A company is defining operational rules for using generative AI in high-impact decisions such as medical diagnosis and credit assessment. After weighing the balance between efficiency and accountability, which practice is the MOST appropriate from a responsible AI perspective?
A question about choosing the responsible practice for high-impact decisions.
Have a human review and approve important decisions before they are finalized.
Correct. For high-impact decisions, the responsible practice is to add a human-in-the-loop, where the AI output is not used directly as the final decision but is reviewed and approved by a human.
Skip human review and finalize automatically when the confidence score is high.
Even with a high confidence score, a model can be confidently wrong.
For high-impact decisions, keeping a human review regardless of the score is the responsible practice, so this is incorrect.
Do not disclose the reasoning and notify the person of the result only.
For high-impact decisions, an explanation of why the result was reached is required.
Not disclosing the reasoning goes against transparency and explainability, so this is incorrect.
Handle corrections of misjudgments in a quarterly batch.
Leaving errors in high-impact decisions unaddressed for months allows the harm to grow in the meantime.
Errors in important decisions require a process for prompt correction, so this is incorrect.
For high-impact decisions (medical, credit, hiring, and so on), the responsible practice is to have a human review and approve important decisions (human-in-the-loop). This prevents AI errors and bias from turning directly into serious consequences. Leaving everything to AI without a human, not disclosing the reasoning or limitations, and not recording errors all violate the principles of responsible AI (accountability, transparency, human oversight).