Practical Implementation of your AI Policy
Policies must only be written when they can be enforced. Keep this in mind when writing your AI Policy. Here are some practical examples on how to implement and enforce your policy.
When procuring an AI system, the procurement team will use an AI assessment framework to evaluate ethical risks and alignment with the policy. Vendors may need to provide details on how they address issues like bias testing, transparency, and security.
Before deploying an AI system, the IT team will document how it complies with each relevant principle in the policy. An oversight committee reviews this documentation to approve deployment.
Data scientists will take bias mitigation steps when assembling datasets to train AI models. Representative data and testing on different demographic groups can improve fairness.
Product managers will consult user experience designers on how to effectively communicate an AI system's capabilities, limitations and key information about its decision-making process through user interfaces.
When an AI system makes or aids consequential decisions, such as credit approval, the human collaborators will review critical cases to ensure appropriate outcomes rather than blindly following AI recommendations.
IT security teams will routinely audit AI systems, monitor for cyber threats, and establish backup and recovery processes in case of failures.
The legal/compliance team will stay updated on AI regulations and modify corporate training programs and system design protocols to reflect legal requirements.
An internal assessment process will evaluate high-risk AI systems for policy compliance, unintended harms and user feedback on a regular basis after deployment.
Employees can raise concerns on irresponsible AI use to managers, ethics boards, or whistleblower hotlines without fear of retaliation. Violations will result in re-training or other corrective actions.