Norm Elicitation

Norm elicitation is a structured methodology originating from social sciences and participatory design, aimed at systematically uncovering implicit or unstated social norms, values, and expectations within a specific group. In AI governance, its core purpose is to address the critical question of 'whose values should AI align with?' Unlike traditional surveys, it emphasizes a deliberative process where participants engage in deep dialogue about specific AI scenarios, fostering informed reflection to extract nuanced collective consensus. This practice directly supports the 'MAP' function of the NIST AI Risk Management Framework (RMF), which requires organizations to understand the socio-economic and ethical context of AI deployment. It also aligns with ISO/IEC 42001 (AI management system) requirements for identifying stakeholder needs, ensuring that diverse societal perspectives are integrated into the AI lifecycle to mitigate risks of bias, discrimination, and unfairness.

Enterprises apply norm elicitation to translate abstract AI ethics principles into concrete operational practices for managing sociotechnical risks. The implementation involves three key steps. First, 'Stakeholder Mapping & Context Definition': identify communities most impacted by the AI system, especially underrepresented groups, and design relevant interaction scenarios. Second, 'Deliberative Workshop Execution': facilitate structured discussions with representative groups, using AI-generated outputs as prompts to elicit judgments on fairness and appropriateness. Third, 'Norm Synthesis & Technical Translation': analyze qualitative data to synthesize norms into clear principles, which then guide AI model fine-tuning or the creation of safety guardrails. For example, a fintech company used this process to uncover potential biases in its loan approval model, leading to a 15% improvement in fairness metrics and successful passage of an AI ethics audit.

Taiwanese enterprises face three main challenges. First, 'Insufficient Stakeholder Representation': recruiting truly diverse participants who can represent various minority groups (e.g., new immigrants, rural residents) is difficult. Second, 'Lack of Interdisciplinary Talent': there is a shortage of professionals skilled in sociology, ethics, and AI to facilitate and analyze these complex discussions. Third, 'Technical Translation Gap': converting qualitative ethical norms into machine-readable rules for AI models is a significant challenge. To overcome these, companies should partner with local NGOs and academic institutions for recruitment, invest in cross-training internal staff or hire external experts, and establish a dedicated AI ethics committee to bridge the gap between ethical principles and technical implementation. A phased approach, starting with pilot projects, is recommended.

Winners Consulting specializes in norm elicitation for Taiwan enterprises, delivering compliant management systems within 90 days. We have successfully served over 100 local companies. Request a free consultation: https://winners.com.tw/contact