Can neuroscience help shape the future of educational AI?
Most educational AI adapts to what students know. But what if it could also recognize how students think, remember, focus, and learn?
This OER explores how neuroscience can inform the design of AI that supports cognitive diversity while helping learners build understanding, confidence, and independence.
What You’ll Explore
Throughout this OER you will:
- explore neuroscience and executive functions
- investigate real classroom scenarios
- evaluate AI support versus AI dependence
- compare emerging educational AI models
- decide where the greatest opportunity for innovation lies
Ready to Begin?
Join the Conversation
After completing the OER, return to this page and share your thoughts in the comments.
Discussion Questions
- How should neuroscience shape the next generation of educational AI?
- Should future AI focus on giving every learner more support, or on first understanding how different learners think, remember, and learn?
- What ethical or practical challenges should designers consider when creating neuroscience-informed AI for education?
The Neuroscience Team: Mike Cafuta, Gerta Heba, Neonatal Rios, Monique Sanders
Hi there! I’m enjoying the OER so far. Just a quick note: on the Research Principles page, the “Neuroscience-informed does not mean neuromyth-based” drop-down doesn’t expand for me, so I wasn’t able to view the content. Otherwise, it’s been a great experience so far!
Thanks so much Denise for your feedback, we fixed this – and we have also done some troubleshooting on the Mentimeter’s and fixed them if you want to submit your response 🙂
Hello! The Mentimeter included in Activity 3 Genially doesn’t ask for a name, and it also says, “Please wait for the presenter to change slides’ so I am not sure if there are more questions to answer after the first one.
Hi Danielle, thank you for this. Both are fixed now, and that was the only question in that activity, so you didn’t miss anything.
After exploring the rest of our OER, I’m curious how you think neuroscience should shape the next generation of educational AI. Should it focus on giving learners a faster path to the right answer, or on understanding why a learner is struggling and delivering the right scaffolding that unblocks any barriers, whether directly or through teacher mediation?
Hi Neonatal, thank you for the information!
How should neuroscience shape the next generation of educational AI?
I think neuroscience research needs to be at the forefront of creating educational AI. Currently, AI has been developed with a focus on the technology aspects such as generative AI, image generation, and now videos. However, if universal design or personalized learning is well-researched and applied to AI’s capabilities, alongside the outline of learning theories, then the next generation of EdAI would be beneficial for both areas of learning.
Should future AI focus on giving every learner more support, or on first understanding how different learners think, remember, and learn?
While each learner has their own strengths, weaknesses, and overall skills, I believe that support shouldn’t be provided just because it’s there. Options of support, challenges, extensions, or interests, like a choice board variety of EdAI, would be the most supportive for all learners. If analytic data is collected on which supports students are accessing during their learning (be it challenges, extensions, or interests), then the supports most used are visibly seen by the student and teacher. This would inform the student that they like the support, build awareness of their AI dependence, and shift them toward other tasks that push them more into the AI support role. E.g. “I noticed you could not remember the parts of a paragraph; can I share a hamburger visual with you?” Then when a student is stumped again, the hint could be, “I like to think of a hamburger when I’m writing a paragraph. Would you like to know why?” This may jog students’ memory or provide an entry point into the visual guideline of a top bun for a topic sentence, interesting details for the patty and toppings, and a bottom bun for the concluding sentence.
What ethical or practical challenges should designers consider when creating neuroscience-informed AI for education?
Some ethical challenges that come to mind focus on privacy, safety with online use and the amount of screen time. As well as when students can access ‘their’ AI only at school and/or home, and even the idea of the personalized AI following them through school or being completely wiped every year. I would like to see the long-term research of consistent AI use by adults, children, and students. Practical challenges include technology access and costs, storing data, and the environmental concerns with AI data centers, such as the one that will be built in Alberta without passing an environmental assessment.
Hi Gerta,
Sorry, one more!
In activity 3, Genially Scenario 1, the “Next” button isn’t working. It was when I went through it yesterday though (I wanted to take another look at the scenarios!)
Also, I am finding that the Mentimeter at the bottom of the Research Principles for Meaningful AI Learning page does not allow me to participate.
You are very welcome! Great work Neuroscience Team. Your OER connected well with last week’s discussion about AI Learning Partners. One question I keep coming back to is usability and sustainability in JK-12 classrooms. Are we moving toward a model where every student has their own AI tutor on their own device or is the greater opportunity AI that supports both teachers and students? The SchoolAI example stood out to me because the AI wasn’t just interacting with students, it was helping the teacher identify patterns in understanding and make instructional decisions. That feels like a more sustainable model and raises an interesting question about next generation of educational AI. Should we focus on replacing parts of teaching, or amplifying teachers’ ability to support every learner?
Great questions. I think you are pointing to one of the biggest tensions in this market: whether AI is moving toward individual tutoring for every student or toward tools that make classroom support more sustainable for both students and teachers.
The one-to-one AI tutor model sounds powerful, but in JK-12 it also depends on devices, supervision, teacher capacity, student independence, and whether the AI support is actually improving learning rather than just helping students finish tasks. That makes me wonder whether the strongest opportunity is not AI that works separately with each learner, but AI that fits into the classroom system.
For cognitive diversity, this matters because a student’s difficulty is not always visible from their answer alone. AI may notice that a learner is struggling, but the teacher still brings context: behaviour, confidence, language, relationships, task design, and what has happened over time.
