A couple of years ago, my wife bought my then four-year-old son a supercool set of wooden ramps, which could be combined with our furniture to create courses through which little balls could run. Building the first course was easy, but, as our ambitions grew, the difficulty level rose. Could we make the balls turn corners? What about generating enough momentum for them to go briefly uphill? Would the ottoman support a ramp? What about the piano bench? The variations seemed endless. Two or three times a month for the next two years, we tweaked our techniques or incorporated new elements into complicated routes. By the time my son got bored with the whole exercise, earlier this year, we’d become ramp experts, capable of seeing untapped ramp potential in almost any random object—an old stuffed animal, a dustpan, a strangely shaped cardboard box.

I hadn’t expected to enjoy the ramps so much. But they turned out to be a portal into a special kind of experience, one that involved an exploratory loop of trying, failing, revising, and trying again. This is a special kind of effortful repetition. In many parts of our lives, we repeat ourselves in order to optimize or perfect a task: once, at a corporate-bonding event, I raced a go-kart around a track for fifty laps, gradually refining my racing line so that my times decreased; a couple of mornings a week, my son makes his baby sister a cheese omelette, and each time he strives to produce a more perfect cylinder. That’s repetition as the dogged pursuit of an ideal.

There’s another kind of repetitive activity, though—one that combines loops of repetition with variation to allow for exploration and discovery. It invites us to change our approach in an open-ended way that, over time, deepens our inner resources. If you cook a stew each week, you don’t have to follow a set recipe each time; you can endlessly adjust your ingredients, developing your intuition as a cook and adding pages to your mental recipe book. When you practice a piece of music over and over, you can vary your performance as you go, conjuring new shades of composition and feeling. When you paint a landscape and then repaint it, you can observe how changes in the light reveal new aesthetic opportunities and aspects of nature. In all these cases, repetition doesn’t lead to a single, Platonic end point. Instead, it contributes to an expanding set of possibilities, which reflect your growing consciousness.

There are practical reasons to engage in activities like these. Some of those stews, performances, and landscapes might be transcendent, perhaps in ways you wouldn’t have envisioned. It’s also illuminating to learn, one repetition at a time, that there are various paths to success. After making a lot of tasty vegetarian stews over the years, you become ready to turn whatever’s in your kitchen—beans, tomatoes, that weird vegetable your wife got at the farmers’ market—into a delicious meal for your vegetarian friends. You’ll have mapped what engineers call the “design space”—the set of winning possibilities inherent in your endeavor. This is, broadly speaking, an evolutionary way of working. In evolution by natural selection, each new generation of creature is almost exactly the same as the previous one, but with subtle variations which can turn out to be valuable when the environment shifts. Because the environment is always shifting, no species can ever achieve perfection; instead, it’s variation that insures survival.

Something similar can be true in our individual lives. We’re drawn to activities that invite us to grow, by trying and trying again, because we want to evolve as people. Life is mostly repetitive—wake, eat, work, sleep, repeat—and each day can feel like an unsatisfying circle. But repetition with variation broadens us. It makes our circular days into spiralling journeys. “The spiral is a spiritualized circle,” Vladimir Nabokov wrote, in “Speak, Memory.” “In the spiral form, the circle, uncoiled, has ceased to be vicious; it has been set free.” This way of being, for which we don’t even have a name, is part of what makes us feel that we’re really living our lives instead of just going through the motions.

We’re so used to trying things for ourselves that it seems bizarre to imagine us ever stopping. And yet, more and more, it’s becoming clear that artificial intelligence can relieve us of the burden of trying and trying again. A.I. systems make it trivially easy to take an existing thing and ask for a new iteration. The technology is still developing, and yet already an A.I. can give you a custom recipe based on a photo of what’s in your fridge. Songwriting A.I.s can generate version after version of a new tune; image-creating A.I.s can tweak an image endlessly. Is the automated exploration of alternatives a good substitute for the organic equivalent? Is this kind of variation-creation the same thing as human creativity? These are important questions to ask because, as A.I. grows more powerful, we will be tempted more and more to give up in advance and let it figure things out for us.

People are adopting A.I. at different rates, and so far only some of us have experienced this temptation. Trust me: it creeps up on you. Not long ago, in her weekly newsletter, the owner of my local bookstore wrote about some trees in front of her shop, which the town had cut down; it had promised to plant new ones but hadn’t yet done so. She suggested that we write to our town supervisor to complain. My wife took the relevant bits of the newsletter, pasted them into Claude—the A.I. system offered by the firm Anthropic—and asked it to redraft them as an e-mail to the county. When the result seemed overwrought, she had it rework the e-mail, and then sent it. She got a prompt and courteous reply from the town, promising that the project would soon resume, which it did. A significant amount of mental effort—of writing, and perhaps rewriting—had been rendered unnecessary, to useful effect.

A couple of weeks later, the day before we were supposed to go on vacation, our son got sick. I suspected that he had norovirus, which is like the stomach flu squared. I postponed our flights and asked ChatGPT to tell me about the likely course of the illness, which is highly contagious, as it marauded through our family. OpenAI’s newest model, o3—which some observers have judged as exhibiting artificial general intelligence, or A.G.I., a humanlike level of cognition—explained what was in store with what I can only describe as verve. (Our toddler daughter probably wouldn’t warn us about her own nausea, it warned: “She might be happily toddling along and then suddenly—bam.”) All this was familiar enough—I already knew that A.I. was good at knowledge retrieval. But then the system offered to plan the crisis for me. “If you want, I can help with a survival checklist for handling both kids if she gets sick too—like which cleaning products actually work, or which surfaces are sneakily germy (hello, doorknobs and light switches),” it wrote. “Or a tongue-in-cheek ‘Parent Plague Protocol.’ You need backup, Josh, and I am here for it.” Needless to say, I’ve navigated many family illnesses without the help of A.I. Those experiences have made me a more capable and confident parent. Still, I wondered whether my approach should change.

