WQF7005 Week 4 Notes: Transparency & Accountability
ETHICAL TRANSPARENCY CRITERIA
This criteria suite explores:
Why Transparency?
What is AIS Ethical Transparency?
Ethical Requirements
Overview of Goals
TRANSPARENCY: WHY IT MATTERS?
Transparency involves:
Preservation of self-determination
Openness
Communication
Disclosure of information
Trust is essential for individuals and social well-being across all cultures.
Transparency fosters trust.
Values that support transparency:
Openness
Explicability
Explainability
Access to data
Auditability
Self-determination
Communication
Disclosure of information
WHAT IS ETHICAL TRANSPARENCY
Definition: Ethical transparency is the preservation of self-determination, openness, communication, and disclosure of information regarding an AIS and the organization and its processes that promote ethical behavior.
Scope:
Explainability, interpretability, and traceability of the system
Openness by an organization of its decision to use AIS
Ethical profile of an AIS
Accessible processes for transparency of use of data and algorithmic decisions
AI developer should be able to sketch a strategy for proper use of data, as well as a strategy on processing data source for quality data
Employee competency to explain the outcomes of AIS
ETHICAL FOUNDATIONAL REQUIREMENTS (EFRS) ETHICAL TRANSPARENCY
Processes and training that:
Support explanation of AIS decisions
Promote ethical data and logic use
Align with human rights and compliance
Requirements to comply with laws and safeguard human rights
GOALS OF ETHICAL TRANSPARENCY
G1 – Governance and capability
G2 – Clarity of operations
G3 – Awareness of AIS interaction
G4 – Confidence in system behavior
G5 – Accessible control and feedback
G6 – Upholding ethical integrity
INHIBITORS OF TRANSPARENCY
G1b – Behavioral obfuscation
G2b – Excessive liability fears
G3b – Trade secret protection
G1 – ORGANIZATIONAL GOVERNANCE, CAPABILITY AND MATURITY
Explain-ability rely on an organizational structure supported by appropriate roles and responsibilities, capably staffed, along with tracking of AIS risk management.
Employees able to explain AIS outcomes
Example: HR using CV automated apps; transparency is if HR can explain how the app works.
G2 – CLARITY OF OPERATIONS
Explain-ability is central to AI regulation and laws.
Openness and accessibility of system design.
Clear representation of organizational and system processes throughout the AIS lifecycle (e.g., system design, data flow between systems).
G3 – AWARENESS OF AIS INTERACTION
Right to know with whom we are communicating – human or machine.
Able to challenge the AIS decisions effectively and efficiently
G4 – CONFIDENCE IN SYSTEM BEHAVIOR
Technical robustness and safe and effective systems are essential in regulations, and for stakeholder trust.
Consistent validation and verification of system accuracy and predictability
Audit and issue mitigation process
G5 – ACCESSIBLE CONTROL AND FEEDBACK
Humans are always to be in control of the AIS and the consequences of its outcomes
Stakeholders are to understand how to reliably and meaningfully opt in or out and the consequences of doing so
G6 –UPHOLDING ETHICAL INTEGRITY
Transparency of the impact of an AIS and our ability to control outcomes of the AIS.
How the organization upholds its ethical profile and evolves to changes in regulations and social norms.
G1B – BEHAVIORAL OBFUSCATION
Obfuscation is unethical behavior.
Make it difficult for observers to accurately interpret the behavior of the obfuscating entity.
G2B – CONCERN WITH LIABILITY
Minimizing liability concerns minimizes responsibility of AIS consequences.
Excessive concern for liability can undermine ethical behavior and result in harm to stakeholders
Transparency should be given priority over concern for exposure at all levels of the organization
Liability dimensions include legal, commercial, financial, and human intervention
G3B – PROTECTION OF TRADE SECRETS
The desire to protect trade secrets can get in the way of transparency
Intellectual property protection should not prevent appropriate audit
Secure oversight is to be assured regardless
KEY TAKEAWAYS OF TRANSPARENCY
The universal human value of self- determination requires transparency of information and the ability for individuals to make informed decisions.
