Intersectionality

Coined by scholar Kimberlé Crenshaw, intersectionality is an analytical framework explaining how an individual's multiple social identities—such as race, gender, class, and disability—overlap to create unique, compounded experiences of discrimination. It argues that single-axis analysis (e.g., focusing only on gender) is insufficient. In enterprise risk management, intersectionality is crucial for assessing algorithmic bias. While not a specific ISO standard, its principles are embedded in key regulations. The EU AI Act, for instance, prohibits discrimination based on intersecting grounds (Article 5), and the NIST AI Risk Management Framework (AI RMF 1.0) mandates evaluating AI impacts on various demographic subgroups to ensure fairness and equity. It provides a lens to uncover hidden risks that only become apparent when analyzing the intersection of different vulnerabilities.

Enterprises can apply intersectionality in AI risk management through concrete steps: 1. **Intersectional Impact Assessments (IIA):** Before deploying an AI system, evaluate its potential risks not just for broad categories but for specific subgroups, such as older women of color. This practice extends the spirit of GDPR's Data Protection Impact Assessment (DPIA) under Article 35 to cover compounded social harms. 2. **Dataset Auditing and Mitigation:** Systematically audit training data for the underrepresentation of intersectional groups. Use techniques like stratified sampling or synthetic data generation to correct imbalances. For example, a company found its hiring AI unfairly penalized candidates who were women from non-elite universities, a bias invisible when analyzing gender or education alone. Correcting this improved hiring diversity by 15%. 3. **Diverse Red Teaming:** Assemble a team with diverse intersectional backgrounds to stress-test the AI system. This team can identify biases and failure modes that a homogeneous team would likely miss, leading to a measurable reduction in fairness-related incidents.

Taiwan enterprises face several key challenges when implementing intersectionality: 1. **Data Scarcity and Privacy:** Local datasets often lack the granular demographic labels required for intersectional analysis, partly due to restrictions under Taiwan's Personal Data Protection Act (PDPA). 2. **Regulatory Ambiguity:** Current laws, like the Act of Gender Equality in Employment, do not explicitly mandate an intersectional approach for algorithmic systems, creating a lack of clear compliance incentives for businesses. 3. **Talent and Skill Gaps:** Effective intersectional analysis requires a blend of expertise in data science, social sciences, and AI ethics—a skill set that is currently scarce in the local market. **Solutions:** * **Priority Action:** Establish an AI ethics board and pilot Intersectional Impact Assessments for high-risk systems. Use Privacy-Enhancing Technologies (PETs) to analyze sensitive data compliantly. * **Strategy:** Benchmark internal policies against global standards like the EU AI Act. Partner with expert consultants like Winners Consulting to train technical and legal teams, bridging the knowledge gap. * **Long-term Goal:** Foster industry-academic collaboration to develop localized fairness metrics and best practices.

Winners Consulting specializes in Intersectionality for Taiwan enterprises, delivering compliant management systems within 90 days. Free consultation: https://winners.com.tw/contact