Robustness is the ability of a system to maintain its level of performance under any circumstances. For AI systems, Robustness is often used to describe the ability of a model to maintain its level of performance under any circumstances, including external interference or harsh environmental conditions encompassing resilience and reliability and potentially more attributes, as related to proper operation of a system as it was intended by its developers. [ISO/IEC DIS 22989].
Robust AI, also known as stable AI or resilient AI, refers to artificial intelligence (AI) systems that are able to function effectively even when operating in unexpected or changing environments.
When considering if an AI is Robust, you will commonly be asking questions like:
- Will it deal with variation in the future?
- Is it vulnerable to manipulation?
- How quickly will it recover from significant variation?