Abstract
Artificial intelligence (AI) has become increasingly integrated into education and holds promise for supporting teachers and students alike. AI has the potential to support special education teachers in particular, who have instructional- and compliance-related tasks in their daily practice that can lead to burnout. Yet, limited empirical research has examined how special education teachers perceive the quality and usability of AI-generated materials and the feasibility of using AI tools in their everyday practice. Grounded in a human-in-the-loop approach and informed by Cognitive Load Theory (CLT; Chandler & Sweller, 1991; Sweller, 2020; Sweller et al., 1998), Cognitive Apprenticeship (Collins et al., 1989), and the Technological Pedagogical Content Knowledge conceptual framework (TPACK; Mishra & Koehler, 2006), the purpose of this case study was to (a) investigate the quality of AI-generated instructional and progress monitoring materials; (b) examine special education teachers’ preferences and perceptions regarding the feasibility, usability, and quality of these materials and platforms; and (c) explore the interplay of engagement with generative AI platforms and teacher expertise through a pilot workshop.
With experienced special education teachers as participants, the case study occurred in three distinct phases with the first two phases employing explanatory sequential mixed methods and the third phase employing convergent mixed methods. In the first phase, seven participants rated anonymized human-created and AI-generated materials from three platforms (ChatGPT, CoPilot, and Magic School). These ratings informed the second phase where six participants participated in either a semi-structured interview or focus group. The results and findings from the first two phases informed the design of the pilot workshop, which occurred in the third phase with four participants. During the workshop, qualitative data was derived from participants reflecting on their experience using either ChatGPT or Magic School. At the end of the workshop, participants completed a survey assessing their perceived quality, feasibility, and usability of the materials they created during the workshop and the AI platforms they used.
Results and findings from all three phases of the study suggest generative AI can support, but not replace, special education teachers with some contingencies. Quantitative findings indicated that AI-generated instructional and progress monitoring materials were rated as comparable in quality to human-created materials, and participants were largely unable to distinguish between them. Qualitative analyses revealed that teachers positioned themselves as evaluators and decision makers, emphasizing the centrality of professional judgment, contextual knowledge, and iterative engagement when using AI. While participants acknowledged the potential for generative AI to streamline aspects of their work, perceived efficiency was contingent on training, familiarity, and task type rather than inherent tool capabilities. The pilot workshop further demonstrated that scaffolded professional development supported feasibility judgments and exploratory engagement, reinforcing the importance of teacher expertise within a human-in-the-loop process. Limitations include a small sample size, a focus on only three generative AI platforms, and the rapidly evolving nature of AI tools and models which constrain generalizability and replication. Implications highlight the need for small-scale longitudinal research respective to generative AI model updates; disaggregation of AI use by tasks; and, above all else, targeted training that centers teacher expertise and prioritizes ethical, human-centered integration of generative AI in special education practice.
