Here is the finding that should make every thoughtful AI user pause. Researchers at the University of Pennsylvania observed something they call "cognitive surrender" — a pattern where people trust AI output over their own judgment, even when the AI is wrong.
A separate study published in 2025 found a strong negative correlation between heavy AI tool usage and critical thinking scores. The more people offloaded their thinking to AI, the worse they performed on standard critical thinking assessments. The BBC reported on this research under a headline that says it all: "How to stop AI from turning your brain to mush."
But here is the nuance that matters. The problem is not AI itself. The problem is what you offload to it.
The safe offload and the risky one.
Cognitive offloading is the term for using a tool to reduce mental effort. You do this all the time. You write down a phone number instead of memorizing it. You set an alarm instead of keeping an appointment in your head. You use a calculator instead of doing long division. These are safe. They free up mental space for harder problems.
The risky offload happens when you hand over your reasoning — not just your memory — to the machine. Asking AI to summarize a complicated email is fine. Asking AI to decide whether that email should make you angry is not. One is retrieval. The other is judgment.
The Oxford Review study found that older participants — those 46 and older — showed lower AI dependence and stronger critical thinking skills compared to younger users. That is not an accident. Decades of making decisions, catching errors, and trusting your own instincts build a muscle that younger workers have not had time to develop.
A simple routine.
Here is a practical way to keep the benefits of AI without the cognitive cost.
Use AI for retrieval: "Summarize this article in three bullet points." "Find the five most important dates in this contract." "Translate this email from Spanish."
Keep judgment for yourself: "Does this summary miss anything important?" "Is this contract fair?" "Does this response match my tone?"
The Stack Overflow blog put it this way: "A system is only as reliable as your ability to identify its failures, which requires keeping the core problem-solving process active."
In other words, use AI as a research assistant, not a brain replacement. Let it handle the legwork. You handle the thinking.
Your move this week.
Pick one task you currently let AI handle end-to-end — writing a whole email, drafting a response, planning a trip. This week, do half of it yourself. Use AI for the first draft or the research. Then rewrite, question, and improve the output using your own judgment. Notice how different the result feels.
Sources: BBC Future, "'Think Outside the Bots': How to Stop AI from Turning Your Brain to Mush"; Oxford Review, "How AI Tools Affect Critical Thinking Skills and Cognitive Offloading"; Stack Overflow Blog, "AI Is Becoming a Second Brain at the Expense of Your First One"; APA Monitor on Psychology