISCUSSION 14
5. Awareness on Responsible AI: Background
Reduce Trust issues and
apprehension of AI systems
Understand capabilities,
strengths and weakness of AI
systems
Learn about Responsible AI
Management practices, tools
and techniques
Reduce Information
asymmetry
Develop skills to identify and
think through ethical problems
Broad aim of awareness programs may be as follows,
Such programs may be entity specific (Public sector, Private sector, Academia, General Public, etc) and
may be customized to the local context
DRAFT FOR DISCUSSION 15
5. Awareness on Responsible AI: Role based training
Training needs may depend on the role
Decision MakerProcurer/
Influencer
Implementing
AgencyUserImpacted
Stakeholder
How AI/ML works
Need for ethical
thinking
Best practices in
procurement
How AI/ML works
Identify and anticipate
ethical problems
Ability to reason on
potential solutions
Ability to communicate
ways of addressing
the problems
Standards, guidelines,
best practices
Tools and techniques
for responsible AI
Grievance redressal
mechanisms, SOPs,
etc
Capabilities of a
specific AI technology
Awareness of its
limitations and safe
usage protocols
Awareness of rights
Awareness of role,
capabilities, limitations
of AI
Awareness of
grievance redressal
mechanisms
*The topics mentioned are representative only. Actual needs may depend on individual context
DRAFT FOR DISCUSSION 16
5. Awareness on Responsible AI: Recommendations
(1/2)
General Public
Local communities and regional social organisations may also
be engaged to study the impact of AI, knowledge gaps and
facilitate targeted awareness campaigns
Public Sector
Independent organisations may be leveraged for needs
assessment, and developing targeted training curriculum for
public sector officials.
Academic Institutions, Private sector, and relevant experts may
also be involved for training on use cases, and best practices.
States, departments and bodies with experience in responsible
deployment may host others and create sister-city agreements
for knowledge transfer
DRAFT FOR DISCUSSION 17
5. Awareness on Responsible AI: Recommendations
(2/2)
Private Sector
Private sector may be encouraged to create open knowledge
resources on risks, case studies and best practices on
responsible AI, in collaboration with academic institutes.
Ethics-by-design standards for Responsible AI may mandate
training for all stakeholders
Academic Institutes
Courses to be introduced at the earliest appropriate level to
develop the skills to think through ethical issues early and learn
to identify effective ways of addressing them
Model curriculum may be created for Universities to leverage-
(developed in collaboration with the Ministry of Education).
In universities where multi-disciplinary faculty is not available,
cross-university collaboration and guest lectures may be
considered.
DRAFT FOR DISCUSSION 18
6. Coordinate with sectoral regulators
Multiple regulators across sectors are regulating data and AI. This requires coordination to prevent
inconsistent policies and ambiguity, especially for cross sectoral AI use cases
Coordinate ApproachesCoordinate approaches across various regulators to avoid
duplication of efforts and inconsistent policies
Identify risksAssist regulators in identifying risks w.r.t AI use cases and
design policies, benchmarks, or ratify standards as applicable
Monitor policies
Work with various civil societies, research institutions, industry
bodies and other relevant agencies to monitor existing policies
and regulations gaps, inconsistencies, and other issues and
provide recommendations
Publish policy papers and promote any such activities that
contribute to realising benefits of AI while managing the risks
DRAFT FOR DISCUSSION 19
7. Represent India in International dialogue on AI
International collaboration on
Responsible AI
Identify avenues for International collaboration on Responsible AI
Provide India's (and other emerging economies) perspective on
responsible AI in International forum
Policies to enable International
collaboration
Assist relevant ministries (MeitY, MEA in development of cross
border data sharing protocols to facilitate collaborative research
Assist in facilitating International University collaborations on
Responsible AI
DRAFT FOR DISCUSSION Design of Oversight Body 21
Global Practices
United KingdomSingapore
Centre for Data Ethics
and Innovation
Advisory Council on
Ethical Use of AI and
Data
Under Department for Digital, Culture,
Media & Sport
Independent Board comprising
expert and influential individuals
from a range of fields relevant to its
mandate
Under Infocomm Media Development
Authority (IMDA)
Eleven council members include
international leaders in AI; advocates
of social and consumer interests;
and leaders of local companies
DRAFT FOR DISCUSSION 22
India’s approach: Highly participatory advisory body
is proposed
Considerations behind design of the Oversight mechanism
●Existing regulatory instruments are best placed to enforce rules, standards and guidelines. The
oversight mechanism may serve in advisory capacity
●It must interface with existing regulators across sectors
●Have dedicated resources to drive each mandate
●Technology easily blends across other technologies and must not be viewed in silo. Ethics should
be seen as not just limited to AI but also other emerging technologies such as ARVR, etc
DRAFT FOR DISCUSSION 23
Proposed composition of the advisory body
For effective functioning, the body must include,
●Computer Science and AI experts,
●Legal experts,
●Relevant sectoral experts,
●Civil societies,
●Humanities and Social Science experts
●Industry representatives
●Representatives from Standard setting bodies
●Government support for interfacing across Ministries and Departments
Additional experts may be opted in by the body depending on the requirement
DRAFT FOR DISCUSSION
A Council for Ethics and Technologyis proposed with a multi-disciplinary composition Institution-wise
Structures for
Enforcement 25
Procurement in Public Sector
Constitute Ethical Committee
An ethics committee may be constituted for the procurement,
development, operations phase of AI systems and be made
accountable for adherence to the Responsible AI principles
Composition depends on use
case
Composition of the committee will depend on the use case. A
model terms of reference and composition of such a committee
is proposed in the following slides
Procurement of AI systems may include a review by an ‘Ethical Committee’
DRAFT FOR DISCUSSION 26
Model Terms of Reference of Ethical Committees (1/2)
Ethical Committees are accountable for enforcement of principles in the AI system’s lifecycle
●EC should assess the “potential of harm” and potential benefits, evaluate plan for mitigating risks
and provide recommendations on whether the AI solution should be approved.
