About the Authors and This Research

Lead author Mateusz Dolata is a researcher at the University of Zurich's Department of Informatics, specializing in human-computer interaction, digital ethics, and the societal impacts of algorithmic systems. With an h-index of 16 and 1,321 cumulative citations, Dolata is a recognized voice in European information systems research. Co-author G. Schwabe, also from the University of Zurich, contributes expertise in digital service design and sociotechnical systems. Since its publication in 2023, this paper has accumulated 9 citations, generating sustained discussion in the qualitative research community on algorithmic fairness.

A key methodological distinction: rather than examining code or performance metrics, Dolata and Schwabe analyze how algorithmic behavior shapes public moral discourse and, in turn, how social forces compel algorithmic adjustments. This socio-constructivist lens offers Taiwan enterprises a perspective often missing from purely technical AI governance frameworks—one that is increasingly demanded by ISO 42001 and the EU AI Act alike.

Core Research Findings: Algorithms as Moral Discourse Architects

The study's central question is how society constructs notions of fairness when algorithms make consequential decisions. The Brooklyn Subway Shooting case reveals that Uber and Lyft's dynamic pricing algorithms, by automatically generating a fivefold fare increase during a public safety crisis, became the primary catalyst for an extensive public debate on corporate ethics, platform accountability, and algorithmic justice—even when the algorithms themselves were not explicitly named in public discourse.

Finding One: Algorithmic Outputs Drive Moral Discourse Without Direct Attribution

The research demonstrates that algorithmic decisions initiate and structure fairness debates independently of corporate communications. The mere existence of a fivefold price surge—regardless of the technical logic behind it—framed public expectations, evoked solidarity with affected groups, and mobilized moral advocacy movements. This has a direct implication for Taiwan enterprises: the "technical neutrality" of an algorithm provides no shield against social and regulatory backlash. EU AI Act Article 13 requires that high-risk AI systems provide sufficient transparency for users to understand and contest outputs. Dolata and Schwabe's findings make clear that this transparency requirement is not bureaucratic box-checking—it is a prerequisite for maintaining social license to operate.

Finding Two: Fairness Is a Socio-Algorithmic Construction, Not a Fixed Standard

The authors propose a Socio-Algorithmic Construction Theory, arguing that fairness norms are no longer purely social constructs. Instead, they emerge from a bidirectional interaction: algorithms shape social expectations and moral frameworks, while social pressure compels algorithmic adjustment. In the Brooklyn case, Uber ultimately revised its surge pricing policy in response to public outcry—a concrete demonstration of social forces recalibrating algorithmic behavior. For enterprises implementing ISO 42001's clause 6.1.2 risk assessment requirements, this theory demands a broadened scope: risk evaluation must include not only technical bias metrics but also the dynamic social contexts in which algorithmic decisions will be interpreted and contested. Algorithmic Impact Assessments should incorporate this socio-algorithmic dimension as a standard component.

Finding Three: Methodological Contribution and Its Limits for Taiwan Application

From Winners Consulting's advisory perspective, it is important to note the study's methodological boundaries. The research is grounded in a single-event qualitative discourse analysis, conducted predominantly in English-language media in a New York context. Taiwan's ride-hailing and delivery platforms (such as Uber Eats and foodpanda) operate within distinct labor structures, consumer culture norms, and regulatory environments. Direct extrapolation of conclusions without localization carries risk. Taiwan enterprises should use the Socio-Algorithmic Construction framework as a conceptual scaffold and supplement it with locally grounded fairness evaluations specific to Taiwan's platform economy and the expectations of Taiwan's regulatory bodies and consumer base.

Implications for Taiwan AI Governance: Algorithmic Fairness as Compliance Obligation

This research elevates algorithmic fairness from an ethical aspiration to a concrete governance imperative. Three regulatory frameworks create immediate obligations for Taiwan enterprises.

