AI as Cognitive Accessibility: Rethinking Neurodivergent Learning in the Age of Artificial Intelligence

IntroductionBeyond the Narrow Questions

Artificial intelligence is rapidly reshaping education, but much of the current debate remains concentrated on a limited set of questions: Does AI improve efficiency? Does it encourage cheating? Will students become dependent on it? Can it replace traditional methods of teaching and learning?

These questions are legitimate, but they may also be too narrow.

Since February 2026, I have been developing a research project titled Unlocking Potential: Artificial Intelligence, Neurodiversity and the Future of Learning; AI-Mediated Cognitive Accessibility and the Hypothesis of Parallel Cognition. The first consolidated conceptual paper has now been completed following an eight-question research process, critical review and citation audit. Its central question is deliberately different from much of the mainstream AI-in-education discussion:

This distinction is especially important in neurodivergent education. For some learners, particularly those with ADHD, autism and other forms of neurodivergence, difficulty in education does not necessarily indicate a lack of understanding. The difficulty may lie in the route through which understanding must be expressed. A learner may grasp a concept, make complex connections and generate original ideas, yet struggle to sequence those thoughts in the rigid form expected by a worksheet, timed assessment or conventional classroom exchange.

The educational system often treats the visible output as the primary evidence of cognition. This research challenges that assumption.

Part 01When the Route Becomes the Barrier

Traditional education frequently requires learners to process information in a predefined sequence: listen, retain, organise, write, recall and reproduce. These stages are commonly treated as though they form a neutral route to demonstrating knowledge. They are not always neutral.

For some learners, the route itself may create unnecessary cognitive friction. Working-memory limitations, attentional variation, processing-speed differences, executive-function demands and difficulties translating non-linear thought into linear language can interfere with performance even when conceptual understanding is present.

This creates a problem of construct validity. If an assessment is intended to measure a learner’s understanding of history, science or mathematics, but performance is significantly constrained by handwriting speed, sequencing difficulty, working-memory pressure or the ability to organise thoughts under time constraints, then the assessment may partly measure something other than the intended construct.

This is not an argument for lowering standards. It is an argument for examining more carefully what, exactly, we are measuring.

AI introduces a new possibility here. Rather than functioning only as a content generator, it may serve as a cognitive-accessibility scaffold: a tool that helps a learner hold onto a thought, reorganise fragmented ideas, rehearse privately, clarify instructions or translate complex internal reasoning into a form that can be communicated. That possibility deserves more serious academic attention than it currently receives.

Part 02AI-Mediated Cognitive Accessibility

This research examines the concept of AI-mediated cognitive accessibility. The idea is relatively straightforward: artificial intelligence may help reduce unnecessary barriers between a learner’s cognition and the learner’s ability to use or communicate that cognition.

The critical issue is not whether AI has assisted the learner. The critical issue is what kind of assistance has taken place.

There is a substantial difference between AI generating an answer on behalf of a learner and AI helping the learner organise an answer they already understand. Current educational discourse does not always distinguish adequately between these two forms of use. That distinction is central to this study.

Part 03The Hypothesis of AI-Mediated Parallel Cognition

One of the central hypotheses developed within the completed conceptual paper is AI-mediated parallel cognition. It proposes, provisionally, that AI may operate as an external interactive workspace that helps preserve and coordinate several learner-originated, semantically related lines of inquiry while human attention moves sequentially among them. It is not presented as an established cognitive phenomenon; it remains a falsifiable hypothesis requiring empirical testing against credible non-AI alternatives.

In practice, this might occur when a learner is thinking in several directions simultaneously but cannot preserve every strand of thought long enough to organise it. An AI system may help capture fragments, compare relationships, ask clarifying questions and support the external organisation of those ideas.

The cognition remains human; the technology assists with preservation, structure and expression.

This is especially relevant where a learner’s thought process is associative, rapid or non-linear. Conventional educational systems have historically privileged linear production: one answer after another, one paragraph after another, one reasoning step after another. Neurodivergent cognition does not always operate in that sequence.

That raises an uncomfortable but necessary question: Should education require every learner to think through the same route, or should it become better at distinguishing between the route used to demonstrate knowledge and the knowledge itself? This is where the research challenges the status quo.

Part 04Challenging the Assumption That Difficulty Is Always Desirable

Education often treats difficulty as inherently valuable. Some difficulty is valuable. Learners need challenge, practice, productive struggle and opportunities to retrieve knowledge independently. Educational psychology provides strong reasons for preserving effort and avoiding excessive assistance.

However, not every difficulty contributes to learning. There is an important distinction between productive difficulty and incidental difficulty.

A mathematics learner should struggle with mathematical reasoning. A learner studying history should be challenged to interpret evidence. A student writing an argument should be required to develop and defend a position. But if the primary obstacle is retaining several instructions simultaneously, physically organising ideas before they disappear or translating an already-understood concept into an imposed communication format, then the learner may be expending cognitive resources on a barrier that is not central to the learning objective.

