Authors ForHumanity Inc.
License CC-BY-NC-ND-4.0
Responsibilities (illustrative) of committees in AI lifecycle
Socio-technical nature of AI, Algorithmic and Autonomous systems means that the human is
embedded in the equation, often through the use of Personal Data, equally as often impact
humans through outcome. The risk arising from AAA systems are unique/ multidisciplinary
skill sets to handle them appropriately and timely efforts and require Diverse Inputs and
Multiple Stakeholders Feedback.
To illustrate the roles of various committees , we are providing a brief guidance on the broader
roles and responsibilities of these committees along the lifecycle of AI, algorithmic or
autonomous systems.
There are five key committees (Data Management Committee, Algorithmic Risk Committee,
Ethics Committee, Childrens Data Oversight Committee and Testing & Evaluation
Committee).
This document provides an overview of illustrative responsibilities of each of these committees
through the lifecycle.
The following are guidance provided to understand the responsibility of the specific
committees and their interrelationships. These are illustrative and not exhaustive,
(especially in respect to specific audit criteria).
Data Management Committee
Phase Process Stage Illustrative responsibilities
Development Data Collection Assess risks associated with data collection including
appropriateness, relevance and representativeness. Report
the risks along with recommendations to ARC as part of
Data management report
Development Data Labeling Assess risks associated with data labeling including data
quality and information quality. Report the risks along with
recommendations to ARC as part of Data management &
information management report
Development Data Cleaning Assess risks associated with data preprocessing . Report the
Development Data risks along with recommendations to ARC as part of Data
transformation management & information management report
& reduction
—-------------
This document is the property of ForHumanity Inc. ©2022, ForHumanity Inc. a 501(c)(3) tax-exempt Public Charity Page 1
All rights reserved. Creative Commons CC-BY-NC-ND Attribution-NonCommerical-NoDerivs
Responsibilities (illustrative) of committees in AI lifecycle
Development Training, test
and validation
split
Development Model design Assess risks associated with the model including the data
quality and info quality of model data and pipeline data.
Report the risks along with recommendations to ARC as
part of Data management & information management
report
Development Model testing Assess risks associated with the model including the data
and validation quality and info quality of model data and pipeline data.
Report the risks along with recommendations to ARC as
part of Data management & information management
report
Deployment Human in the Assess risks associated with the adequacy and
loop / on the loop appropriateness of data and its related processing provided
to the HTL for action including the issues associated with
risk of cognitive bias contributed by the outcome data
representation to the HTL. Report the risks along with
recommendations to ARC as part of Data management &
information management report
Algorithmic Risk Committee
Phase Process Stage Illustrative responsibilities
Design Scope-Nature-Con Appropriateness of the design and approving
cept-Purpose processing of personal data. No reports at this
design stage. Feedback provided to business/ data science
teams as appropriate.
Design Necessity & Review and approve the reports. No reports at this
Proportionality stage
Development Data Collection Assess risks associated with privacy and bias with
DI&MSF, mitigate risks, ensure that residual risks
are within risk tolerance. Include inputs as part of
the Data transparency report. Report the risks,
treatment and residual risk management as part of
—-------------
This document is the property of ForHumanity Inc. ©2022, ForHumanity Inc. a 501(c)(3) tax-exempt Public Charity Page 2
All rights reserved. Creative Commons CC-BY-NC-ND Attribution-NonCommerical-NoDerivs
Responsibilities (illustrative) of committees in AI lifecycle
ARA.
Development Data Labeling
Development Data Cleaning
Development Data Assess risks associated with privacy, bias, data
transformation & governance, data & information quality with
reduction DI&MSF, mitigate risks, ensure that residual risks
Development Training, test and are within risk tolerance. Include inputs as part of
validation split Data transparency report. Report the risks,
treatment and residual risk management as part of
Development Model design ARA. Consider reports including ERA, TEC
Development Model testing and At-risk, Data Management and Info management
validation and CDOC recommendations in this regard
Development Model tuning
Deployment Model deployment
Deployment Model integration Assess risks associated with data and information
/ interface quality during integration, interface connectivity.