I guess this leaves me thinking that the strongest opportunity may be somewhere in the middle. AI can support individual learners, but it also needs to help teachers understand what is happening across the classroom. For JK-12, that balance might matter more than simply giving every student a separate AI tutor.
Hi dmouton, thanks for your comment and great questions! I couldn’t help but reminisce about Woolgar’s Configuring the User when you brought up usability 🙂 I think you’ve asked a really important question.
As for sustainability, a model that supports both teachers and students definitely has my vote. I also wonder where parents and guardians fit into that picture. How much involvement should they have? What role might they play? And how might that change as educational AI continues to evolve? It’s engaging to think about what the next 10 years of educational AI might look like.
P.S. Thanks again for helping us catch those early snags. We really appreciated it!
Hi Neuroscience Team,
Wonderful work on your OER! I appreciated the way your team approached neuroscience-informed AI as an opportunity forecast instead of asking whether AI belongs in education. Framing the activities around cognitive diversity, executive functioning, retrieval practice, and metacognitive encourages us to think more critically about what meaningful AI support should look like.
One thing that stood out to me was the distinction you make between AI that adapts to learner performance and AI that seeks to understand why a learner is struggling. That shift is important because designing AI that builds learner independence rather than supporting task completion, is paramount! However, I also found myself thinking about how important the role of the teacher/educator remains in all of this. They need to be able to provide the context and interpret patterns in order to make better informed instructional decisions.
I have a couple of suggestions that I think would strengthen your OER:
-While working through Activity 3, I wasn’t always sure whether I was completing the activity correctly. The Genially appeared to rely on interactive elements that either did function correctly for me, or didn’t provide clear feedback, and I was left wondering if I completed it successfully. Adding clearer instructions or confirmation of the expected interaction would improve the learner experience.
– I also found some of the embedded videos cognitively demanding. For example, in Activity 4’s Market Analysis, the OER directs learners to begin listening at the 16-minute mark of a 56-minute podcast, but it isn’t clear how much of the video they are expected to watch. Since your OER emphasizes cognitive load, I think it would be helpful to either clip the relevant segment or explicitly state something like “Watch approximately 2-3 minutes beginning at 16:00”. That small change would better align the instructional design with the learning principles discussed throughout the resource.
To answer your discussion questions:
I don’t think neuroscience alone can or should shape the next generation of educational AI. Instead, I see the strongest opportunity at the interaction of neuroscience, learning sciences, UDL, and educator expertise. Understanding how learners think, remember, and learn should inform the type of support AI provides, but that support should continue to promote productive struggle, retrieval practice, and learner agency rather than replacing those processes.
I also think designers need to carefully consider the ethical issues of their AI tools such as learner privacy, profiling transparency, and the risk of unintentionally labeling students based on inferred cognitive patterns. One question I keep coming back to is how confidently AI infers why a learner is struggling. In practice though, this is harder to prove. For example, some students guess, rush through tasks, or randomly click on responses, while others demonstrate similar performance because of genuine cognitive or language barriers. Those behaviours might produce similar data but have very different underlying causes. This reinforces to me why neuroscience-informed AI should augment (not replace) teacher judgment, and should be implemented in ways that preserve learner autonomy and independence.
Hi Quarters, thanks so much. Both the practical notes and the discussion piece are genuinely valuable. We’ve updated Activity 3’s interactive flow so it’s clearer when an interaction has registered and tightened guidance on which segments of longer embedded videos to watch (worth another look if you get a chance).
I really liked your point about the limits of what AI can actually infer. It’s not just that AI needs more data to figure out why a student is struggling, but that sometimes there may not be a clean signal to find at all, in the case of distinguishing guesswork from genuine cognitive limitations for instance. This also connects to your astute point that how support is delivered, preserving productive struggle and learner agency, matters too, not just identifying the source of struggle and the right type of support to offer.
Available research I’ve come across doesn’t settle this exactly, but a few things I found keep circling back to the same idea, teacher involvement matters, which also speaks to the key tension raised earlier: should the market be moving toward individual AI tutoring for every learner, or toward AI that supports the classroom system as a whole?
Sharing a few findings for the discussion: one study of high school students found unrestricted AI-alone access could harm learning, as students who relied on it performed worse once it was taken away than students who never had it, while a version with teacher-built safeguards prevented that harm entirely (Bastani et al., 2025). Similar story in an earlier three-condition study in postsecondary math where AI-alone significantly underperformed both teacher-mediated and teacher-only instruction (Dasari et al., 2024). Also, a recent K-12 working paper looked at real-world use of teacher-facing AI, and found it can backfire specifically when used for task delegation, especially among teachers less equipped to catch and correct what it produced (Sungu et al., 2026). Still a thin evidence base, so call it early signal, but it’s got me thinking teacher involvement matters in more than one place (how a tool gets built, and how it actually gets used) and the risk isn’t just in AI being wrong, it’s also in AI sounding compelling enough that nobody stops to check.
This brings me to the question of whether, in ambiguous-signal situations when learner struggle is detected, AI learning tools should be designed to flag when they’re uncertain and call for teacher mediation. Would that actually help, or just add friction most teachers don’t have time for?