Educational systems have often treated these barriers as though overcoming them were automatically part of learning. That assumption deserves scrutiny. AI may force education to confront it.

Part 05Support Must Not Become Substitution

The case for AI-mediated accessibility does not mean that all AI use is educationally beneficial. Dependency remains a legitimate concern. If learners consistently outsource interpretation, reasoning, writing and decision-making to AI, then the technology may weaken rather than strengthen learning.

The study therefore places substantial emphasis on the distinction between scaffolding and substitution. Useful scaffolding should gradually support greater learner independence. A learner may initially need AI to organise ideas into a coherent structure. Over time, the learner should increasingly understand that structure and become more capable of producing it independently.

The aim is not polished AI-assisted output. The aim is human mastery.

This requires educators to evaluate more than the final product. Teachers may need to ask learners to explain their reasoning orally, reconstruct arguments without assistance, identify errors, defend decisions and demonstrate retained understanding. AI should not make verification of learning less important. It makes verification more important.

Part 06Why the Current Policy Response May Be Too Simple

Many institutions initially responded to generative AI through prohibition. This response is understandable. Schools and universities faced a technology capable of producing essays, solving problems and generating apparently sophisticated answers within seconds.

However, blanket prohibition risks placing very different forms of AI use into the same category. The learner who asks AI to complete an assignment and the learner who uses AI to organise their own fragmented thoughts are not necessarily engaged in the same educational act. Yet existing rules may treat them as equivalent.

This distinction has implications for assessment, academic integrity, special educational needs, disability accommodation, curriculum design and teacher education. This research does not propose that educators abandon caution. It argues that caution should become more precise.

Part 07The Teacher Remains Central

A future of AI-supported learning does not remove the need for teachers. It increases the importance of professional judgement.

Teachers remain responsible for determining the intended learning construct, deciding when assistance is appropriate, verifying that understanding is genuine and reducing support as capability develops. Human relationships, observation, dialogue and contextual knowledge remain essential because an AI system cannot independently determine whether a learner has mastered a concept in the educational sense that matters.

For SEN educators in particular, the opportunity is not simply to add another digital tool to an already crowded classroom. It is to reconsider whether some technologies can be used deliberately as accessibility mechanisms while preserving challenge, accountability and independence.

Part 08From Conceptual Research to the Next Stage

The completed conceptual phase examines these questions across school-age, tertiary and adult learning contexts. It brings together cognitive accessibility, construct validity, cognitive offloading, self-efficacy, Universal Design for Learning, learner agency, dependency risk, private rehearsal, academic integrity and the distinction between assisted performance and independent mastery. The next stage is not to assume that the framework is correct, but to subject its claims to empirical testing.

The consolidated paper also includes a deliberately limited legislative and educational-policy advisory. It does not propose a new law or an immediate AI mandate. Instead, it argues that lawmakers, education departments, universities, assessment bodies and other relevant role players should begin creating legitimate space for research and carefully governed consideration of AI as cognitive accessibility. The key policy challenge is to distinguish accessibility support from cognitive substitution while preserving assessment validity, privacy, fairness, human oversight and learner responsibility.

The research intentionally challenges established assumptions while retaining strict evidentiary boundaries. Neurodivergent learners have too often been required to adapt themselves to educational systems designed around comparatively narrow expectations of attention, sequencing, communication and performance. Emerging AI tools provide an opportunity to ask whether some of those systems should also adapt to the learner — but any such change must ultimately be justified by evidence of genuine learning, ownership and transfer.

ConclusionA Different Question for the AI Era

The purpose of this work is not to argue that AI should replace teachers, traditional learning or human effort. The completed conceptual paper instead asks whether education may be overlooking an emerging accessibility function because very different forms of AI use are too often evaluated through the same lens.

If AI can help a learner retain an idea that would otherwise disappear, organise reasoning that already exists, rehearse privately before participating publicly or communicate understanding more accurately, then the educational question changes.

The question is no longer only, “Did AI help produce this?” It must also become:

That question now moves from conceptual argument toward testing and wider academic discussion — and it may require education to reconsider some of its most familiar assumptions about ability, assistance and learning.

About the author

Francois Greyling

SEN Educator and Tutor, Real Teachers Pty Ltd · ITA Member & Project Manager

Francois Greyling is an SEN educator and tutor at Real Teachers Pty Ltd and a Project Manager of the International Teachers Association. His research focuses on artificial intelligence, neurodiversity and cognitive accessibility, including the hypothesis of AI-mediated parallel cognition. His current work examines how emerging technologies may challenge conventional assumptions about learning, assessment and the recognition of learner capability while preserving human agency and independent mastery.

• Neurodiversity
• Special Educational Needs
• Cognitive Accessibility
• AI Research

Connect on LinkedIn