"Minimizing bias requires diverse datasets, use of fairness metrics, and human oversight throughout development and deployment." This clearly confirms that the proper approach is a combination of diverse data, fairness metrics, and oversight.
References: Microsoft Responsible AI and GitHub Copilot fairness guidelines.