Thinking takes effort. In a 2024 paper titled “The Unpleasantness of Thinking: A Meta-Analytic Review of the Association Between Mental Effort and Negative Affect,” three psychologists reviewed a hundred and seventy studies conducted in twenty-nine countries and concluded that, for almost everyone everywhere, “mental effort is inherently aversive”—that is, no fun. We might add that redoing one’s mental effort is even less fun. Once you’ve written an e-mail, or coded an app, the last thing you want to do is rewrite or recode it. An A.I., by contrast, can’t feel cognitive discomfort (or anything else); if you ask it to redo its work differently, it will oblige not just instantly and repeatedly but without tiring. Commanding an A.I. to do something only once could even be considered a waste. If you do that, you’re like a Victorian traveller who’s impressed by how effectively a locomotive can carry passengers and hand luggage; you’re missing the fact that it can also ferry many tons of coal and steel. An A.I. can haul a vast cognitive load—instead of asking for simply one recipe, you should ask for ten. The whole painful cycle of trying, failing, revising, judging, and redoing can be replaced with something simpler: plucking the most suitable result out of a pile.

In the physical realm, we’re familiar with the costs of laziness. If you drive everywhere and never walk, if you binge-watch TV shows instead of playing sports, if you get lost in video games instead of going on hikes, then you grow lethargic, unfit, and inflexible. You become less willing to stride to the top of the hill, and more likely to lose your grip on the handrail when you trip on the stairs. We know all this, but we’re lazy anyway, because the technologies that encourage physical inactivity offer so many practical benefits. In the cognitive realm, artificial intelligence is similarly double-edged. The same technology that allows us to skip the unpleasant work of thinking for ourselves can also help us automate the writing of repetitive e-mails and the discovery of new drugs. Tasks that once took hours can be accomplished in minutes; problems can be instantly analyzed; befuddling subjects can be made welcoming through conversation. It may be hard to take advantage of these opportunities without losing the inclination to climb mental hills under our own power.

The gym offers one model for mental effort in the age of A.I. Perhaps we’ll come to see thinking for ourselves as an optional and semi-recreational form of self-improvement, something we choose to do because we want to make our minds stronger. But gym-going has turned out to have advantages and disadvantages. For some people—the super-committed—it can open the door to extreme fitness. But it also produces weekend warriors—muscle-bound bros who can bench their own weight but tweak their backs while collecting toys on the lawn. Intellectual gym-going risks leaving certain mental muscles untrained. Perhaps there are especially unpleasant mental tasks (for example, learning how to adjust and readjust your family’s travel plans when illness strikes) that offer benefits we too readily overlook: the cultivation of patience, the taming of frustration, attention to detail, the calibration of optimism and pessimism.

One of the paradoxes created by physical automation is that people can be physically effective and physically weak at the same time. Even if you’re in bad shape, you can easily transport hundreds of pounds of luggage in your car. Similarly, if A.I. turns out to be as effective as many researchers think it will be, then people who use it well may be able to produce effective intellectual products—reports, experiments, business strategies, and the like—without themselves doing intellectual work. In such a future, how will we gauge our own mental vitality? People who are serious about fitness have all sorts of ways to keep track of their performance: they wear heart-rate monitors, or try to lower their marathon times, or take on new challenges meant to reveal their weak points. We’re not used to scrutinizing our levels of mental engagement that way. We may have to start.

There is, meanwhile, an inner dimension to performing mental tasks for yourself, or deciding not to. Just as physical effort reshapes our bodies, mental effort reshapes our minds and, therefore, our identities and our selves. Consider two cooks. One proceeds in the traditional way, learning how to cook through years of experimentation, ascending from the mastery of individual recipes to a broader and more intuitive feel for how ingredients and techniques go together. The other relies on an A.I. to generate recipes one by one, often based on whatever happens to be on sale or in the fridge. An A.I. can do this successfully because, just like the first human cook, it’s been exposed to countless recipes and used them to develop intuitions about cooking. It, too, has ascended, through a training process, from particulars to generalities. In contrast, the second human cook never has to develop those intuitions; he stays at the level of individual recipes. Of the three chefs—the first cook, the second cook, and the A.I.—he is actually the least well trained.

Does this mean that the second cook is different, as a person, from the first cook? Certainly, his mind, his capabilities, and his story are different. The way in which he makes choices is different—it’s one thing when a masterly cook makes you a meal based on a recipe that he treasures after a lifetime of cooking, and another when someone uses a recipe an A.I. has chosen. And we might say that, to some degree, his character is different. The second cook might make a good dinner, but he’s not someone who has tried to learn to cook, failed, and eventually succeeded. He hasn’t really lived life as a cook; he’s just going through the motions.

Suppose the differences between these two cooks were repeated in many domains of intellectual labor. Imagine, as an extreme case, two individuals, one of whom attempts to solve problems by herself, and the other of whom often enlists the help of an A.I. when mental labor is required. They would be quite different people. One would be a thinker, the other a consumer. One would have a mind shaped by learning; the other, a mind shaped by preferences. One would have a wide range of evolved, adaptable, internalized competencies; the other, a sense of what to ask for. In real life, of course, these won’t be two separate people: they’ll be two potentialities within each of us. In how much of our thinking lives will we be passengers, rather than pilots?