Report all risks as part of the Deployment Release
Report. Consider reports including ERA, TEC
At-risk, Data Management and Info management
and CDOC report in this regard
Deployment Human in the Assess risks, mitigate risks, ensure that residual
loop / on the loop risks are within risk tolerance. Include inputs as
part of the Data Transparency Report. Report the
risks, treatment and residual risk management as
part of ARA. Consider reports including ERA, TEC
At-risk, Data Management and Info management
and CDOC report in this regard
Deployment Model health,
fitness &
monitoring
Deployment Post market Assess risks (including potential harms, adverse
insights/ feedback events, emergent risks), mitigate risks, ensure that
residual risks are within risk tolerance. Consider
inputs from AIRS, Blackbox and other insights
from TEC.
—-------------
This document is the property of ForHumanity Inc. ©2022, ForHumanity Inc. a 501(c)(3) tax-exempt Public Charity Page 3
All rights reserved. Creative Commons CC-BY-NC-ND Attribution-NonCommerical-NoDerivs
Responsibilities (illustrative) of committees in AI lifecycle
Decommissioning Model Assess risks associated with data sharing consent,
decommissioning unplanned data loss, disaster recovery and BCP.
Provide and implement appropriate related risk
controls. Maintain / update in AAA inventory list.
Testing & Evaluation Committee
Phase Process Stage Illustrative responsibilities
Development Data Labeling Assess risks associated with data quality
(specifically data labeling) and information quality
including the risks arising from the associated
processes or systems used for the said purpose.
Report the risks, treatment and residual risks as
part of TEC At-Risk report
Development Data Cleaning Assess risks associated with data quality
Development Data (specifically data pre-processing) and information
transformation & quality including the risks arising from the
reduction associated processes or systems used for the said
purpose. Report the risks, treatment and residual
Development Training, test and risks as part of TEC At-Risk report
validation split
Development Model design Assess risks associated with safety, security,
accountability, governance and accessibility in the
model design stage with DI&MSF. Report the risks,
treatments and residual risks as part of TEC
AT-Risk Report
Development Model testing and Assess risks associated with bias, safety, security,
validation accountability, governance, transparency,
explainability and accessibility in the model testing
and validity stage with DI&MSF. Report the risks,
treatments and residual risks as part of TEC
AT-Risk Report
Development Model tuning Assess risks associated with bias, diversity,
accountability, explainability. Report the risks,
treatment, residual risks as part of TEC AT-risk
report.
—-------------
This document is the property of ForHumanity Inc. ©2022, ForHumanity Inc. a 501(c)(3) tax-exempt Public Charity Page 4
All rights reserved. Creative Commons CC-BY-NC-ND Attribution-NonCommerical-NoDerivs
Responsibilities (illustrative) of committees in AI lifecycle
Deployment Model deployment Assess risks associated with model interpretability
& drift as part of governance and explainability.
Report the risks, treatment, residual risks as part
of TEC At-Risk report
Deployment Model integration / Assess risks associated with model integration,
interface interfaces in terms of privacy, cybersecurity,
transparency, accountability, accuracy, auditability,
integrity. Report these risks, treatments, and
residual risks as part of TEC-AT - Risk Report.
Deployment Human in the loop Assess risks associated with HTL including
/ on the loop effectiveness of HTL as part of Governance with
DI&MSF. Report the risks, treatments and residual
risks as part of TEC AT-Risk Report. Also include
the process and mitigatable risks or residual risks
as part of the HTL integration report.
Deployment Model health, Assess risks associated with bias, safety, security,
fitness & accountability, governance, transparency,
monitoring explainability and accessibility in the model
monitoring stage with DI&MSF (including inputs
from AIRS, Stress testing, edge case testing etc).
Report the risks, treatments and residual risks as
part of TEC AT-Risk Report. Include guidance as
part of interpretability report
Deployment Post market Assess risks associated with the model based on
insights/ feedback inputs from AIRS. Report the risks, treatments and
residual risks as part of TEC AT-Risk Report.
Contribute with insights and mitigatable risks or
residual risks as part of a Post deployment model
management report prepared by ARC.
Decommissioning Model Assess risks associated with AAA , its integration
decommissioning and interface loss/ absence. Assess risks related to
partial/ full data unavailability to other systems/
operations. Appropriate risk mitigation controls
will be communicated to ARC as part of the
Decommissioning Report.
