Contemporary special education research emphasizes artificial intelligence as a tool for automating pedagogical administration and identifying student deficits more efficiently. This paper presents an alternative theoretical framework, Genius Recognition Through AI-Enabled Dialogue (GRAID), which reconceptualizes the role of artificial intelligence in special education. Drawing upon epistemic justice theory, neurodiversity studies, and phenomenological philosophy, this paper argues that current AI implementations, despite beneficial intentions, perpetuate systematic epistemic injustice by automating the diagnostic gaze that renders disabled students' ways of knowing invisible. GRAID proposes a theoretical inversion: rather than using AI to identify deficits, AI should liberate educators from compliance bureaucracy, creating temporal and cognitive space for genuine dialogue with students about their actual thinking processes. This framework articulates four interconnected theoretical dimensions: first, the concept of diagnostic epistemic closure and its displacement through sustained dialogue; second, neurodiversity reconceived not as a disorder but as a cognitive architecture constituting measurable forms of genius—systemic thinking, spatial reasoning, and rapid ideation; third, phenomenological dialogue as a practice of recognition distinct from conventional pedagogical interaction; and fourth, transformative identity formation that repositions disabled students from subjects of deficit discourse to intellectual agents. The paper further develops original conceptual innovations including the Genius-Deficit Paradox, the Epistemic Paradox of Diagnosis, and a relational ontology of disability. This theoretical framework offers philosophical foundations for educational justice without requiring empirical validation, contributing new conceptual architecture for understanding intelligence, recognition, and technology's proper role in transforming special education.