Embedded Files
What does the brain actually tell us about meaningful AI support?
What the Research Says
What the Research Says
This OER asks whether AI learning tools can do more than respond to student performance. When a learner is stuck, relying on hints, avoiding a task, or producing stronger work with AI, we need to ask: why is this happening?
Research on learning, cognition, and AI helps us look beneath the surface. Learning is not only about completing tasks or getting correct answers. Students also need opportunities to retrieve information, manage cognitive load, monitor their understanding, and transfer skills independently.
Because this OER is an opportunity forecast, the research is not included only to explain learning. It is used to judge whether neuroscience-informed AI is a meaningful emerging market opportunity, and what conditions would make these tools educationally valuable rather than just technologically impressive.
The research below supports this OER’s main argument: the strongest opportunity is not simply more AI support, but better-designed AI support that builds cognitive independence over time.
Learner Variability: The Starting Point
Learner Variability: The Starting Point
Learners do not all process, remember, communicate, or show understanding in the same way. Universal Design for Learning emphasizes that learner variability should be expected from the beginning, not treated as an exception after students struggle (CAST, 2024).
Why this matters for AI
AI learning tools need to respond to more than visible performance. A student’s struggle may come from attention, memory, language, sensory processing, task design, or self-regulation rather than a lack of understanding.
Connects to: Activity 2, where participants compare learner profiles and consider what a standard AI tutor might miss.
Image generated with ChatGPT by OpenAI, 2026.
Cognitive diversity includes many ways of processing, remembering, communicating, regulating, and demonstrating understanding. This is the starting point for designing AI tools that support learner variability rather than assuming one “average” learner.
Cognitive Diversity Includes Executive Function
Executive functions help learners hold information in mind, manage attention, shift between tasks, plan, organize, and monitor progress. Diamond (2013) identifies three core areas of executive function: working memory, inhibitory control, and cognitive flexibility.
Why this matters for AI
If AI only sees the final answer, it may miss the real reason a learner is struggling. The learner may need content support, but they may also need help with memory, attention, task-switching, planning, or self-monitoring.
Image generated with ChatGPT by OpenAI, 2026.
Executive function matters for AI design because a learner’s visible struggle may come from working memory, attention, task-switching, planning, or self-regulation rather than a lack of understanding.
Executive Function in Practice
Executive Function in Practice
The infographic above shows the main executive function areas. The video below extends this idea by showing how executive functions can affect learning behaviours such as starting tasks, organizing information, managing attention, and adjusting strategies.
As you watch, think about this question:
If an AI tool sees that a student is struggling, how would it know whether the problem is understanding, memory, attention, task-switching, or self-regulation?
Video created for this OER, 2026.
Connects to: Activities 2 and 4 because learners may need different forms of support, and an investable AI learning tool should do more than respond to performance—it should help identify what kind of support each learner actually needs.
Learning Is Not Just Task Completion
Learning Is Not Just Task Completion
Completing a task does not always mean learning has happened. Students still need opportunities to retrieve, explain, apply, and transfer ideas independently. Retrieval practice research shows that actively pulling information from memory can strengthen long-term learning (Roediger & Karpicke, 2006).
Why this matters for AI
AI can improve the final product while reducing the learner’s chance to practice remembering, explaining, or rebuilding the idea themselves.
Connects to: Activities 1 and 3 because both ask whether AI is helping learners build understanding and independence, or simply helping them complete the task.
Cognitive Load, Retrieval Practice, and Metacognition
Cognitive Load, Retrieval Practice, and Metacognition
AI support can be helpful when it reduces unnecessary cognitive load, but it can become risky when it removes the thinking learners need to do. Cognitive load theory reminds us that learners can only process a limited amount of information at one time (Sweller, 1988). Retrieval practice shows that learners strengthen memory when they actively pull information from memory, and metacognition helps learners monitor what they understand and what they still need to practise (Roediger & Karpicke, 2006; Koriat, 1997).