Ethics Committee
—-------------
This document is the property of ForHumanity Inc. ©2022, ForHumanity Inc. a 501(c)(3) tax-exempt Public Charity Page 5
All rights reserved. Creative Commons CC-BY-NC-ND Attribution-NonCommerical-NoDerivs
Responsibilities (illustrative) of committees in AI lifecycle
Phase Process Stage Illustrative responsibilities
Design Scope-Nature-Conc Appropriateness and inclusion of ethics in design.
ept-Purpose design No reports at this stage. Feedback provided to
business/ data science teams as appropriate
Development Data Collection Assess ethical issues in the data including bias and
ethical use of the data. Recommend mitigations/
suggestions to the ethical risks to ARC as part of
ERA.
Development Data Labeling Assess ethical issues in the data including bias,
representativeness and compliance with other
Development Data Cleaning
aspects covered under Code of Data Ethics.
Development Data Recommend mitigations/ suggestions to the ethical
transformation & risks to ARC as part of ERA.
reduction
Development Training, test and
validation split
Development Model design Assess ethical issues associated with the model
including accessibility and human agency (HTL &
Development Model testing and
overseer) and other relevant aspects covered as
validation
part of Code of Data Ethics. Recommend
Development Model tuning mitigations/ suggestions to the ethical risks to the
Deployment Model deployment ARC as part of ERA
Deployment Model integration / Assess the ethical issues of nudging,
interface appropriateness of data use in the context of
accountability and governance and highlight these
risks along with mitigations/ suggestions to ARC as
part of ERA
Deployment Human in the loop Assess adequacy and appropriateness of HTL in
/ on the loop the process and highlight ethical issues along with
recommended mitigations/ suggestions to ARC as
part of ERA
Deployment Model health, Assess risks associated with ethics based on
fitness & mitigatable risks or residual risks identified during
monitoring the KRI monitoring. Ethics risks along with
recommendations/ suggestions are reported to ARC
as part of ERA.
—-------------
This document is the property of ForHumanity Inc. ©2022, ForHumanity Inc. a 501(c)(3) tax-exempt Public Charity Page 6
All rights reserved. Creative Commons CC-BY-NC-ND Attribution-NonCommerical-NoDerivs
Responsibilities (illustrative) of committees in AI lifecycle
Deployment Post market Accumulate and report risks reported as part of the
insights/ feedback post market insights . ERC will assess related
ethical risks and propose appropriate risk
mitigation controls into AAA system where
applicable to ARC.
Decommissioning Model Assess risks associated with data transparency,
decommissioning unplanned data subject loss, disaster recovery and
BCP with respect to subjects. Provide and
implement appropriate related risk controls to
ARC. Maintain / update in AAA inventory list.
Children’s Data Oversight Committee
Phase Process Stage CDOC
Design Scope-Nature-Conc Appropriateness of considerations for children in the
ept-Purpose design design. No reports at this stage. Feedback provided
to business/ data science teams as appropriate
Development Data Collection Assess issues in the children's data including age
appropriateness, ethical use of the data and other
childrens data collection issues (eg. geolocation).
Highlight risks along with recommendations to ARA
and EC as appropriate in form of CDOC report.
Development Data Labeling Assess issues in the children's data including age
Development Data Cleaning appropriateness, ethical use of the data and other
childrens data collection issues (eg. geolocation).
Development Data Highlight risks along with recommendations to ARA
transformation & and EC as appropriate in form of CDOC report.
reduction
Development Training, test and
validation split
Development Model design Assess risks associated with processing of childrens
data (aligned with age appropriate design
Development Model testing and
guidelines). Highlight risks along with
validation
recommendations to ARA and EC as appropriate in
Development Model tuning form of CDOC report
Deployment Model deployment
—-------------
This document is the property of ForHumanity Inc. ©2022, ForHumanity Inc. a 501(c)(3) tax-exempt Public Charity Page 7
All rights reserved. Creative Commons CC-BY-NC-ND Attribution-NonCommerical-NoDerivs
Responsibilities (illustrative) of committees in AI lifecycle
Deployment Human in the loop Assess risks associated with parental controls and
/ on the loop oversight as relevant for children. Highlight risks
along with recommendations to ARA and EC as
appropriate in form of CDOC report
Deployment Post market Assess risks associated with processing of childrens
insights/ feedback data (aligned with age appropriate design
guidelines). Consider inputs from AIRS as part of
the process as applicable. Highlight risks along with
recommendations to ARA and EC as appropriate in
form of CDOC report
—-------------
This document is the property of ForHumanity Inc. ©2022, ForHumanity Inc. a 501(c)(3) tax-exempt Public Charity Page 8
All rights reserved. Creative Commons CC-BY-NC-ND Attribution-NonCommerical-NoDerivs