Keeping ethical bias out of artificial intelligence models remains a largely unsolved problem.
Building an ethical model based on sound policies is a great and essential first step, but is not enough to maintain ethical fairness throughout the model life cycle. To prevent bias from developing, a model must be actively monitored from the time it is put into production until the moment it is retired. Monitoring must be supported with proper governance and operations processes. Even that is not enough, because organizations will still need to present the fairness metrics to validate their models and retain them (with traceability to the model version) to meet audit requirements. Finally, organizations need to be able to do all this at scale, for all the models they have in production.
This paper presents an overview of how organizations can manage their AI models to ensure they are ethical, fair and unbiased throughout their entire life cycle. It puts special emphasis on the production stage and how enterprises can prevent bias from developing over time. It explains how actionable model monitoring can maintain ethical fairness and auditability, and describes how the ModelOp Center model monitoring solution supports ethical AI by automating and orchestrating many of the required tasks.