●Ethical Committees (EC) must ensure the AI system is developed, deployed, operated and
maintained in accordance with the Principles
●EC should determine the extent of review needed for an AI system depending on inherent risks and
benefits including but not limited to external audit.
●EC should ensure accessible and affordable grievance redressal mechanisms for decisions made by
the AI system.
DRAFT FOR DISCUSSION 27
Model Terms of Reference of Ethical Committees (2/2)
●EC should ensure creation of structures within the entity for protection of ‘whistleblowers’ reporting
unethical practices
●Every EC should have a documented Standard Operating Protocol (SOP) on functioning. The SOP
may be reviewed and updated periodically to reflect changing requirements
●Every EC review must be documented, including the risks identified, mitigation strategy, and
comments from the committee members
DRAFT FOR DISCUSSION 28
Model Composition of Ethical Committee (1/2)
Ethical Committees should have multi-disciplinary composition without Conflict of Interest
MemberDefinition
Chairperson●Nodal point of contact, accountable for independent and efficient
functioning of the committee
●Must be able to ensure active participation of all members in
discussions and deliberations
●Ratify minutes of EC meetings
Member Secretary●Must be a member of the organization or institute and should be
able to dedicate time for EC reviews
●Ensure effective procedures and protocols for EC review
Data Science and/or AI expert
(one or more depending on
requirement)
●Must be a qualified data scientist
●Must identify procedural or technical risks during development and
deployment including, data collection, annotation, management,
storage, processing, training, maintenance, and monitoring.
DRAFT FOR DISCUSSION 29
Model Composition of Ethical Committee (2/2)
MemberDefinition
Sector expert●Must have expertise in the sector and wide ranging deployment
scenarios
●Must evaluate safety, reliability, access and affordability of grievance
redressal mechanism
Legal expert●Must have expertise in relevant rules and regulations relevant to the
AI system
●Must evaluate legal considerations for the AI system
Social scientist/ ethicist (one or
more depending on requirement)
●Must have background in social or behavioural science or relevant
expertise. Must be sensitive to local cultural and moral values.
●Must assess impact on community, socio-cultural, religious,
philosophical context
Representative of Stakeholder
community (one or more,
depending on requirement)
Must be a stakeholder of the AI solution. Serve as a representative of the
user community
DRAFT FOR DISCUSSION 30
Private Sector (1/ 3)
Encourage Self-Regulation in
general
India currently lags in terms of private sector investment in AI
and risks under local context are yet to be fully understood (AI
Index, 2019).
In many countries around the world, public awareness and
market forces have incentivised the private sector to self
regulate
Voluntary self-regulation may be a good starting point for India
as well. This may evolve as the risks become clear
Private sector may be encouraged to use ethics-by-design structures (defined by standards bodies) in the
organisation. Adherence may further be incentivised through a carrot-and-stick approach
DRAFT FOR DISCUSSION 31
Private Sector (2/ 3)
Mandate adherence for high
risk use cases through
guidelines, standards and
other instruments
High risk use cases include all such use cases that have the potential to
cause significant harm to individuals
Such use cases, guidelines and adherence mechanisms may be defined
by the ‘Council for Ethics and Technology’ in consultation with sectoral
regulators and experts. Adherence may be through self-declaration or
through an independent third party audit, depending on the level of risk
International standards may not always to relevant, exhaustive or
available for Indian context. Hence it is critical that the Government
plays a role in ensuring the definition of 'acceptable behaviour' is clear
For high risk use cases*, adherence mechanisms may be mandated
DRAFT FOR DISCUSSION
*We invite comments as a part of public consultation on a framework to identify high risk applications and practical means toensure adherence 32
Private Sector (3/ 3)
Compliance to standards and guidelines has sometimes raised concerns in terms of creating a barrier to
entry for the startups. However, startups around the world have found unique ways to manage the costs,
some of them include,
a assigning accountability for ethics to their existing leadership team;
b) leveraging online courses, workshops, open materials so the entire team is aware of the risks and
develop the skill to ask the right questions;
c) leverage open tools and techniques;
Investment firms around the world are also being sensitized about the economic cost of non-adherence
Cost of compliance for ethical structures
DRAFT FOR DISCUSSION 33
Research Institutions
For research, a model ToR, composition and review mechanism for AI research may be developed by the
‘Council for Ethics and Technology’ in collaboration with the Ministry of Education
The existing Institute Review Board and peer review mechanism may be augmented with necessary
experts and cross-disciplinary skills. Cross-University collaboration may be considered in case the relevant
skills are not available
Journal and Conference may be recommended to include of ‘Statement of Ethical consideration’ in all
submissions
Government funding and fellowship opportunities on AI offered by various Ministries and Departments may
mandate institutional adherence to responsible AI structures
Existing Institute Review Board and Peer-review mechanism may be augmented
DRAFT FOR DISCUSSION 34
Acknowledgement
Prof Mayank Vatsa (IIT Jodhpur)
Arghya Sengupta and Ameen Jauhar (Vidhi Centre For Legal Policy)
Google Team
John C Havens (IEEE)
We acknowledge the contribution of the following experts for reviewing this document and providing
comments Thank You