Sources:
Bastani et al., 2025 – https://www.pnas.org/doi/10.1073/pnas.2422633122
Dasari et al., 2024 – https://doi.org/10.3389/feduc.2023.1295413
Sungu et al., 2026 – https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7007339
Hi Neuroscience team,
Your OER brought up some really interesting points and made think more deeply on how we, as edtech innovators, can facilitate learning for neurodiversity. In reading your content, it actually gave me an idea for my assignment 3, so thanks!
“Should future AI focus on giving every learner more support, or on first understanding how different learners think, remember, and learn?”
I think the necessary first step is understanding how different learners think, remember, and learn. More support is only beneficial once the teacher or school board has an understanding of how that support is best offered, and doing so effectively requires an understanding of each student’s learning profile. Otherwise, AI risks providing generic assistance that is no more effective than what a currently overworked teacher can offer, but without the human-to-human connection.
I also feel that this understanding doesn’t necessarily need to overstep privacy boundaries, which is a legitimate concern if potentially vulnerable students’ sensitive information is uploaded. The AI could track, for example, how quickly someone reads and, if it falls below a certain threshold, subtly adjust the font, size, or colour contrast to see whether those changes improve comprehension. The AI would be trained on that student’s reading behaviours rather than on a label such as, “This student has dyslexia; interact accordingly.” The ideal situation is one in which AI provides every student with support proportionate to its understanding of the individual and continuously refines that understanding over time.
This also shifts the conversation from diagnosis to responsiveness. Rather than attempting to categorize learners, this personalization becomes less about identifying deficits and more about removing specific, possibly overlapping barriers to learning. If implemented thoughtfully, this approach could create more equitable learning environments by responding to what students actually do when studying, listening, or reading, rather than what they have been formally diagnosed with or how they compare to their peers.
The question above is posed as an either/or, but I think it should be approached (aptly) as a spectrum. At one end is ‘learn about the learner,’ and the goal of this technology should be to ensure the process begins at this end. At the other end is ‘ provide personalized support,’ and the AI should gradually move up the spectrum as it interacts with, and better understands, the learner.
Hi jlehu, thanks for your comments, and glad this also sparked an idea for your own Assignment 3.
Your point about shaping how the AI learning tool understands a student through their behavior, rather than a diagnosed label, really resonates. It aligns with our framing, and it’s a direct answer to a labeling risk others have raised in this thread.
What stands out to me most in your proposed design is the focus on measuring observed behavior and iteratively improving responsiveness from there: instead of sorting a learner into a category, the model tracks something concrete (like reading speed), tests a barrier-removal response (adjusting font or contrast), and refines based on whether comprehension actually improves. That refinement loop is itself a soft form of proof, where each cycle confirms whether an adjustment is working, though I’d still distinguish that from evidence the understanding transfers without the AI learning tool later, which is the higher bar our analysis also explored.
This all brings me to the design question of what should happen with cognitive limitations and other learning barriers that don’t map cleanly to observable behavior (attention maybe more than most, since gaze can be tracked directly, though even that misses when someone’s mentally checked out while still looking at the screen). Would closing that gap require something more invasive, like monitoring brain activity directly, and if so, how much privacy would be a reasonable tradeoff to get there?
Hi Neonatal,
Your question is something I was thinking a lot about while going through your OER.
I have no expertise in neurodivergence, but I feel that a properly designed and trained AI could discover these limitations through a variety of means. Tracking gaze is one way, as you mentioned, but even then, AI could track frequency of eye movements, distance, and possibly even pupil dilation, all calibrated to the individual’s normal eye movements. The same could be done for mouse movements, typing habits…
I think my point is that more invasive techniques may eventually be part of this equation, but there is a ton of data that can be collected by less invasive means first.
Some posibilities:
– Non-invasive alerts and reminders for ADHD learners to assist with staying on task
– Subtly lessening rigidity of routine for learners with Autism
– Scaling pace of content for gifted learners to deal with boredom
And as I mentioned earlier, I think all of these could be implemented without diagnosis, but with a carefully designed AI that recognizes when learners are more and less effective, and adjust accordingly.
What are your thoughts on this?
Hi jlehu, I agree, non-invasive tracking applied to AI learning tool design like this seems like an underexplored area, and a compelling one to dig into from an investment angle. If pilot data showed measurable achievement gains from behavior-based adaptation, without touching diagnostic labels or requiring a pre-diagnosis baseline, that could be a promising pitch in the making.
From a de-risking standpoint, it might also be worth weighing where this kind of behavioral data lands relative to current privacy regulatory frameworks, whether there are instances where it has counted as biometric or re-identifiable data, or stays clear of such designations. Beyond good ethics, as others in the thread have mentioned, privacy-respecting designs offer a valuable risk moat from an investment standpoint, and can be the deciding factor when investors weigh similar opportunities.
Hello team!
I think the questions you pose are important because I believe education is at a crossroads. Many educators are trying to decide how AI fits into teaching and learning, but too often those decisions are driven by excitement or skepticism rather than evidence. Before teachers can confidently embrace AI, we need accessible, defensible research that helps us understand when AI genuinely supports learning and when it may unintentionally replace important thinking.
1. As a teacher, I’ve learned that two students can struggle with the same task for completely different reasons, and the learner profiles reinforced that idea. I think neuroscience should help AI understand why a learner is struggling so it can provide the right support without taking over the thinking.
