Building Supervised AI Models with AI Assurance Framework
Assuring safety of autonomous systems by standardizing AI algorithm testing
AI is finding increased application across industries. However, it still lacks the ability to process complex situations and take decisions. AI also finds it challenging to rationalize whether a task is appropriate or ethical. For AI systems to be successful, testers need to define the operational boundaries of AI and monitor them periodically to pre-empt any breaches. Leveraging the proposed approach towards AI assurance utilizes both human expertise and technology monitoring to help drive superior AI performance. Key points to consider while deploying this approach include:
- Pre-deployment phase - Choose a data set that closely resembles the production system, identify tools for feedback data, eliminate data biases, execute non-functional testing, and prioritize data sanity.
- Post-deployment phase - Review output from continuous feedback, establish failure threshold, use AI-monitoring platform to identify code progressions, classify required level of changes, and identify new data parameters.