Human Intervention in Artificial Intelligence-Based Automated Decision-Making
European Constitutional Law Association · 30.04.2026 · 50 · 161-187
DOI: 10.21592/eucj.2026.50.161 ↗Kurz gefasst
Fachartikel in der Zeitschrift European Constitutional Law Association, online erschienen am 30.04.2026. Verfasst von Aeryung Jung.
Abstract (Originalsprache)
This study examines the constitutional implications arising from the increasing use of artificial intelligence (AI)-based automated decision-making (ADM) and aims to clarify the significance and limits of the “right to human intervention” in safeguarding human dignity and the right to self-determination. While AI-driven decision-making systems are widely utilized across various sectors—such as finance, employment, and social welfare—due to their efficiency and consistency, they pose significant risks to fundamental rights, as they operate through opaque algorithmic structures and data-driven inferences that can substantially affect individuals’ lives. First, this study analyzes the decision-making process of AI systems, which consists of data collection and preprocessing, model training, inference, and output generation, and highlights the structural reduction or formalization of human involvement throughout these stages. In particular, AI systems are characterized not merely by automation but by autonomy and inference capabilities, generating new information beyond predefined rules. This feature gives rise to qualitatively different risks compared to traditional human decision-making. Such risks include reliance on spurious correlations, infringements of “predictive privacy,” probabilistic assessments based on group characteristics, and a lack of explainability due to the so-called “black box” effect. These phenomena may reduce individuals to objects of data processing and deprive them of opportunities for self-regulation, thereby threatening human dignity. Against this background, the study reviews the legal framework governing automated decision-making under Article 22 of the General Data Protection Regulation (GDPR), focusing on the legal basis of the right to human intervention. It further analyzes the Court of Justice of the European Union’s Schufa judgment, which provides important interpretative guidance on the scope of “automated decisions” and the requirements for “meaningful human intervention.” The judgment emphasizes that human involvement must be substantive, involving genuine decision-making authority and critical assessment, rather than merely formal or symbolic participation. Furthermore, this study explores the multi-layered regulatory approach to ADM by comparing the GDPR with the human oversight provisions under the EU Artificial Intelligence Act (AI Act). In particular, the AI Act’s requirement of human oversight throughout the lifecycle of high-risk AI systems—from design to operation and ex post control—represents a significant development that complements the individual rights-based approach of the GDPR. In conclusion, human intervention in AI-based automated decision-making should not be understood as a mere technical safeguard but as a constitutional requirement essential to protecting human dignity and the right to self-determination. Accordingly, it is necessary for the domestic legal system to explicitly recognize the right to human intervention and to establish institutional mechanisms that ensure substantive oversight and accountability beyond formalistic involvement.
Zitieren
APA
Jung, A. (2026). Human Intervention in Artificial Intelligence-Based Automated Decision-Making. European Constitutional Law Association, 50, 161-187. https://doi.org/10.21592/eucj.2026.50.161
Vancouver
Jung A. Human Intervention in Artificial Intelligence-Based Automated Decision-Making. European Constitutional Law Association. 2026;50:161-187. doi:10.21592/eucj.2026.50.161
BibTeX
@article{Jung2026,
author = {Jung, Aeryung},
title = {Human Intervention in Artificial Intelligence-Based Automated Decision-Making},
journal = {European Constitutional Law Association},
year = {2026},
volume = {50},
pages = {161--187},
doi = {10.21592/eucj.2026.50.161}
}
EUVIMED