2. I think understanding the learner has to come first. The sections on cognitive diversity and executive function reminded me that effective teaching isn’t about giving students more support—it’s about giving them the right support at the right time.
3. Maya’s story really stood out to me because it showed how easily helpful support can become dependence if AI does too much. Designers need to build tools that are grounded in research, avoid neuromyths, and gradually develop learner independence.
Overall, this OER reinforced my belief that the greatest opportunity isn’t creating AI that thinks for students, but creating AI that helps teachers better understand learners while empowering students to become more capable and independent thinkers.
Many thanks for such a great learning experience!
Thank you, Cherri-Lynn! We’re really glad you found the OER to be a great learning experience. It was definitely fun trying to wrap our “brains” around the topic!
I think you captured where education is right now with your point about excitement and skepticism. It’s easy to get caught up in the newest AI tools while also wondering where they truly fit in our classrooms.
I also completely agree with your comment about designers and the tools they’re building. They have a pretty big responsibility, and so do we as educators, to understand these new technologies before deciding how they fit into teaching and learning.
Things are moving so quickly that it will be really interesting to see what educational AI looks like even a few years from now. Thanks again for your reflection!
Hey team, a great OER, thanks for a great week!
How should neuroscience shape the next generation of educational AI?
Neuroscience should help guide the design of educational AI by helping us better understand how students learn, process information, and overcome barriers. AI should not just become better at giving answers; it should become better at supporting the learning process.
Should future AI focus on giving every learner more support, or on first understanding how different learners think, remember, and learn?
I think AI needs to start by understanding how different learners think and learn. Cognitive diversity means students do not all struggle or succeed in the same ways, so effective AI should adapt to the learner rather than expecting the learner to adapt to the technology. The goal should be providing support while keeping the learner actively thinking and building independence.
What ethical or practical challenges should designers consider when creating neuroscience-informed AI for education?
Designers need to consider privacy, accuracy, and the risk of labeling students based on AI predictions. There is also the challenge of making sure AI supports teachers and learners rather than replacing human relationships or creating dependence. The biggest question is whether the AI is reducing barriers while still allowing students to think, practise, and grow.
* I may have submitted just my name with a few of the Mentimeters and then may have submitted the responses without my name, please reach out if you do not have all of my responses! Thanks again!
Hi jessy701, thanks for your reflection! I also found myself wondering what will happen with all of the information that students may eventually share with AI tutors. Will privacy really remain private? And what happens if AI begins labeling students based on patterns it sees? Maybe labels become a thing of the past, or maybe they become even more common behind the scenes. I also wonder about accuracy. If a student answers questions differently depending on how they’re feeling that day, will AI be able to recognize that, or will it make assumptions based on incomplete information? It’ll be interesting to see how designers balance personalization with privacy and ethics.
Thanks for the note about the Mentimeter responses. We’ll reach out if we notice anything missing, but hopefully everything came through. Thanks again!
Hi everyone,
Great OER this week.
To discuss the question “Should future AI focus on giving every learner more support, or on first understanding how different learners think, remember, and learn?” According to the OER, all neurodiversity exists on a spectrum. The OER also states that the spectrum on which all neurodiversity occurs is always shifting depending on context.
Because of this huge variation amongst every individual learner and amongst every learning context, I think AI tutors are best suited to focusing on giving every learner more support as opposed to trying to understand how each individual thinks, remembers, and learns.
I believe the universal support that the AI tutors should focus on are the executive functioning skills. As was stated in the OER, executive functioning skills help every learner to focus, plan, and get things done, and that these are skills that every learner needs to work on developing.
I also think that AI tutors might be better suited for teachers to use instead of students. If teachers used these tools to plan their lessons and units, the AI tutors might be able to help the teachers find ways to incorporate executive functioning skills into their lessons. This will limit the amount of time the students are interacting with technology, and increase the time students are interacting with each which is where they can learn to understand emotions, manage impulses, adapt to change, consider multiple solutions, and work on organization, all key executive functioning skills according to the OER.
I think that educational psychologists are best suited to diagnose individual learners with diversities like autism, dislexia, ADHD, and others. But there is definitely an opportunity for the psychologist and teachers to use AI to come up with lessons and learning strategies that are best suited to assist with those individual learning differences.
Hi Bryce, thanks for your reflection…especially how you connected executive functioning back to the OER. I think it’s a really important piece of this whole neuroscience discussion, and one that will only become more relevant as AI and education continue to evolve.
I’m still only scratching the surface of how to use AI more efficiently and effectively, but I already see the potential it has to help teachers improve their practice. Our students are going to be using AI someday, so I think it’s important that we stay ahead of the game and be as knowledgeable as possible so we can help guide them in using it responsibly.
García-Campos et al. (2020) from our reference list dives deeper into executive functioning and inclusive education, and I’d encourage anyone interested in the topic to have a looksy.
Thanks again!
Hi everyone,
Great job this week!
How should neuroscience shape the next generation of educational AI?
I think neuroscience can play an important role in shaping the next generation of educational AI because understanding how the brain learns can help provide more personalized support. However, I also think other factors, such as pedagogy and the role of teachers, should be considered when designing educational AI.
Should future AI focus on giving every learner more support, or on first understanding how different learners think, remember, and learn?