ISO 42001: Clause 6.1.2 requires systematic assessment of AI system risks to individuals and groups, explicitly including unfair treatment risks. The Socio-Algorithmic Construction framework informs a more robust interpretation of this clause—risk assessment must capture not only measurable technical bias but also context-dependent social perceptions of fairness that may emerge in crisis or contested situations.

EU AI Act: Effective since 2024, the Act classifies dynamic pricing systems and automated hiring tools as high-risk AI applications, requiring explainability of decision logic, continuous monitoring mechanisms, and meaningful human oversight. The Brooklyn case illustrates precisely what happens when these safeguards are absent: algorithmic decisions made without explainability provisions become lightning rods for public outrage and regulatory scrutiny. Taiwan enterprises with EU market exposure face direct compliance obligations; those without should treat the EU AI Act as a leading indicator of the regulatory direction Taiwan's own frameworks will follow.

Taiwan AI Basic Law (台灣 AI 基本法): The 2024 draft explicitly enshrines fairness and accountability as core principles for AI deployment. Automated systems affecting consumer pricing, employment decisions, or public resource allocation must establish clear accountability mechanisms—ensuring algorithmic behavior can be explained, challenged, and subject to human correction. The research finding that social forces successfully compelled Uber to modify its algorithm is a case study in why these accountability mechanisms need to be proactive, not reactive.

How Winners Consulting Supports Taiwan Enterprises

積穗科研股份有限公司（Winners Consulting Services Co. Ltd.）provides comprehensive support for Taiwan enterprises building AI governance systems aligned with ISO 42001 and EU AI Act requirements, including AI risk classification assessments and compliance frameworks under Taiwan's AI Basic Law. Based on the socio-algorithmic fairness challenge identified in this research, we recommend three concrete actions:

1. Conduct a Socio-Contextual Algorithmic Impact Assessment: For all automated decision systems affecting pricing, hiring, credit, or resource allocation, establish a risk assessment process per ISO 42001 clause 6.1.2 that explicitly includes a "social context fairness perception" dimension—evaluating how algorithmic outputs may be interpreted across different stakeholder groups and crisis scenarios, not merely measuring technical bias metrics.
2. Design Explainability and Public Communication Mechanisms: In response to EU AI Act transparency requirements for high-risk AI, develop clear decision explanation interfaces and stakeholder communication protocols for automated systems. Ensure that when algorithmic decisions generate controversy, the enterprise can credibly and transparently articulate the system's logic, boundaries, and oversight mechanisms.
3. Build a Rapid Review and Adjustment Process for Social Pressure Scenarios: Drawing on the research finding that Uber was compelled to revise its pricing policy under social pressure, proactively design an internal rapid-response governance process for situations where algorithmic decisions trigger public controversy. This process should be embedded within ISO 42001's continuous improvement requirements, ensuring human oversight can effectively intervene before reputational damage becomes irreversible.

Frequently Asked Questions

Under what circumstances might our dynamic pricing algorithm be deemed "unfair" and trigger compliance risk?

Based on the Socio-Algorithmic Construction Theory from this research, algorithmic fairness is not determined solely by technical benchmarks—it is co-shaped by social context, the characteristics of affected groups, and public expectations at the moment of decision. When automated systems maintain purely profit-driven logic in crisis or vulnerable-population contexts (such as public safety emergencies or natural disasters), they risk being judged socially unfair even if technically legal. Under the EU AI Act, automated pricing systems affecting essential consumer services may be classified as high-risk AI, requiring explainability, human oversight, and transparency disclosures. Taiwan enterprises should proactively conduct Algorithmic Impact Assessments to map potential social controversy points across different operating scenarios, not just under normal conditions.

What are the most common compliance challenges Taiwan enterprises face when implementing algorithmic fairness under ISO 42001?