Why this matters for AI
Good AI support should reduce barriers without replacing the cognitive work. A tool may be helpful if it organizes information, reduces distractions, or supports reflection, but less helpful if it gives the answer before the learner has a chance to think.
Connects to: Activity 3 because it asks learners to judge whether AI is reducing unnecessary barriers while still preserving the retrieval, reflection, and independent thinking needed for meaningful learning.
Image generated with ChatGPT by OpenAI, 2026.
From Research to Market Judgement
From Research to Market Judgement
The research above gives us a way to judge AI learning tools more carefully. It is not enough for a tool to claim that it is personalized, brain-based, adaptive, or inclusive. For this OER, the stronger question is whether the tool actually supports cognition, reduces barriers, and builds independence over time.
Three cautions help shape that judgement.
Adaptation is not the same as understanding
Adaptation is not the same as understanding
AI in education can already provide feedback, adjust difficulty, suggest next steps, and create more individualized pathways. Holmes, Bialik, and Fadel (2019) describe intelligent tutoring systems as tools that can give automatic feedback and guide learners through activities based on their responses.
However, responding to performance data is not the same as understanding why a learner is struggling. A student may need content support, but they may also be affected by memory, attention, language processing, confidence, task-switching, sensory demands, or the design of the learning environment itself.
Market connection:
This matters for Activity 4 because the strongest AI learning tools will need to do more than react to errors, speed, or accuracy. They will need to support learner variability while still preserving teacher judgement and meaningful thinking.
Neuroscience-informed does not mean neuromyth-based
Neuroscience-informed does not mean neuromyth-based
Cognitive diversity is not a myth, nor are conditions or learner profiles such as Autism, ADHD, Dyslexia, Dyspraxia, Giftedness, or Anxiety. However, neuroscience-related myths in education, often called neuromyths, do exist.
One example is the meshing hypothesis, which is the idea that instruction is best delivered in a format that matches a learner’s preferred learning style, such as teaching a “visual learner” mainly through visual materials. Pashler et al. (2008) found no adequate empirical evidence that matching instruction to learning-style preferences improves learning outcomes. Massa and Mayer (2006) tested this directly with matched visual and verbal versions of the same lesson and found no advantage when instruction matched a student’s stated preference.
A related caution comes from Howard Gardner’s theory of multiple intelligences. Although the theory has been highly influential in education, the idea that teaching should be matched to separate, brain-based intelligences remains contested. Waterhouse (2023) argues that claims of independent intelligences with distinct brain bases lack strong empirical support.
Market connection:
This matters for Activity 4 because an investable AI tool should not sort learners into fixed brain types or make unsupported claims about how each “kind” of learner learns best.
Planning for variability is not adapting to it
Planning for variability is not adapting to it
Universal Design for Learning addresses learner variability by designing multiple ways for learners to engage with learning, access information, and show what they know (CAST, 2024). UDL also connects to executive function. García-Campos, Canabal, and Alba-Pastor (2020) found that many UDL checkpoints already support executive function skills, especially planning and metacognition. CAST also addressed executive function in their 2024 UDL Guidelines 3.0, naming it one of three design categories, alongside strategy development and emotional capacity (CAST, 2024).
Market connection:
This matters for Activity 4 because flexible entry points and support for executive functions such as working memory, inhibitory control, and cognitive flexibility are two different design bets, not two features already combined in one tool. Neither should be overlooked when judging how fully a tool actually addresses learner variability.
Research Takeaways
Research Takeaways
Taken together, the research points to one design challenge: AI learning tools need to support cognition without replacing it.
For an emerging market, this matters because the most promising tools will not simply offer more hints, faster answers, or more personalization. They will need to show that they can support learner variability, reduce unnecessary barriers, preserve meaningful thinking, and build independence over time.
This is the foundation for Activity 4. Participants are not just choosing which AI tool sounds most impressive. They are judging which kind of AI learning tool is actually designed for cognitive diversity, and which ones are only using the language of personalization.
Scan the QR code above to participate in our Mentimeter survey on Intelligent Tutoring Systems.
Page updated
Google Sites
Report abuse