I’ll admit something that bothers me a little. Most mornings now, my first move when I hit a hard sentence is not to think harder. It’s to open a tab and ask a machine. I caught myself doing it last week, reaching for the shortcut before I’d even tried the climb, and it stopped me — not because the machine gives bad answers, but because I’d skipped the part of the work that used to be the work.
I am not a cognitive scientist or a psychologist, so take what follows as one curious person reading the research and noticing things in my own day, not as advice or settled fact. The studies here are findings from particular groups of people, often based on what they say about themselves or on single trials, not universal rules about you or me.
The retention finding
A 2025 trial put numbers on the thing I’d been feeling. The study, run by André Barcaui and published in the journal Social Sciences & Humanities Open, randomised 120 undergraduate business students at a Brazilian university into two groups: one studied with ChatGPT freely, the other used only traditional methods. Their task was preparing a 10-minute presentation on AI and machine learning. Forty-five days later they sat a surprise test. As PsyPost reported, the students who studied with the chatbot answered only 57.5 percent of the questions correctly. The ones who studied the old-fashioned way answered 68.5 percent.
As the study authors write, “This suggests that unrestricted ChatGPT use impaired long-term retention, likely by reducing the cognitive effort that supports durable memory.” In plain terms: if the chatbot does the hard mental work, your memory has less to hold onto later. The usual caveats apply — one university, one cohort of business students.
It reminds me of a secondary-school teacher I had who used to say that if you truly understand something, you won’t forget it. I’ve tested that line against my own life for years, and it mostly holds. Perhaps, that’s the difference here.
The wider pattern
One trial on its own is just one trial. What makes me sit up is that other 2025 work points the same way, from a different angle. Michael Gerlich surveyed 666 people and, writing in Societies, reported that “The findings revealed a significant negative correlation between frequent AI tool usage and critical thinking abilities, mediated by increased cognitive offloading.” That last phrase means people handed more of their thinking to the tools. This is a link, not proof that AI causes anything. People who lean on the tools more tend to score lower on critical thinking. The study can’t tell us which way the arrow runs.
A separate survey of 319 knowledge workers, run by Microsoft Research and Carnegie Mellon, found that “higher confidence in AI was associated with less critical thinking, while higher self-confidence in one’s own abilities was associated with more critical thinking.” Again, a link, not a cause, and based on self-report. But the shape keeps repeating. The more you trust the tool to do the thinking, the less of it you seem to do yourself.
Why the gap makes sense to me
If the effort is what builds the memory, then skipping the effort and keeping the answer is a strange trade. You end up holding the output and none of the structure that would let you rebuild it. The reason the retention study points to, less mental effort, is just a tidy way of saying the part you skipped was the part doing the learning.
The line I’ve drawn in my own work
A recent evening made this concrete. I’d written a post and the editing didn’t feel right, the kind of wrong you can’t name. The easy route was to clean it up and ship it. Instead I went down a research rabbit hole for a couple of hours, didn’t post, and came out the other side with something I believed. It cost me an evening, and the easier version would have cost me nothing and taught me nothing.
I’m not drawing a rule for you — I have no standing to. But for myself the line has gotten clearer. The machine can carry what’s cheap, and I’ll keep doing the part that’s expensive, because the expense seems to be where the understanding lives. Whether I can hold that line is another question. The pull of the easier route is steady, and most days I’ll probably still reach for the tab before I should. What I’m trying to do is notice which reaches I can afford and which one is quietly taking the thing I came for.
The post Heavier reliance on AI correlates with weaker critical thinking and recall. One 2025 trial saw AI-assisted learners retain just 57.5% on a delayed test versus 68.5% for those who went without appeared first on Space Daily.