Which attribute in trustworthy AI addresses bias management?

• A. Fair with Harmful Bias Managed
• B. Real-Time Adaptation
• C. Maximum Throughput
• D. Closed-Source Licensing

Answer: (A)

Bias management in trustworthy AI focuses on ensuring fairness and actively handling biases that can appear in data and model decisions. The attribute described as Fair with Harmful Bias Managed embodies this by requiring the system to be fair and to include controls that detect, audit, and mitigate harmful biases. It covers steps like examining training data for representation gaps, applying fairness metrics (such as demographic parity or equalized odds), and implementing mitigation techniques, along with ongoing monitoring and governance to prevent biased outcomes from creeping in over time. This emphasis on identifying and addressing bias distinguishes it from other attributes that deal with speed, adaptation, or licensing. Real-Time Adaptation is about updating the model as conditions change, Maximum Throughput concerns performance and efficiency, and Closed-Source Licensing relates to ownership terms.