System Subtainability
Checklist / Questions to ask your team
Di you have a method to detect and alert any model drift ie. keep track of performance degradation and accuracy of model
Are you monitoring your input for skews?
Have you tested arbitration when system failures or introduces bias
Have you considered using A|B test models to determine if a challenger model is ready to replace a deployed model?
Do you capture input data (features) and output predictions while in production
Have you consider long-term impact on human and environment?
Do you have a method to detect, report and asses AI harm incidents
2. External Resources - Tools to use
Feature monitoring libraries: Great Expectations and Deequ
System Logs
Test Score Card to include model accuracy, results/predictions features and raw inputs
3. Templates (Samples)
AU (Public) Incident databases: AIAAIC , I Incident Database, AI - Global
4. Case Studies
5 Further Readings