Lead with transparency: Clearly communicate how and why AI gets used in your products, services, and operations. Ongoing transparency demonstrates commitment to ethics.

Perform impact assessments: Analyze the potential risks AI systems pose to stakeholders like customers or employees. Assessments should cover unintended biases, privacy implications, and effects on decision autonomy.

Empower oversight teams: Assign people to assess AI risks, audit systems, and monitor compliance. Provide oversight teams the authority and access required to ensure accountability.

Minimize biased data: Scrutinize training data to detect and mitigate any biases or underrepresentation that could propagate unfair outcomes. Seek diverse input on data evaluation.

Test for unwanted bias: Proactively test AI systems for potential discrimination across race, gender, age, disabilities, and other protected characteristics not part of the decision logic.

Enable human oversight: Keep humans involved in the decision loop for high-risk scenarios instead of allowing full automation. Humans can evaluate results for fairness and reasonableness.

Focus on inclusiveness: Seek diverse perspectives when designing AI solutions. Inclusive teams build more thoughtful, empathetic, and ethical systems.

Respect user agency: Carefully consider when automation may overly constrain user autonomy or freedom of choice. Build in checks and overrides giving users control.

Plan responsible rollbacks: Devise winding down procedures for experimental or high-risk AI systems in case harms emerge. Avoid creating dependency on systems that prove ethically problematic.

Continual improvement process: Institute data and ethics reviews as integral to development lifecycles rather than one-off activities. This enables ongoing refinement guided by ethics.