Identify a potential data source for AI-based performance appraisal and discuss a bias that could arise from it.

Identify a potential data source for AI-based performance appraisal and discuss a bias that could arise from it.

• A. Annual salary data; bias risk includes salary inversion.
• B. Email or chat communications; bias risk includes misinterpreting tone or sarcasm, and language/style differences across groups.
• C. Attendance logs; bias risk includes weather effects.
• D. Personal hobbies; bias risk includes entertainment preferences.

Answer: (B)

Relying on email or chat communications as input data for AI-based performance appraisal hinges on capturing everyday work interactions that reflect how people perform and collaborate. But a big bias risk is that the AI may misread linguistic cues like tone, sarcasm, or humor. Tone and communication style vary widely across individuals and groups—language, culture, dialect, and even organizational norms shape how messages come across. The model might infer lower performance from a terse or direct style in one group while interpreting a more expansive style in another as higher engagement, even if the actual work outcomes are the same. This potential for misinterpretation creates systematic bias that can unfairly advantage or disadvantage certain employees. To mitigate it, pair such data with objective performance indicators and context, ensure diverse training data, and involve human judgment to interpret communications within their real-world context.

Relying on email or chat communications as input data for AI-based performance appraisal hinges on capturing everyday work interactions that reflect how people perform and collaborate. But a big bias risk is that the AI may misread linguistic cues like tone, sarcasm, or humor. Tone and communication style vary widely across individuals and groups—language, culture, dialect, and even organizational norms shape how messages come across. The model might infer lower performance from a terse or direct style in one group while interpreting a more expansive style in another as higher engagement, even if the actual work outcomes are the same. This potential for misinterpretation creates systematic bias that can unfairly advantage or disadvantage certain employees. To mitigate it, pair such data with objective performance indicators and context, ensure diverse training data, and involve human judgment to interpret communications within their real-world context.