I think AI should first understand how different learners think, remember, and learn because it can then better differentiate instruction and match support to students’ individual needs.
What ethical or practical challenges should designers consider when creating neuroscience-informed AI for education?
There are several ethical and practical challenges to consider. Privacy and data security are important because AI could collect sensitive information about how students think and learn. This data needs to be protected and used responsibly. Bias and fairness are also important, as AI could make inaccurate assumptions about students or reinforce existing biases. Finally, parents should be informed and give consent when AI is being used to collect or analyze student data.
Hi Jodee. Thanks for your comment! You have provided some awesome insights. I like your points especially in the last paragraph about responsible use of AI. We as educators need to be teaching students about how to use AI responsibly if we will be using it in schools as a supplement to our teaching and their learning. School districts approve certain AI tools and that is usually what is suggested that students use. I know that you can change settings in an AI tool so it does not connect the students identity to the questions they’re asking, but I’m sure it is retrievable still, and that’s what makes using AI a sensitive issue especially for kids. Parents definitely need to be in the loop as students are impressionable and they need to be taught the skills to critically analyze what they are reading. AI is a great supplement to teaching and learning, but can be harmful if not used properly.
Thank you for sharing your OER!
Here’s a few ideas based on the discussion questions below:
How should neuroscience shape the next generation of educational AI?
-My answer to this is very carefully. Right now we are becoming increasingly surrounded by data-tracking tools and while this can be informative and helpful, there are also potential challenges to this as well. Always on devices and SLAM (Simultaneous Localization and Mapping) sensors can cause ethical concerns especially when entrusting sensitive data to larger companies. I was intrigued by the UDL AI approach while working through this OER and I’m going to be thinking about ways that I can incorporate UDL principles more easily through AI tools.
Should future AI focus on giving every learner more support, or on first understanding how different learners think, remember, and learn?
– I’m mixed on this. I think that if people want an answer to a question, they shouldn’t have to beat around the bush to get an informative answer; however, I also think that the importance of using in AI in other ways in support of cognitive neuroscience is also important. They sound like different tools to me.
What ethical or practical challenges should designers consider when creating neuroscience-informed AI for education?
-I think shifting district policy, permissions, data storage and usage, and privacy are all important factors.
Thanks again for this OER!
Hi Kane. Thanks for your comment. Yes, a mixed approach as you suggested is important. An ITS system that focuses on providing strict answers can serve it’s purpose when needed, and when scaffolding and actual cognitive support is needed, an ITS system can serve it’s purpose in this way too. I am just as concerned as you are with how much we are being tracked nowadays. Even with wearables, and other tracking devices people have presented on in the AI x Neuroscience field.. these all raise concerns. I think ITS systems will only get smarter with helping students with their learning, but hopefully it is not too much at the expense of the students privacy.
Hi neuroscience team,
I really enjoyed working through your OER this week! The role of neuroscience in the development of future AI is fascinating, and I believe that it will become increasingly important as new AI tools enter the market. I think the biggest role it has is in recognizing the different ways that learners process information.
I was really intrigued by Tobey’s Tutor founder, Arlyn Gajilan, describing how the AI platform builds customized learner profiles that provide scaffolding and support tailored to each student’s specific strengths and areas of growth. While I think there is a real challenge in making educational AI that meets students’ individual needs, the opportunity for learners and educators outweigh these challenges. We have seen students rapidly access AI tools to simply get the right answer, get a good grade, and complete assignments, all to the detriment of the actual learning process. AI tools such as NeuroScaffold AI potentially provide the missing piece in educational AI – personalization that builds independence for learners.
As education systems become better at understanding students’ needs, AI tools should become better at supporting those needs. While this type of AI development may not be as quick and profitable in the beginning, I do believe it holds massive potential in helping to develop learning skills in all students versus replacing the actual learning.
Thank you so much for taking the time to work through our OER. I really liked your point about AI supporting the different ways students process information. That was one of the biggest ideas we kept coming back to while creating the project.
Your point about students using AI to finish work without actually learning really stood out to me too. I think that is one of the biggest challenges with AI in education right now. The goal should not be to remove all difficulty, but to give students the right support at the right time and then gradually step back.
I also agree that this kind of tool may take longer to develop and may not be the quickest route to profit, but it could have a much more meaningful impact on learning in the long term. I wonder how we can make sure personalized AI helps students become more independent rather than creating another kind of dependence.
This is a great OER! Here are some thoughts regarding the discussion questions.
Q: How should neuroscience shape the next gen of educational AI?
I think the big shift is moving from “did they get it right” to “why are they stuck.” Most AI tools right now just react to performance with wrong answer feedback, maybe giving a hint. But if you bring in stuff like executive function research (working memory, attention, self-regulation), AI could actually start noticing patterns in how someone struggles, not just that they’re struggling. A student who loses track of a multi-step task needs something totally different from a student who just can’t hold info in working memory.
More support vs. understanding first?
Understanding first. More support sounds good on paper, but if the AI doesn’t know why someone’s struggling, it’s just as likely to create dependence as actually help. Like, a student could “get through” an assignment with AI doing a lot of the cognitive lifting and never actually build the skill. A tool can look helpful on the surface, even though it’s doing the thinking for the student instead of with them. Once AI can actually tell how someone’s thinking, then support makes sense, and ideally, it fades out over time instead of becoming a permanent crutch.
