How to design ML Observability for high-risk AI use cases

MLOps simplified the baseline processes making it easy to build models at scale today. But there has little or no focus on ML acceptance. Any AI/ML model can fail, models are not explainable by design, models can carry the risk of usage during production and model auditing is very complex. Deploying AI for mission-critical use cases requires additional layers like explainability, monitoring, auditability, data privacy and risk mitigation to ensure the AI solution is acceptable to all stakeholders.
Agenda:

1. Introducing ML Observability
2. Using ML Observability for model monitoring, model explainability and auditing.
3. Designing the policy layers to manage model usage risk in ML Observability.