International regulations require the AIS to be explainable and to be understood by all stakeholders
Making the AIS explainable and understandable requires disclosure of information regarding an AI and how the organization and its processes promote ethical behavior
ETHICAL ACCOUNTABILITY CRITERIA
This criteria suite explores:
Why accountability?
What is AIS accountability?
Ethical requirements
Overview of goals
ACCOUNTABILITY: WHY IT MATTERS?
Being answerable and responsible for actions, products, decisions and policies and accepting liability and blame.
Values supporting accountability:
Autonomy
Agency
Dependability
Responsibility
WHAT IS ETHICAL ACCOUNTABILITY
Answerability, blameworthiness, and liability an organization accepts as it takes responsibility for the AIS and the consequences of its outcomes.
Scope:
Accountability of all inputs from all sources within the AIS environment and responsibility of all outputs
Employee competency to be accountable of outcomes of AIS
Ethical profile of an AIS
Accessible processes for transparency of use of data and algorithmic decisions
Employee competency to explain the outcomes of AIS
Human accountability: Ethical accountability of an AIS is human centric and intended to keep the human in the loop
ETHICAL FOUNDATIONAL REQUIREMENTS (EFRS) ETHICAL ACCOUNTABILITY
Organizations must:
Define roles and responsibilities
Ensure decision traceability
Maintain system boundaries
Enable audit and review
GOALS OF ETHICAL ACCOUNTABILITY
G1 – Governance and maturity
G2 – Clarity of operations
G3 – Human oversight
G4 – User challenge mechanisms
G5 – Ethical organizational alignment
INHIBITORS OF ACCOUNTABILITY
G1b – Random/ systematic errors
G2b – Ethics rubber-stamping
G3b – Poor or missing documentation and records
G1 – ORGANIZATIONAL GOVERNANCE, CAPABILITY AND MATURITY
Capable oversight of AIS.
Roles have sufficient accountability to determine outcomes of the AIS.
Requirements include accountability for ethical governance of suppliers that influence the AIS.
G2 – CLARITY OF OPERATIONS
Understanding the context of the system within all its scenarios at all stages of its lifecycle.
The operational environments, stakeholders, and contexts must be analyzed under various scenarios and clearly specified for design, deployment, and maintenance modes for the product, service, or system
G3 – HUMAN OVERSIGHT
The AIS cannot be accountable to itself.
Ethical Accountability requires a human to have override capability and clear risk assessments during the operation of the AIS.
G4 –USER INTERACTIONS/USER AWARENESS
To trust, humans need to know who or what they are trusting –a machine or a human.
Human in the loop.
G5 –UPHOLDING ETHICAL PROFILE
Ethical integrity is tied to the ethical profile being maintained.
Requirements include resources, processes and policies to monitor and mitigate ethical risks.
G1B – RANDOM AND SYSTEMATIC ERRORS
There will always be random errors that create risks.
Human-related systematic errors of omission and commission during the life cycle are included.
A structured process is required to monitor and manage errors
G2B – RUBBER STAMPING
When an ethics policy exists in name only, there is minimal effort to be accountable.
Ethics policy is implementable, measurable, and publicly visible
Regular executive oversight to prevent ethics washing.
G3B – INADEQUATE OR NON-EXISTENT RECORDS
If it is not recorded, it cannot be tracked. And if it cannot be tracked, it is difficult to be accountable.
Requirement of reasonable and adequate recording keeping, aligned to the risk and impact of the AIS.
VALUES SUPPORTING ACCOUNTABILITY
Autonomy
Agency
Dependability
Responsibility
KEY TAKEAWAYS OF ACCOUNTABILITY
The universal value of autonomy requires accountability by humans.
Humans are responsible for the consequences of the AIS.
An AIS is NEVER accountable.