Ethical/practical challenges?
Some that jump out to me:
Neuromyths are a real risk. “Learning styles” sounds real and is very widespread as a concept, but research doesn’t really back it up, and I could see AI companies leaning on shaky pop-science because it markets well.
Hard to design something that knows when to help vs. when to back off
Tracking attention/memory/behaviour = sensitive data, so privacy and consent (especially with kids) is a big deal.
Designing for “cognitive diversity” means testing across a bunch of different learner types, not just tweaking a tool built for the average student, which is easier said than done.
Overall, insightful OER that I found particularly engaging as someone who really benefitted from internet tools and online schooling because the standard education system didn’t really accomodate the way that my brain works, so I’ve alwys been interested in adaptive learning and tutoring systems that work with the student rather than the student having to overcome the challenge of aligning themselves with a system that doesn’t suit them before they can even start considering the tasks and learning themselves.
Thank you for such a thoughtful response. I really liked the way you described the shift from asking whether a student got the answer right to asking why they are stuck. That feels like such an important difference, and it really gets at what we were trying to explore in the OER.
I also agree that understanding has to come before simply giving students more support. An AI tool can make it look like a student is succeeding because the work gets finished, but that does not always mean the student is actually learning. Your point about support fading over time really stood out to me. Ideally, the AI should help students build confidence and skills, then gradually step back.
You also raised some important concerns around neuromyths, privacy, and how difficult it is to design for real cognitive diversity. I can easily imagine companies using words like personalized or brain-based because they sound appealing, even when the research behind them is weak.
I really appreciated you sharing your own experience too. It was a good reminder that adaptive learning is not just about convenience. For some learners, it can make the difference between being able to engage with the learning and feeling like they have to fight the system before they can even begin.
Hello team! Good work on creating a well-organized and fun OER! I love how you specifically aligned the information to the activities. Here are my thoughts to the three questions you posed:
1. How should neuroscience shape the next generation of educational AI?
From a past project on neuroscience I completed in my last MET course, I have come to understand that for neuroscience to work in education (and to avoid neuro myths), we need experts’ involvement in the design process to get the best use out of tools for the classroom. In this case, it might be a combination of child psychiatrists, special education teachers, and neuroscientists. I think there is definitely a place for AI tools to help neurodivergent learners; my concern would be whether the tool is helping, holding a student back, or even hurting a student in their learning if it can’t fully measure / assess the students’ needs and provide the right support.
Should future AI focus on giving every learner more support, or on first understanding how different learners think, remember, and learn?
I think it would be great having both options. Not everyone would be comfortable with an AI tool assessing how they think, remember and learn. It would be great to have a tool that you could turn to for customized support. However, many people like that separation of neuroscience and AI when they are using an AI tool.
What ethical or practical challenges should designers consider when creating neuroscience-informed AI for education?
I think the biggest concern for me would be overreliance on an AI tool to provide expert support to students with neurodivergent needs. An expert, like a special education teacher or psychologist, would probably need to monitor to make sure students are getting the right kind of support. An AI tool can’t read body language and connect on a human level. I think AI can be very helpful but shouldn’t be a replacement for human interaction and support. Privacy might also be an issue if students are being asked to describe themselves and disclose sensitive information.
Great work, Team!
Edith
Thank you for this thoughtful response. Your point about involving different experts in the design process made me think more about how complex this kind of tool would be to create well. It would need more than strong technology. It would also need a real understanding of students, classrooms, learning differences, and the limits of what AI can actually interpret.
I also found your point about choice interesting. I had not thought as much about the fact that some learners may want personalized support without being comfortable with an AI tool assessing how they think or learn. That tension between personalization and privacy seems like it would be difficult to navigate.
Your concern about overreliance also really connects with the questions we were trying to explore in the OER. AI might be able to identify patterns or offer support, but it can still miss so much of what a teacher, psychologist, or other specialist notices through conversation and human connection.
I think your response highlights how important it is to see AI as one part of a support system rather than the whole solution.
Strong work. This is an important and underappreciated side of education. In the college system it’s clear to anyone who cares to look that support and accommodation of neurodivergent students is essential to maintaining a vibrant, diverse workforce.
How should neuroscience shape the next generation of educational AI?
I think the most important thing is the neuroscience vs. neuromyths distinction you point out in your research section. This is particularly so because these will be commercial products, and even if neuroscientists design them, the people responsible for selling them won’t be. Marketing requires good stories and good stories are simple, so nuance and complexity generally don’t make it into marketing materials. As Mark Twain said, “A falsehood is halfway around the world before the truth has its pants on.” And yes, I know Twain didn’t actually say that, but it feels like he could have said it so lots of people think he did. It’s the same reason many people believe the learning styles myth – because it sounds right. And even when the underlying neuroscience is correct, it doesn’t mean the tool is effective. A product can be designed for people with ADHD but that doesn’t mean the AI that powers it can correctly identify and respond to specific executive function issues. Naming a population isn’t the same as being grounded in how that population’s cognition works. As you said in Key Takeaways, “The gap between what’s proven and what’s promised is still open.” Unproven neuroscience claims are themselves a design risk, especially if the general public already thinks those claims are true.
Should future AI focus on giving every learner more support, or on first understanding how different learners think, remember, and learn?
