Artificial Intelligence (AI) is rapidly changing the way we learn, teach, and interact with information. Large Language Models (LLMs) like ChatGPT are now being used in education, professional development, and everyday decision-making.
But how does using AI impact the human brain? More specifically, what does it do to our cognitive load, and does that tell us anything about the quality of our learning?
A recent paper titled “This is Your Brain on ChatGPT: Understanding Cognitive Load when Using Large Language Models” (Kos’myna et al., 2025) from the MIT Media Lab seeks to answer just that. Using functional near-infrared spectroscopy (fNIRS), the researchers measured participants’ brain activity while completing tasks with and without ChatGPT. The results are compelling—and they raise important questions for educators, instructional designers, and society at large.
This blog post breaks down the key findings of the paper, explores the role of instructional design in managing cognitive load, unpacks the different types of cognitive load, and reflects on whether reduced cognitive load necessarily leads to better learning. It also invites us to consider the ethical and practical implications of using AI as a cognitive partner.
What Did the Study Examine?
The researchers designed a series of text-based tasks, including programming challenges and comprehension exercises. Participants were asked to complete these tasks either independently or with the assistance of ChatGPT. While doing so, their brain activity was monitored using fNIRS, a non-invasive method that detects changes in blood flow in the prefrontal cortex—a region associated with cognitive effort.
The study found that using ChatGPT significantly reduced cognitive load during complex tasks. Participants exerted less mental effort when aided by the AI. However, this benefit was less clear for simpler tasks, where ChatGPT’s assistance sometimes led to decreased engagement or critical thinking.
The authors suggest that while LLMs can reduce strain on working memory and improve task efficiency, they may also discourage deeper processing when the task does not demand it.
The Three Types of Cognitive Load
To understand these findings, it’s helpful to revisit the framework provided by Cognitive Load Theory (CLT), which breaks cognitive load into three types:
Hover over each flipcard to learn more.
Hover over each flip card to learn more.
Instrinsic Load
/
Intrinsic Load
This is the mental effort required by the inherent complexity of the material. For example, learning calculus or grammar rules naturally requires effort.
Extraneous Load
/
Extraneous Load
This comes from poor instructional design, such as confusing instructions, distracting visuals, or irrelevant content. This type of load is unnecessary and often inhibits learning.
Germane Load
/
Germane Load
This is the productive effort learners expend to form and refine mental models (schemas). It supports deep learning and should be encouraged.
Effective instructional design seeks to reduce extraneous load, manage intrinsic load, and promote germane load.
The Role of Instructional Design
Instructional designers (also known as learning experience designers and learning architects) play a crucial role in managing cognitive load. By thoughtfully structuring content and interactions, designers can create learning experiences that are not only efficient but also effective.
Here are some ways instructional design can support healthy cognitive load:
- Scaffolding complex material to manage intrinsic load.
- Removing distractions and simplifying navigation to reduce extraneous load.
- Incorporating retrieval practice, reflection prompts, and real-world application tasks to increase germane load.
- Using multimedia in alignment with the Cognitive Theory of Multimedia Learning to distribute processing across visual and auditory channels.
When used intentionally, AI tools like ChatGPT can become part of the design. For example, an AI chatbot can scaffold complex topics or provide feedback, helping learners stay in the zone of proximal development. But careless integration can increase extraneous load or eliminate valuable germane load opportunities.
Does Lower Cognitive Load = Better Learning?
Not necessarily.
While a reduction in extraneous load is a good thing, too little overall cognitive effort may signal disengagement or surface-level processing. The study (Kosmyna et al., 2025) found that participants using ChatGPT in simple tasks tended to rely on the model without deeply engaging with the material.
This aligns with what we know from Cognitive Load Theory (CLT): productive struggle is necessary for schema construction. If AI does all the thinking, learners may skip the very mental processes that lead to long-term retention and understanding. We need to differentiate between using AI as a tool for thinking and using it as a replacement for thinking.
This insight is especially important in educational contexts where critical thinking, synthesis, and metacognition are goals. Instructional designers and educators must be mindful of when AI is a helpful scaffold—and when it becomes a shortcut.
The Tools and the Brain: Measurement vs. Meaning
The study used fNIRS to detect activity in the prefrontal cortex, correlating changes in blood oxygenation with cognitive load. This is a reliable way to measure mental effort, but it doesn’t tell us everything.
Here’s what these measurements can tell us:
- How much mental effort someone is using
- How task difficulty or support systems affect effort levels
Here’s what they can’t fully capture:
- Whether the person is learning deeply or just working hard
- Whether the effort is building useful mental models
- Whether the learner is engaged, curious, or motivated
That’s why we should view cognitive load measurement as a partial lens, not a complete picture. We need qualitative insights, performance outcomes, and learner feedback to triangulate meaningful learning.
Ethical and Societal Reflections
We’re still figuring out the role that AI will play in our cognitive, educational, and professional lives. There are three key tensions we must navigate:
- Support vs. Dependence – If AI reduces cognitive load, are we gaining efficiency or losing capacity?
- Skill Development vs. Skill Shifting – Will AI shift the skills learners need (e.g., from problem-solving to prompt engineering)? Are we inadvertently outsourcing foundational skills?
- Access vs. Equity – Who benefits from AI-powered tools? Will access to ChatGPT or similar models be equitably distributed across educational contexts?
As instructional designers and learning professionals, we must be proactive in addressing these tensions. This means designing with intention, evaluating outcomes critically, and listening to learners’ experiences.
Final Thoughts: Designing the Future of Thinking
The findings from “This is Your Brain on ChatGPT” highlight a crucial point: AI doesn’t just change what we do—it changes how we do things. It can lighten the cognitive load, allowing us to focus more deeply on what matters. But it can also lull us into passivity if not used thoughtfully.
Instructional design has a key role to play in this new era. By crafting experiences that balance support with challenge, and by integrating AI tools wisely, we can create learning that is not just efficient—but meaningful.
As a society, we’re still exploring what it means to co-think with machines. We don’t yet have all the answers, but asking the right questions is a good start.
UpSkill Designs is a learning design agency specializing in custom course development, training strategy, and instructional technology.
References
- Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2024). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. MIT Media Lab.
- Mayer, R. E. (2020). Multimedia Learning (3rd Ed.). Cambridge University Press.
- Sweller, J., Ayres, P., & Kalyuga, S. (2011). Cognitive Load Theory. Springer.
