WHEN THE ALGORITHM ALSO EXCLUDES: ALGORITHMIC ABLEISM AND PERSONALIZATION IN INCLUSIVE SPECIAL EDUCATION

Abstract

Artificial intelligence systems used in education are often associated with individual adaptation, accessibility, and teacher support. This study examines an issue that precedes the question of these tools' efficiency: how the categories used to represent students can carry normative expectations regarding ability, learning, communication, and performance. It is a theoretical, qualitative, bibliographic, and documentary study, conducted as a problem-oriented critical narrative review, featuring a targeted, non-exhaustive survey updated through August 2026. The analysis links studies on learner models, algorithmic harm, ableism, and automated advising with Brazilian Special Education documents. The findings indicate that personalization requires selecting attributes and transforming that selection into inferences; thus, a useful recommendation can become reductionist when it extends beyond what the data warrant or when it adopts a specific mode of operation as an implicit standard. It is also observed that the presence of human judgment does not, in itself, eliminate algorithmic influence. Based on this synthesis, the article outlines seven critical-pedagogical criteria: representation, purpose, context, comprehensibility, contestability, proportionality, and participation. The analysis suggests that inclusion depends less on the "personalized" label and more on the quality of the criteria, the limits of the inferences, and the way the recommendation is integrated into contextual pedagogical assessment.