- Lumiera
- Posts
- 🔆 Ways to prevent AI from making you dumber: "Your Brain on ChatGPT"
🔆 Ways to prevent AI from making you dumber: "Your Brain on ChatGPT"
Making sense of the popular research paper, tech leaders in the US Army, and more
Emma G Moller & Allegra Guinan
June 26, 2025
Was this email forwarded to you? Sign up here
🗞️ Issue 76 // ⏱️ Read Time: 7 min
In this week's newsletter
What we’re talking about: The neural and behavioral consequences of AI-assisted thinking, based on the first study to measure brain activity during AI tool usage. We are also highlighting the limitations of this study.
How it’s relevant: With AI becoming ubiquitous in education and work, understanding its cognitive impact isn't just academic, it's essential for anyone who wants to stay sharp while staying productive. Insufficient summaries of the study have been spread widely on the internet over the past weeks, so we are making sure you get a better understanding of what’s actually going on.
Why it matters: This research suggests that how we use AI today could fundamentally reshape how our brains process information tomorrow. The choices we make now about AI integration could determine whether we become more capable thinkers or more dependent ones.
Hello 👋
A groundbreaking MIT study has been spreading like wildfire across social media, and for good reason. It's the first to peek inside our brains while we're actually using AI tools, revealing some uncomfortable truths about cognitive debt. Today, we're breaking down what this research really means, and more importantly, how you can harness AI's power without sacrificing your mental muscle.
Picture this: You're writing an essay with ChatGPT's help, feeling efficient and productive. But what if I told you that 83% of people who use AI for writing can't accurately quote from essays they wrote just minutes earlier? Meanwhile, only 11% of people writing without AI tools face this same problem. This isn't just about memory. It's about what happens to our brains when we let AI do the heavy lifting.
Why This Research Spread Like Wildfire
This study hit a nerve because it addresses something many of us have quietly wondered: Are we getting smarter with AI, or just feeling smarter? As AI tools become standard in classrooms and workplaces, we're essentially conducting a massive, uncontrolled experiment on human cognition.
The researchers didn't just ask people how they felt after using AI: They actually measured what was happening in the brains of 54 participants while they wrote essays under three different conditions: Using ChatGPT, using search engines, or relying only on their brains.
What makes this research particularly alarming is that the cognitive effects persisted even after participants stopped using AI tools, suggesting potentially lasting changes to how our brains process information.
The Study: Your Brain on ChatGPT
The Experimental Design
Image of the participant setup for the study
Researchers divided participants into three groups:
Brain-only group: No external tools allowed
Search Engine group: Could use Google and similar platforms
ChatGPT group: Could use the AI assistant
Each group completed three sessions under their assigned condition. Then came the twist: In a fourth session, ChatGPT users were switched to brain-only mode, while brain-only users got access to ChatGPT.
Neural Connectivity: A Hierarchy of Mental Engagement
Brain scans revealed a clear hierarchy. The brain-only group showed the strongest neural networks, like a bustling city with multiple highways connecting districts. The numbers are striking: brain-only participants showed theta connectivity scores of 0.644 compared to just 0.331 for search engine users. Even more dramatically, 22 neural connections showed stronger activity in the brain-only group versus only 4 connections favouring the search engine group. Theta band activity is closely linked to working memory load and executive control. In fact, frontal theta power and connectivity increase linearly with the demands on working memory and cognitive control.
ChatGPT users showed the weakest overall neural coupling, with their brains looking more like isolated suburbs with limited connections between regions.
Memory Formation Failing when using ChatGPT
83% of ChatGPT users couldn't accurately quote from essays they had just written, compared to only 11% in other groups. The information never properly integrated into their memory networks, what researchers call "shallow encoding", due to outsourced cognitive processing.
Productivity Comes at a Cost
Here's the flip side: LLM users were 60% more productive overall. However, this efficiency came at a cost: Their brains showed "neural efficiency" patterns, suggesting they were relying on automated processes rather than deep cognitive engagement.
Cognitive Switching: Creating Dependencies
When ChatGPT users were forced to work without AI, their brains showed "under-engagement" - weaker brain wave activity indicating reduced cognitive processing. Their previous AI reliance seemed to create cognitive dependency.
Dictionary: Key Terms Made Simple
We are not brain scientists, so if you aren’t either, we won’t judge you (even if we know that we have at least a couple of brain experts reading this newsletter on a regular basis, so please feel free to add your input!).
Here’s a simple list of the most important words in this study, that we think might be useful for you as you continue your week.
Cognitive Debt: Like financial debt, but for your brain. The long-term reduction in critical thinking skills from over-relying on AI tools.
Cognitive Load: The mental effort your brain expends to process information—your brain's bandwidth.
Germane Cognitive Load: The "good" mental effort that builds lasting knowledge. AI can cause you to miss this crucial brain exercise.
Neural Connectivity: How well different brain regions communicate. Stronger connectivity generally means better cognitive function.
Metacognitive Laziness: When you stop monitoring your own thinking because AI is doing it for you.
The Problems with This Study
This is Groundbreaking, But Preliminary
This represents a crucial first: the first research to measure brain activity while people use AI tools in real-time. However, it's not yet peer-reviewed, meaning it hasn't undergone rigorous scientific scrutiny.