Now That AI Can Answer, What Should We Still Train Ourselves To Do?

I started thinking about how a person grows up in the age of AI not from some new tool, but from a line on Moonshot Academy’s site: raise inwardly rich individuals, and actively engaged citizens. It sounds like it’s about children. But the more I sat with it, the more it felt like a question turned back on those of us who already left school: if AI gets better and better at answering, summarizing, writing, and generating plans, what is left for us to train?

Start by taking apart a word. “AI can answer” — but answering means completely different things in different places. In an exam it’s matching the official key. At work it’s delivering a workable plan. In life, answering is bearing a choice. AI is superb at the first few. But in the sense of living in the world that comes after the choice, it isn’t answering at all — it produces the answer without having to live inside it.

When answers get cheap, the scarce thing is no longer the reply, but whether anyone is still willing to pose a question worth answering — and to live in the world that answer brings.

AI is a magnifier, not a ladder

The most obvious response is: then use AI to train yourself — make it a sparring partner, not a stand-in. Have it attack your weakest claim, hand you counterexamples, force you to write the ugly first draft before it gets to tear it apart. This is right, and there’s evidence for it: one much-cited study found that the more people trusted AI, the less they thought for themselves — while the people more confident in their own judgment were the ones who wrestled hardest with what it produced.

But the same fact has a shadow side. Using AI as a sparring partner and using it as a crutch are the same motion — both remove friction. Removing friction is the entire point of the tool: it writes for you, thinks for you, smooths your sentences for you. And judgment grows precisely in the friction of doing it yourself, getting it wrong, and going back over it. Once the friction is gone, most people don’t “free up energy for higher training” — they just stop training. So “using AI to train yourself” isn’t a stable state; it’s swimming upstream against the grain of the tool, a choice you have to make again every single day. Its default gravity is always pulling you down.

“Then design the environment, don’t rely on willpower” — which also doesn’t hold

So someone says: don’t count on self-discipline, build the training into your defaults — for anything that matters, write down your own judgment before letting AI in; have it surface the opposing case by default; show your work to real people every week. This is genuinely sturdier than just shouting “try harder.”

But it dodges two things. First, installing that anti-laziness scaffolding for yourself takes exactly the foresight and persistence we just said a tired person doesn’t have — it doesn’t abolish willpower, it only relocates it to the one-time moment of setup. Second, the people who’d bother to build such an environment are usually already the ones with the strongest judgment, the ones who least need saving; the people who most need the training are precisely the ones who’d never think to build it. The benefit, once again, goes to those who already have.

Maybe I’ve been asking the wrong question

Having pulled all of that apart, I started doubting this essay’s own title.

Because over a long enough horizon, “can AI do it” probably stops being the boundary — almost everything qualifies. The only boundary left is this: not whether it can do a thing for me, but whether I still want to do that thing myself.

Whatever I truly want to do, I’ll use AI to make myself stronger at it, and then do it myself; whatever I don’t want to do, I’ll let it do for me.

It’s a beautiful line, and it ruthlessly filters out a lot of fake loves — plenty of people think they love writing when what they love is being seen. Once the result can be generated cheaply, only one question is left: are you still willing to live through the clumsy, stuck, repetitive process?

But it has a last and deepest crack: it treats desire as a fixed asset you can take inventory of, as if sitting quietly and asking yourself were enough. Yet if the tool reshapes the person who uses it, then “what do I want to do myself” has no stable answer — it drifts, and it drifts downhill, because “do it myself” is always the more expensive option. The cruelest atrophy isn’t outsourcing the things you don’t want to do — that’s healthy. It’s that the act of outsourcing, year after year, quietly rewrites what you want at all. You don’t decide one day to give up the thing you swore you’d never hand over. You hand it over a little at a time, on one tired evening after another, without ever noticing.

Take away your skill and you’ll be angry. Wear away your wanting, and you won’t even miss it.

So I won’t give you a conclusion. I should even admit this: an essay arguing don’t hand your judgment to AI was itself revised together with AI — I can’t pretend it’s hand-made, only that every uncomfortable tradeoff in it is one I pressed down on myself. If anything reliable is left, it’s probably just this: go back and ask yourself that question on a schedule, and honestly note which way the answer is drifting. Not to defend some fixed answer — but so that, when it’s being quietly rewritten, you can at least notice.

Sources

• Moonshot Academy: frames itself around holistic, character-centered, innovative education, with the stated aim of “raising inwardly rich individuals and actively engaged citizens.”
• Lee et al., The Impact of Generative AI on Critical Thinking (Microsoft Research & Carnegie Mellon University, CHI 2025): higher confidence in AI is associated with less critical thinking; higher confidence in one’s own ability with more. ACM paper
• See also research on cognitive offloading: more frequent AI use correlates with lower critical-thinking scores, with heavier dependence among younger users.