Appropriate support requires knowing the individual, not just their performance. Responding to performance data is not the same as understanding why a learner is struggling. In Activity 1, the learner’s grades improve with AI support but she can’t organize an essay independently without it. That’s what happens when support doesn’t calibrate or fade appropriately over time. This raises the thorny issue of what we mean by “support” particularly because what the learner wants in the
What ethical or practical challenges should designers consider when creating neuroscience-informed AI for education?
The LLMs that power AI edtech tools are themselves commercial products that are part of a larger surveillance capitalism infrastructure. This will create both ethical and practical challenges where the needs of the learner do not align with the goals of the LLM vendor. Ethical design has to account for who profits and how, since that shapes whether a tool ever fades support or keeps the learner engaged and dependent instead. Dependence can be more profitable than independence, especially for a tech industry that has shifted almost entirely to a subscription-based revenue model.
From an investment perspective there is also another, less obvious potential challenge. Governments are becoming increasingly concerned about the data security and privacy implications of big tech. Ontario’s new EDSTA legislation is a good example of the trend toward requiring more oversight, safeguards, and reporting. The Investment Analysis section asks “Would you fund the design most worth building, or the design most proven to work?” In isolation my answer is “most proven to work” but in the current climate, an essential question to consider is how likely the core functionality of a tool is to run afoul of future regulation. An investor who fails to consider more restrictive legislation in future may find they have funded a great tool that many institutions are no longer allowed to use.
Hi John, thanks so much for your thoughtful response.
Your point on the need for a clear definition of “support” is one I keep coming back to. Support from an AI learning tool that removes barriers and builds agency is a different product than support that just improves scores, and an investor skipping pilot data on whether a specific fading mechanism actually works, beyond what’s already known about human-led scaffolding, risks funding something that looks strong on paper and underperforms in practice. Your emphasis on the need for evidence backing design claims is thus the standard for prudent investment that bets on educational value rather than short-term hype.
What also really resonated was how you connected subscription economics to ethical conflicts and regulatory risk. Subscription models create an internal incentive to keep learners dependent, and that tension sits next to a regulatory climate increasingly focused on oversight of how these tools operate. This makes fading a compliance question as well as a design one. Investor-backed EdTech vendors treating regulatory durability as core architecture now, rather than addressing it later, are likely best positioned for long-term sustainability.
Which brings me to wonder, when weighing regulatory risk in this niche, is it likely centered more on data security and privacy, or on whether outcomes transfer beyond AI learning tools?
So far, at least in Ontario, regulation is 100% focused on privacy and security (the new law is called the Enhancing Digital Security and Trust Act), but I think they’re really the same thing in the context of personalized outcomes like fading because you need persistent memory of the student to calibrate their progression over time, which means you need to store data about the student. The richer the data set the tool has access to, the greater the data and privacy risk it carries.
How should neuroscience shape the next generation of educational AI?
Neuroscience should be used to create meaningful learning experiences for individual students. One of the biggest weaknesses we have as teachers is there is only one of us. In this way we can augment the individual learning of students when we are busy helping other students. It won’t replace a teacher’s help, but can augment their natural learning process when the teacher is busy. For example, I think that the most important value for this tool in my context is as an individualized tutor/assistant after a topic has been learned and students are practicing. I do not always have time to get to every student and some students need more constant support when working on material. This way they can get that individualized support for little questions, and can call me over if they are stuck on the bigger ones. It can also help for when I really need to work with one or two kids, as it can answer the questions of the other students temporarily while I am working with those specific individuals.
Should future AI focus on giving every learner more support, or on first understanding how different learners think, remember, and learn?
If you have been in education long enough, or learning any new skill, you learn the phrase go slow to go fast. In education’s case, generally this means that you take the time to do something properly the first time, and even though it initially takes more time in the beginning, it saves you time later on, as you won’t have to repeat it. In this same way, if an AI understands how different learners think, remember, and learn, it can teach better and understand and help students who are having difficulties better, even if it takes more time initially.
What ethical or practical challenges should designers consider when creating neuroscience-informed AI for education?
There might be a push to replace teachers with AI or that the influence of AI means that bigger classes can be created, but AI is still just a tool that a teacher uses, and is not in itself something that allows the teacher to teach more individuals. Additionally, many tools claim to work to help individuals through science or neuroscience, but few actually use current research (such as the learning styles myth you mentioned), so we need to be careful to only use products that use actual neuroscience research to back up their use, rather than just common perception or outdated knowledge.
Hi lschoepp,
Thanks for your post!
I like your idea of AI helping answer the smaller questions while the teacher is working with students who need more support. That seems like one of the most practical ways AI could actually make a difference in a classroom…at least in the beginning. Go slow to go fast 🙂
I also agree that we need to be careful about products throwing around the word “neuroscience.” It’s easy for companies to make big claims, but those claims should be backed by solid research, not just good marketing.
Thanks again for your comment!
I related strongly to the research on neurodivergent learning and AI support because I have recognized many ADHD-related traits in myself. When I was younger, no one identified these challenges as possible differences in attention, working memory, or executive functioning. Instead, adults often treated them as personality problems. This is why the topic felt very real and personally meaningful to me.
