Twice this week I opened my AI system, the one my minions nicknamed Claudius Decimus Maximus, and used it to finish a piece of work that belonged to somebody else. One job took fifteen minutes. The other took two, barely enough time to make coffee. Both were the kind of thing that would have eaten a person’s hours, maybe weeks, if they’d sat down and done it themselves. I handed both back finished. Neither person used what I gave them.
That’s the whole event. It’s small. It’s also been bothering me all week.
What actually happened
Nobody asked me to do this. I saw two people stuck on something I could see the shape of immediately, and I did it because I could, not because anyone requested it. I sent the finished work over the way you’d hand someone their coat. Here, this is done, go do the next thing.
Both declined. Not rudely. Just quietly, the way you decline a second helping you don’t really want. Each person went back to the version that costs hours instead of the version that costs nothing.
The part I understand
If you don’t know how to get good output from these tools, refusing them makes sense. You’ve probably tried an AI system once, gotten something generic or wrong, and decided the whole category isn’t worth your time. That’s a reasonable conclusion from a bad sample. I’ve had output from my own systems that I wouldn’t hand to anyone, and I built the thing. Skepticism about a tool you haven’t seen used well is just good judgment.
That’s not what happened here. I read both jobs myself before I sent them over. Nothing to fix, nothing to guess about. The person could see it was right. They just didn’t want it.
The part I don’t
Somebody hands you a finished, verified answer and you say no thanks, I’ll take the long way. That’s not a decision about AI at all. It’s a decision to keep struggling on purpose when the struggling wasn’t buying anything.
Researchers call a piece of this algorithm aversion. Berkeley Dietvorst ran a set of experiments showing people trust an algorithm less than a human after watching it make one mistake, even when the algorithm is demonstrably more accurate overall. Watch a person screw up and you shrug it off as human. Watch a machine screw up once and you write off the whole category. That explains some resistance to AI in general. It doesn’t explain refusing a specific result you’ve already confirmed is correct.
The earning-it problem
I think it’s the same instinct that makes a nursing student want to do the dressing change herself instead of watching the preceptor do it faster. Competence gets earned by doing the thing, not by receiving the output of the thing having been done. Accepting a finished piece of work from somebody else’s AI system isn’t just accepting help. It’s accepting that the hours you would have spent proving you could do it yourself weren’t necessary, and that’s a harder thing to swallow than it sounds.
Neither person called it cheating. But the shape of the refusal matches what cheating avoidance looks like everywhere else. You don’t take the answer even when it’s sitting in front of you, because taking it means admitting the long way wasn’t required.
I built NursingEdAI because faculty were losing the hours they needed for the parts of the job that actually require a person: the judgment calls, the mentoring, the conversations that don’t fit a template. Most of the resistance I see to it isn’t about whether it works. It’s about whether using it counts.
I don’t have a clean answer for why a gift you can verify is correct feels worse to accept than the same result earned the slow way. I know the two people I helped this week aren’t unusual. Most people reading this have turned down the fifteen-minute version and picked up the fifteen-hour one at least once. Ask why, and the honest answer keeps landing in the same place. It felt like cheating.