The Scaffold That Disappears: AI Tools for Executive Function in Science Classrooms

Recent advances in AI have introduced instructional tools that support planning, organization, data analysis, and communication (Holmes et al., 2019). When paired with thoughtful instructional design, AI can serve as a cognitive scaffold, removing procedural demands from a student's working memory so that cognitive effort can focus on the content itself. Cognitive Load Theory holds that instruction is most effective when it reduces unnecessary demands on working memory, freeing learners to direct their attention toward reasoning and problem-solving (Sweller, 2024).

The following scenarios draw on a classroom-tested middle school inquiry unit, AI in Focus: Micro-Worlds and Macro-Impact (Szentmiklosi, Garza, Bush, & Marino, in press), which positioned students as campus eco-researchers who collected and analyzed biological samples using AI classification tools.

In the first scenario, a student with ADHD struggled with a field sample collection protocol involving multiple sequential phases. The teacher introduced a digital checklist generated through an AI prompt, then reviewed and edited it for vocabulary and grade-level clarity. The checklist made the sequence visible. Each step was short and action-oriented, with a brief reflection prompt at the end of each phase: Did I complete every item? Is there anything I need to fix? Research on Self-Regulated Strategy Development supports this design, finding that students benefit from explicit support for planning, monitoring progress, and evaluating their own performance (Rogers et al., 2020). The checklist did not do the science. It organized the process so that attention could focus on observation and analysis. By the final phase of the unit, the student had enough data to build a visual infographic comparing biodiversity across two sample sites and was the first in the class to note a difference in organism diversity between locations (Garza et al., in press).

The second scenario involved a student with a specific learning disability in reading. Dense AI-generated classification results created a literacy barrier during the analysis phase. The teacher introduced Microsoft Immersive Reader and Read & Write for Google Chrome to provide text-to-speech and vocabulary support. Browser-based accessibility extensions of this kind can reduce literacy barriers while maintaining grade-level content expectations (Ok & Rao, 2019). Because these tools were available to every student, no one was singled out for using them. For the final product, the student chose a narrated video combining images, captions, and a spoken explanation of the data. Multimedia formats that allow students to express their understanding through audio and visual annotations still require scientific reasoning and the integration of evidence. They shift the mode of access, not the intellectual demand (Ok & Rao, 2019).