Course contentsShow
← Machine Learning and Deep Learning
Lesson 3480 of 3,53876. Ethics and Societal ImpactFree lesson

Eliciting Stakeholder Values and Concerns

Techniques for discovering what stakeholders care about: fairness definitions, privacy preferences, usability needs, and harm avoidance.

Eliciting Stakeholder Values and Concerns

You'll learn how to systematically discover what different stakeholders truly care about when your ML system affects their lives.

Why Discovery Techniques Matter

Building on identifying stakeholders and understanding power dynamics, you now need structured methods to uncover what people actually value. Different groups care about different things—and they may not volunteer these concerns unless you ask the right questions.

Core Techniques

Structured Interviews and Workshops

Hold focused conversations with representative stakeholders. Ask open-ended questions: "What would 'fair' look like to you in this system?" Different groups may define fairness differently (equal accuracy vs. equal opportunity vs. equal outcomes).

Value-Sensitive Design Probes

Use concrete scenarios and prototypes. Show stakeholders examples of system behavior and ask: "What troubles you here?" or "What would you change?" This reveals unstated assumptions about privacy preferences, acceptable error types, and harm thresholds.

Participatory Methods

Invite stakeholders into co-design sessions. When people sketch interfaces or describe ideal workflows, you discover usability needs they couldn't articulate in abstract terms.

Red Lines and Trade-Off Exercises

Present stakeholders with difficult choices: "Would you accept 2% lower accuracy to eliminate demographic disparities?" This reveals priority hierarchies and which values are non-negotiable versus flexible.

Documenting Findings

Create a values map showing which stakeholders care about what: fairness definitions (statistical parity, equalized odds), privacy preferences (data minimization, consent models), usability needs, and specific harms to avoid (false negatives in medical diagnosis, false positives in hiring).

Key Takeaway: Stakeholders can't design your system for you, but systematic elicitation techniques—interviews, scenarios, co-design, and trade-off exercises—reveal the diverse values and concerns your ML system must respect to avoid real-world harm.