What steps can be taken to prevent bias in AI systems?

• A. Implement bias mitigation strategies and monitoring outcomes to ensure fairness.
• B. Remove all data features.
• C. Rely solely on model complexity.
• D. Ignore monitoring after deployment.

Answer: (A)

Bias can enter AI systems from data, features, and how the model is applied, so preventing it requires both proactive mitigation and ongoing evaluation. Implement bias mitigation strategies across the model lifecycle—preprocessing to balance or de-bias the data, in-processing to impose fairness constraints during learning, and post-processing to adjust outputs—along with monitoring outcomes using fairness and performance metrics. Continuous monitoring after deployment lets you detect data drift, shifts in usage, or new biases as contexts change, enabling timely remediation. This combination of proactive measures and ongoing oversight is why the option that calls for bias mitigation plus ongoing monitoring is the best choice. The other options either strip away essential information, overemphasize model complexity, or skip monitoring, all of which can introduce or fail to address bias.