Right to Struggle
Some friction is not a problem to be solved. Sometimes the struggle is the whole point, and I want to keep the right to choose it.
Two friends recently told me they had stopped using AI. Not because they think it is useless. The opposite. They had both found it useful enough to become uncomfortable with how useful it was.
One is making a visual novel. He had been using AI to help with parts of the production, until something started to bother him. Whenever he got stuck and reached for it, the work took on the same smoothness and sameness he sees everywhere now. And he was not getting any better at making the things himself. So he stopped. He went back to learning the rendering tools properly, spends more time getting things wrong, and the results are slower and messier. But they are his, and he is learning something on the way.
The other stopped asking AI questions at all. She went back to Google because she wanted to decide for herself which sources were credible. There is something old-fashioned about that, and I mean it as a compliment. She did not want an answer. She wanted to do the work of finding one. The irony is that Google now puts an AI-generated answer at the top of most searches, which makes even this small act of resistance harder every month.
That part bothers me. Not because I think AI is bad, but because choosing not to use it should stay a real option.
There is an old story that feels remarkably current. In Plato’s Phaedrus, Socrates recounts the Egyptian god Theuth presenting writing to King Thamus as a great invention, something that would improve memory and wisdom. Thamus disagreed. Writing, he said, would weaken memory, because people would come to rely on marks outside themselves instead of developing the ability to remember. He was not wrong. Writing did change human memory. We became less dependent on holding everything in our heads, and we did it anyway. Thankfully.
The same argument comes back every time a technology takes a piece of mental work off our hands. Calculators mean we do not practise arithmetic the way our grandparents did. Spellcheck means a lot of us cannot reliably spell anymore. GPS means I can cross a city I have never visited without once knowing where I am. Nobody seriously argues we should abandon these things because they make us worse at the underlying skill. “This tool makes me dumber” may be the oldest complaint in technological history, and sometimes it is even correct. We decide that being a little worse at one thing is worth being dramatically better at something else. That is how progress works.
So I am not going to argue that we should reject AI. But something different is happening here, and it is worth being precise about what.
A calculator gives me the answer to a calculation. AI can give me the calculation, explain it, write the report that contains it, build the presentation explaining the report, and then offer three alternative ways to make the same argument. It does not remove a mechanical step. It can remove the entire process. There is a difference between reaching for a calculator because I do not want to spend thirty seconds multiplying two numbers, and reaching for AI because I do not want to learn how to do something at all. The first is efficiency. The second can quietly become dependency, and the dangerous part is that from the outside they look identical. I can produce something excellent without becoming any better at producing it. The output improves while the person producing it stays exactly where they were. I do not think we have language for that trade-off yet.
I have been building a game in Godot. I am still learning the engine, which means most of the process consists of discovering that I have misunderstood something, trying three increasingly desperate solutions, and eventually finding the cause was something embarrassingly simple. Recently I hit a problem I could not solve. So I asked Claude. I gave it the material and it found the fault immediately. Then it offered to fix it. I said no.
That might sound irrational. I had just asked the machine to find the answer, so why not let it change the code as well? Because I was not trying to ship a game. I was trying to learn how to make one. If Claude fixes every problem I hit, I get a working game. If Claude shows me why my code is broken and I fix it myself, I get a slightly better understanding of Godot. The second outcome is slower, and it is the entire reason I am doing this. There is a specific satisfaction in solving a problem yourself after an hour of struggling with it. It is completely irrational from a productivity standpoint, and it is also one of the reasons people make things at all.
This is where the conversation about AI usually goes wrong. We have become very good at spotting friction and removing it, and most of the time that is a good thing. Waiting for a taxi, replaced by an app. Standing in line, replaced by ordering online. Typing the same report three times, solved with automation. Most of that is progress. But friction is not always a problem to be solved. Sometimes the struggle is the activity. A musician practising scales is not suffering through an inefficient way of producing music. A programmer debugging their own code is not always wasting time that should have been automated. An artist learning how light behaves is not just failing to use the available tools. The struggle is building the capability, and sometimes the capability matters more than the result in front of you.
That is why I understand my friends. They are not saying AI is bad. They are saying there are things they want to be able to do themselves, which is a much more interesting argument and a harder one to wave away.
My friend’s return to Google is the same argument in different clothes. Why deliberately swap an assistant that can summarise dozens of sources for a search engine that makes you click through them one by one? Because she was not optimising for speed. She was optimising for judgment. When an AI hands me an answer, I can weigh the answer, but I cannot always see what happened before it arrived. Which sources were chosen and which were ignored. What assumptions were made. Whether the sources disagreed. Whether some important qualification got compressed away in the summary. If I search myself, I am forced to meet those questions. It is slower, but the friction is doing something useful. I am not just taking an answer someone handed me, I am practising the skill of deciding what deserves to be believed.
And then we reach the slightly absurd situation where even that choice is quietly disappearing. Search for something, and more and more the machine answers before you have decided whether you wanted an answer from a machine at all. That is where my discomfort actually lives. Not with AI. With the removal of the choice.
Technology will keep taking skills from us, and that is part of the point, not a bug. I do not navigate by the stars because I have GPS. I do not memorise phone numbers because my phone does. There are capabilities I have willingly handed over and do not miss. But I want to be the one who decides which ones. That may be the whole distinction. Sometimes I want the machine to do the work. Sometimes I want it to show me how. Sometimes I want it to point at my mistake and then leave me alone to fix it. AI makes all three possible, and the challenge is remembering they are different goals, and refusing to measure every activity by the quality of what it puts out.
Socrates was right that writing changed what humans remembered. He was wrong if the lesson was that we should therefore refuse it. The better lesson is that every powerful tool changes us, and the question is never whether it will. It is what we are willing to become in exchange for what it lets us do. In the age of AI, one of the most useful skills left to us may be knowing when to put the tool down.