I think neuroscience should shape the next generation of educational AI by helping designers understand how learners think, remember, focus, and process information. The goal should not be to label or diagnose learners, but to recognize that difficulties may come from different cognitive barriers, such as weak working memory, limited attention, difficulty organizing ideas, or problems retrieving information. This kind of support could help learners understand their thinking patterns and develop strategies they can eventually use independently.
At the same time, designers need to avoid making learners too dependent on AI. For someone with weak working memory or difficulty organizing ideas, it may be tempting to rely on AI to summarize, plan, or generate every response. Effective AI should provide temporary scaffolding, prompts, and structure while still requiring learners to make decisions, practise retrieval, and express ideas in their own words. The goal should be to help learners become more confident, self-aware, and independent.
There are also important ethical and practical challenges. AI may misinterpret distraction, slower responses, or difficulty completing tasks as evidence of a fixed condition. Designers should avoid allowing AI to diagnose learners and must consider privacy, bias, data collection, accessibility across devices, and transparency about how learner information is used. Support should also be flexible and gradually reduced as learners become more independent.
In terms of the learning experience, I found the activity smooth, engaging, and easy to follow. The structure helped me move through the ideas step by step without feeling overwhelmed. However, I struggled with the word limit and with answering the questions on my phone. This reinforced the topic for me because the response format itself can create an additional cognitive barrier.
Overall, this activity helped me see that neurodivergent learners may not lack ability or motivation. Sometimes the learning environment, task design, or response format creates barriers that others do not notice. Neuroscience-informed AI has the potential to make learning more inclusive, but it should be designed with empathy, flexibility, privacy, and cognitive independence in mind.
Hi unidon,
Thanks for sharing your story. It’s interesting to hear how your perspective has changed looking back. I also really liked your point about AI being a scaffold instead of a shortcut. If it’s helping students become more independent over time instead of relying on it for every task, I think we’re headed in the right direction.
Thank you for the word limit feedback as well. It will help us improve the OER and think about how small design choices can create unnecessary barriers for some learners.
Many thanks again for your comments!
Hi Unidon,
I’m so glad this topic was meaningful and personal to you. It was to us too, as we are seeing these traits exhibited in the students we are teaching today, and we are more cognizant of it than educators in the past.
I have seen with myself that I can rely on AI to plan things for me sometimes when I do not know how to approach it, but this weakens my critical thinking and creative autonomy and so as you said effective AI should not make the users dependent on the tool.
What ethical or practical challenges should designers consider when creating neuroscience-informed AI for education?
I think that designers face a real tension between ethics and market realities when building neuroscience-informed AI, particularly around the continuum of AI support versus AI dependence. I see it in the following areas:
AI Support Level: As we learned in the OER, designers should build frameworks that gradually fade their cognitive scaffolding out over time to avoid AI dependence. If the software does the heavy lifting, long-term retrieval and cognitive independence is weakened.
The Commercialization Conflict: I can see a possible ethical friction between educational dependence and venture economics. While educators want to wean students off a tool to foster self-reliance, investors may see high user retention as a market opportunity. While there will always be new students each year who will need support, some business investors may see a benefit to more people being dependent on the tool as it creates a larger user market.
A Possible Solution: I think there is an opportunity for designers to distinguish between learning gaps that can be bridged with the tool, and permanent disabilities where the tool could be used long term. When thinking about the tool as serving two potential markets, the AI could be intentionally designed to assist students with an optional fading out feature, but for those who need it longer, it could also transition as a permanent accessibility accommodation that extends past K-12 and into adult employment settings.
Thanks so much for an amazing OER!
You make a good point with your commercialization conflict. A lot of these AI business ventures are a business at the end of the day, and so they will find a market area that needs to be filled and cater to it. Hopefully this means that these educational ventures will become more customized to each varying type of learner using it, and so high user retention is not an issue for educators as well as these tools are genuinely helping learners thrive.
I think neuroscience should play a bigger role in educational AI because it can help create learning experiences that better align with how our brains actually learn. Instead of simply giving every learner more support, I think AI should first understand how different people think, remember, and process information. Once it understands the learner, it can provide support that is much more personalized and meaningful.
Also, I really enjoyed your site! I liked how you started with a question and a scenario before transitioning into the topic—it was a great way to get me thinking right away. I was wondering if there was only one way to access the Mentimeter activity, or if it could have been offered in another format as well. It might be helpful to include a direct link or an alternative way to participate for anyone who has trouble accessing the activity.
I think neuroscience should shape the next generation of educational AI by reducing cognitive barriers while building learner independence, rather than simply improving task completion. Instead of giving every learner the same support, future AI should first understand how different learners think, remember, and learn, then provide personalised scaffolding while still offering multiple pathways for everyone to access learning. At the same time, designers must ensure that AI does not diagnose learners or replace teacher judgment. They should also consider privacy, bias, and the risk of creating dependence, so that AI supports thinking rather than doing the thinking for students.
As AI continues to build upon its strengths, I can see how future iterations will be able to have a more critical understanding of how different learners think, remember, and learn. This is a facet of AI that is rapidly evolving as users add more and more feedback. Once a more thorough understanding of how learners think is developed, AI can then make the appropriate adaptations to incorporate unique learning styles and needs. AI already contributes a significant amount of learning support, but it is really missing the critical component of complete understanding. The real question is how close will AI need to be in order to utilize its vast repository of information and learning tactics to tap into an individuals way of thinking.