A Mobility Scooter for the Mind?

Building Human Competence in an Age of AI

Growing up in a deeply conservative district at the height of the 1990s welfare reform debates, I lost count of the number of times I heard the maxim, “If you give a man a fish, you feed him for a day; if you teach a man to fish, you feed him for life.” Whatever the sincerity of the sentiment as a matter of care for the poor, taken at face value it reflected a conviction about the importance of human competence, a conviction that conservatives seem to have since forsaken. What mattered was not merely that all were fed, clothed, and sheltered; more important was that a man knew how to feed, clothe, and shelter himself and his family—that he could lay claim to the competence to make his way in the world, and the independence it made possible.

Today, I worry, we are at risk of forgetting this insight. The tech industry’s boosters enthuse about a world where machines can do almost everything for us, unleashing unprecedented productivity that showers each of us, like the disciples on Galilee, with more fish than we know what to do with. In many ways, we are already living in this world, enjoying more and more material affluence while able or willing to do less and less for ourselves (see my recent essay “Welcome to WALL-E’s World”).

Welcome to WALL-E's World

But of course it doesn’t have to be this way. Steve Jobs famously described the personal computer in the early ‘80s as a “bicycle for the mind”—which is to say a technology that both increases our output and our competence at the same time. And so the computer was for many users, at least until the advent of high-speed internet, which so accelerated access to information that it became more of an automobile for the mind, reducing us to a state of mental obesity that soon demanded a “mobility scooter for the mind,” which has now arrived in the form of generative AI. But could AI too function as a bicycle for the mind? Could it serve not just to boost human output, but to cultivate human competence? Many tech optimists think so, and point to their own experience.

They’re not wrong, but we also should be honest: when a mobility scooter is sitting in every driveway, it may be harder to mount that bicycle. Anthropic recently ran a study of how their coders used AI and their own brains—and the results weren’t pretty. In this essay, I will offer a typology of technology, seeking to discern how and when it serves human competence or destroys it, and will seek to apply this analysis to AI. I will warn you that this is my longest Substack post yet, but I figured you’d rather read it all at once than as a tedious two-parter. I will promise, at any rate, to omit needless words.

Why Competence Matters

First, though, why does competence matter? Why isn’t it enough to just have more stuff?

The first answer that comes to mind might be “efficiency”: even if it takes more time at the outset to teach a man to fish, it will save countless hours over the long run to spend a day in patient instruction than to return day after day with another dole of fish. Not only does the competent person no longer need a steady supply of outside help and inputs to perform a task, but he is apt to become ever better at the task on his own. For this is part of what competence is: the skill not just of knowing how to act or how to think in a given context, but of knowing how to learn from mistakes and become ever more competent. This is why AI forecasters look with such dread or excitement on the beginning of “recursive self-improvement,” when the AIs become competent enough to keep training themselves.

But I think we can go deeper. Competence matters because we are not mere machines. We are not merely concerned with output; we want to feel that it is our output. Thus the child proudly splitting logs or hammering in nails alongside his father doesn’t mind that his father can do it ten times as fast; what matters is not that the nails were all hammered in but that he hammered them in. As John Locke might say, “he mixed his labor” with the boards, so that something of himself is now in them—and just as importantly, they are now in him. He has developed some small measure of skill in the process, so that he knows that next time, he will be able to hammer a little faster and a little straighter. Indeed, this awareness of our own skill is part and parcel of our satisfaction in our work: it is why we derive less satisfaction from an achievement unlocked by “dumb luck,” as when a basketball player smiles sheepishly when a badly missed shot banks in at an unplanned angle. I don’t care so much about doing something, but about knowing I can do it, about having the competency to do it next time. There is some pleasure in arriving safe and sound at my destination after being guided by GPS; there is a greater pleasure in arriving by my own wits, even if I get lost and stumble on the address by luck; there is the greatest pleasure by far in finding my way there and knowing that I now know it, that I have taken personal dominion over this little piece of the world. I would wager that one of the deepest reasons for our societal malaise today—spiking depression, deaths of despair, plunging birth rates—is the loss of such satisfaction and self-worth, as we outsource our agency to Siri, Alexa, and Google. I shudder to think of the despair that awaits us a few years hence if we continue on this same path.

Third, though, competence matters because we do not live merely for ourselves, but for others. “Let the thief no longer steal, but rather let him labor, doing honest work with his own hands, so that he may have something to share with anyone in need,” writes St. Paul. The conservatism of my youth often espoused an ideal of a kind of rugged independence. This is far better than slovenly dependence, but better than both is interdependence, which is after all the good that independence should serve. I cannot either give a man a fish or teach a man to fish unless I have first mastered the art of fishing myself. Having once mastered it, I take joy in sharing a day’s catch with a friend, and an even greater greatest joy in sharing the secrets of my craft with a new beginner. “It is more blessed to give than to receive,” and I will usually have little enough to give unless I have first cultivated some competence. Today, we seem content merely to point others to a fish dispensary. On the very rare occasion when someone else still asks us for directions or information, we pull out our phone. Parents enthuse that AI now means they don’t have to hem and haw over answering their kids’ questions; they just put ChatGPT in front of them instead. Though I sympathize with the temptation, I want to weep every time I read such accounts.

