Practical Approach Recommendations to Developers
Proposing Improvements for Ethical Violations
Once ethical issues are identified and analyzed, the next step involves proposing actionable improvements. Students can present their recommendations as ‘AI Developer Advisory Reports’ or simulate discussions through ‘Ethics Advisory Meetings’ in a role-play format.
Recommendations should include:
– Specify Violations: Which regulations/principles were violated and how
– Propose Improvements:
– Review and provide transparency over how developers collect data, including the source of data and whether proper consent was collected
– Enhance AI explainability (XAI)
– Clarify responsibility
– Improvement Implementation Roadmap:
– When (deadline)
– Who (responsible party)
– How (specific methodology)
Ensuring the Feasibility of Improvement Proposals
Developers may struggle to accept necessary changes or may dismiss overly idealistic or
vague criticisms. Therefore, it’s important to:
– Consider technical limitations and costs, including human resources and time
– Present realistic and feasible solutions
This will also serve as practice for students in finding real-world sensibility and points of
compromise.
Improvement Ideas by Case
Recruitment AI:
– Remove information that may introduce biases, such as gender or race, while also
ensuring the absence of reverse discrimination.
– Develop alternative indicators (e.g., objective indicators focused on work capabilities)
Traffic AI:
– Reflect data for transportation vulnerable (disabled, elderly, etc.)
– Install error detection modules to prevent malfunctions
Medical AI:
– Apply anonymization (De-identification) technology when using sensitive information
– Mandate human in the loop for important decisions and final confirmation by human
medical staff
