AI Feedback Summarization Biases

Let’s say we’re rolling out an AI tool that summarizes written feedback from managers to employees, and we want to be proactive about uncovering any biases in how the tool generates these summaries.

How would you systematically identify and evaluate potential biases in the tool’s outputs?