AI dependency vs. just using a good tool: where's the line?
Calculators didn't make anyone bad at arithmetic in some fixed, universal sense. But they did change what "good at arithmetic" means. AI is doing something similar, faster and across more domains at once.
The comparison people reach for is always a calculator, or spellcheck, or GPS. Tools that took over a mental task and nobody mourned it. The comparison isn't wrong, exactly. It's just incomplete. Calculators replaced one narrow skill. Conversational AI can replace drafting, analyzing, deciding, and remembering, sometimes all in the same afternoon. Scale changes what's at stake.
A working definition
Here's a distinction that holds up reasonably well across the research on this: a tool augments a skill you keep. A dependency replaces a skill you lose. The test isn't how often you use something. It's whether you could still do the underlying task, reasonably well, if the tool vanished tomorrow.
By that test, GPS is a genuine gray area, most people's sense of direction has measurably weakened since turn-by-turn navigation became standard. But GPS failure is rare and low-stakes for most people's daily function. AI failure, or AI unavailability, touches far more of a typical knowledge worker's day.
Three questions that cut through the noise
- Could you still do this without it, at a reasonable standard, today? Not perfectly. Reasonably.
- Are you checking its output, or trusting it? Checking means the skill is still yours. Trusting without checking means it's moved to the tool.
- Has your standard for "good enough" quietly dropped to whatever the tool produces? This one's sneaky. It doesn't feel like a loss because the output still looks fine.
What the research actually shows
A 2026 preregistered study out of FGV and UFRJ found an 11 percentage point gap in what students retained 45 days later, comparing AI-assisted learners to unassisted ones. Not a small effect. And a RAND report from the same year found 62 percent of students now use AI for homework, up from 48 percent just seven months earlier, with two-thirds of those students believing, correctly according to the data, that it's affecting their critical thinking.
None of that means AI use is bad. It means the convenience is real and so is the cost, and pretending only one of those is true doesn't help anyone make a good decision about their own use.
Where this leaves you
Not every AI habit needs fixing. Most people reading this are somewhere in the middle, using AI heavily for some things and barely at all for others, which is a perfectly reasonable place to land. The goal isn't zero AI use. It's knowing, specifically, which of your skills you're keeping and which ones you've quietly handed off.
Frequently asked questions
Isn't this just moral panic about a new technology?
Every new tool gets some of that, and some of the concern about AI probably is overstated. But the retention and critical-thinking research cited above is measured, not speculative, so it's worth taking seriously without treating it as catastrophic.
What if I need AI for my job and can't just cut back?
That's most people. The point isn't to use AI less across the board. It's to be deliberate about which specific skills you're willing to lose and which ones you want to keep sharp on purpose.
← Back to the assessment · Related: What happens to your brain when AI does your thinking for you