Question 1070
A chatbot's error rate increases across a new language population but no production metrics detect it. A risk owner is documenting the concern for management. What should the responsible team do first to correct the specific weakness?
Explanation
Correct Answer
Implement subgroup and use-context performance monitoring with action thresholds. It addresses the failure that the deployment lacks feedback on real-world reliability and possible disparate impact. The expected evidence is that monitoring alerts on language-specific degradation and prompts a corrective response.
Incorrect Answers
That action can help where operational success metrics omitted foreseeable adverse societal impacts. It does not correct the current weakness: the deployment lacks feedback on real-world reliability and possible disparate impact.
That action can help where a deployed consequential workflow lacks meaningful operational oversight. It does not correct the current weakness: the deployment lacks feedback on real-world reliability and possible disparate impact.
That action can help where risk response is ineffective without an operational stop mechanism. It does not correct the current weakness: the deployment lacks feedback on real-world reliability and possible disparate impact.