Three challenges consistently appear in our advisory engagements. First, enterprises tend to define fairness narrowly as a technical metric (model accuracy or error rates), missing ISO 42001 clause 6.1.2's broader requirement for stakeholder impact assessment—leading to risk evaluations that are formally compliant but substantively inadequate. Second, cross-functional governance gaps exist between legal, data science, and business teams, who often hold divergent definitions of fairness, making it difficult to establish coherent governance standards. Third, fairness assessment is treated as a one-time pre-deployment exercise rather than a continuous monitoring obligation—failing to capture how algorithmic behavior evolves across changing social contexts. Taiwan's AI Basic Law draft similarly emphasizes ongoing accountability, requiring enterprises to embed fairness evaluation as a routine governance process.

What are ISO 42001's specific requirements on algorithmic fairness, and what are the practical implementation steps and timelines?

ISO 42001 clause 6.1.2 requires enterprises to identify potential risks of AI systems to individuals, groups, and society at large, and establish corresponding control measures. Clause 9.1 mandates continuous performance monitoring, including fairness indicators. Practical implementation follows three phases: Phase 1 (months 1–2): Current state diagnosis—inventory existing AI systems and conduct an ISO 42001 gap analysis, identifying fairness-related gaps. Phase 2 (months 3–5): Governance mechanism design—develop Algorithmic Impact Assessment procedures, fairness indicator frameworks, and human oversight protocols. Phase 3 (months 6–12): System implementation, internal audit, and certification preparation. For most mid-to-large Taiwan enterprises, the full process from initiation to ISO 42001 certification requires 7 to 12 months, adjusted based on existing data governance maturity.

How should we evaluate the cost and expected benefits of implementing algorithmic fairness governance mechanisms?

Costs vary by enterprise size and existing governance foundation, but benefits should be assessed across three dimensions. First, compliance benefit: avoiding EU AI Act violations, which can carry penalties of up to 3% to 6% of global annual turnover for serious infractions—making implementation investment significantly less costly than potential fines for Taiwan enterprises with EU market exposure. Second, reputational benefit: the Brooklyn case demonstrates that unmanaged algorithmic fairness risks generate reputational damage in crisis scenarios that is difficult to quantify but typically far exceeds the cost of system improvement. Third, operational efficiency benefit: enterprises that establish systematic AI risk management frameworks report average reductions of 30% to 40% in review time for subsequent AI system deployments, as the governance infrastructure is standardized and reusable. We recommend evaluating the investment decision through a risk-adjusted return lens rather than a pure cost frame.

Why engage Winners Consulting Services Co. Ltd. for AI governance?

Winners Consulting Services Co. Ltd. (積穗科研股份有限公司) is among Taiwan's few consulting firms combining ISO 42001 implementation advisory credentials, EU AI Act compliance expertise, and the capability to translate cutting-edge academic research—such as Dolata and Schwabe's socio-algorithmic fairness theory—into actionable enterprise governance practices. Our advisory team continuously monitors international AI governance research to ensure client frameworks anticipate regulatory evolution, not merely satisfy current requirements. We provide end-to-end services from gap analysis, mechanism design, and personnel training to certification accompaniment, supporting Taiwan enterprises in building ISO 42001-compliant AI management systems within 7 to 12 months while simultaneously addressing dual-track compliance requirements under the EU AI Act and Taiwan's AI Basic Law.

積穗科研：社会－アルゴリズム的公平性の構築――台湾企業のAIガバナンスへの示唆

積穗科研股份有限公司（Winners Consulting Services Co. Ltd.）は、台湾のAIガバナンス分野における専門機関として、企業リーダーに重要な知見をお届けします。価格設定、採用、または与信審査に関わるアルゴリズムが意思決定を行う際、「公平性」は静的な技術的属性ではなく、アルゴリズムと社会が相互に構築する動態的なプロセスです。Dolata と Schwabe が 2023 年に発表した研究は、2022 年 4 月 12 日のブルックリン地下鉄銃撃事件後に配車アプリのアルゴリズムが引き起こした 5 倍の運賃急騰を事例として分析し、アルゴリズムの行動が社会の公平性認識を根本的に再形成することを明らかにしました。ISO 42001 認証および EU AI Act への対応を検討している台湾企業にとって、この研究は不可欠な理論的基盤を提供します。