<![CDATA[The application of artificial intelligence (AI) in education can improve efficiency and consistency through automated assessment and personalized learning, yet it can also introduce algorithmic bias that amplifies educational inequities. This study examines how algorithmic bias manifests in AI-based assessment systems and personalized learning, and it proposes an algorithmic fairness framework to support equitable implementation of educational technology. This study applies a library-based desk study using a qualitative descriptive-analytical approach. It systematically collects and documents evidence from indexed journal articles (Scopus, Web of Science, IEEE Xplore, ACM Digital Library), textbooks on AI ethics and educational technology, and policy reports from UNESCO and OECD using targeted keywords, then it selects sources based on relevance and credibility. The analysis uses content analysis that (1) identifies core themes on bias sources, injustice manifestations, and fairness frameworks, (2) categorizes findings across technical, social, and ethical dimensions, (3) critically interprets patterns and gaps, and (4) synthesizes a conceptual framework with practical recommendations. The findings show that bias arises in automated essay scoring (AES) and adaptive learning through unrepresentative training data, proxy variables, and feedback loops that reinforce prior disparities. The proposed framework integrates individual, group, and counterfactual fairness, and it clarifies the ethical trade-offs that emerge when these criteria conflict. Bias mitigation strategies include stronger transparency and documentation, participatory design with affected stakeholders, routine algorithmic audits, human-in-the-loop review for high-stakes decisions, and AI literacy training for educators. This study offers practical guidance for developers, educators, and policymakers to build fairer and more transparent educational AI systems.]]>
Copyrights © 2025