Let Me Google that For You

Technology and the Crisis of Competence

It is easy to anticipate the retort that is waiting on some lips: “isn’t this just what technology does?” Isn’t it the purpose of technology to replace human labor by doing more than we can do ourselves? It extends human capabilities by in some measure outsourcing them, allowing us to produce more while making us painfully aware that it is no longer exactly we who are producing, and that while richer, we are less and less capable than our rugged homesteading ancestors. This is simply the Faustian bargain of progress, isn’t it?

We might reply in two ways. First, we might recognize this, with C.S. Lewis, as the “Irishman’s Two Stoves” phenomenon. Till now, technology has invited us to hand over the work of our lower faculties to machines, so that we might have more leisure to develop our higher faculties. But what happens when, in search of ever more output with ever less labor, we hand over our very highest faculties? What then? “This time,” observes Lewis, “the being who stood to gain and the being who has been sacrificed are one and the same.”

Second, however, we could contest the premise—and that is what I intend to do here. It is not in fact the case that technology always involves handing over human capabilities to a machine. I mentioned hammering nails and splitting logs above. Neither of these are things that human beings could do at all without technology; our bodies are simply not constructed for it. And neither of them are activities we associate with weak, flabby men; rather, they both tend to build muscle. Like the bicycle, they seem to be technologies that extend not merely our output but enhance our native capacities. This, then, is the urgent question for us today, standing on the threshold (or the precipice) of transformative AI: how do we design, deploy, and use more technologies as hammers or bicycles, rather than conveyor belts to WALL-E’s world?

A Typology of Technology

First, let’s make sure we understand the options clearly. I have spoken of bicycles and mobility scooters, but I would like to fill out this typology of technology with five machines of locomotion: the bicycle, the motorcycle, the car, the mobility scooter, and the exercise bike.

The Bicycle:

What makes the bicycle so special? Why do we enjoy it so much, and envy those societies (like the Dutch or the Danes) that seem to use it as their primary mode of transportation? The bicycle leverages existing muscles so as to both accomplish more and build more muscle mass; it increases output and enhances native capabilities. Why? If a bike allows me to travel five miles using the same energy as I would expend in running just one mile, then won’t I just expend less energy overall? Perhaps, but usually not. Most of us find biking not just easier on our joints, but actually more pleasurable than running: There is something exhilarating about feeling the wind on your face, about the subtle interplay of centrifugal forces as you navigate a turn at high speed. As a result, we may well bike eight miles instead of running just one—we will actually exercise more than we otherwise would have!

The Motorcycle:

The motorcycle clearly represents a step change improvement on the bicycle from the standpoint of pure efficiency in locomotion. Rather than spending 500 calories to travel ten miles, I might spend twenty. The motorcycle’s piston engines do not amplify or leverage the activity of my bodily muscles, but simply replace them, allowing me to travel much further, but at the cost of sacrificing my own engaged agency and cultivated competence. Or do they? Well no, it’s not that simple. Here I must rely on Matthew Crawford, for I am certainly no motorcyclist—and the reason why is quite simply that I don’t remotely have the competence or the will to cultivate it. Motorcycling—at least as often practiced—is a thrilling, high-risk, high-skill activity. It takes extraordinary hand-eye coordination, muscle memory, and balance to bank a sharp turn at 70 mph—not to mention courage and grit. The motorcycle, then, while almost entirely replacing our native bodily capabilities, makes possible the cultivation of a new and in some measure higher set of capabilities. I may, if I am not careful, allow my muscles to atrophy, but I will acquire new skills in which I may take genuine pride. The motorcycle, then, replaces existing muscles so as to accomplish more and develop new human capabilities.

The Car:

Now, while the car can be similarly used to cultivate such exhilarating skill, as in the case of the race-car driver or the off-road pickup truck adventurer, it generally does not, especially in its more recent iterations. The car exists above all to get us from Point A to Point B more efficiently. Like the motorcycle, its internal combustion engine is in no meaningful way an extension of my own muscles after the initial turn of the key or push of the button; after that, the barest movements of my arms and hands suffice to guide this two-ton metal capsule down the highway. Once, this still required the cultivation of considerable skill, but much less today in the age of dynamic cruise control and stay-in-lane steering overrides; soon, full self-driving mode will render the skill of human drivers entirely obsolete. What we will have then is a pure example of a labor-replacing technology that is justified in terms of the other things it frees us up to do. I may be increasingly clumsy and obese, but at least I can get to work faster and design cool things on my computer, or drive my kids to sports games so I can watch them exercise muscles I have long since given up and hold off their own inevitable obesity a little longer. The car, then, replaces existing muscles so as to accomplish more and free us up to potentially develop alternative capabilities; it is a necessary evil of sorts.

The Mobility Scooter:

What, then, of the mobility scooter, which has become an increasingly ubiquitous sight in the grocery store or the airport? Clearly, it serves a valuable purpose for those who have lost mobility through no fault of their own, and in terms of convenience at least marks a significant improvement on the wheelchair. However, we might well ask whether even the permanently disabled are better off operating a push-button device than wowing us with the dexterity and upper-arm strength they’ve developed from years in a wheelchair. But most users have been disabled not by disease, but by technology. The automobile, the screen, and the food-processing factory have together conspired to render them all but immobile; they are premature denizens of WALL-E’s world who hold up to each of us a mirror of the future that awaits us if we are not careful. Having used technology to deprive ourselves of the power of self-locomotion which we once shared in common with the lowest animals, we now use it as a permanent substitute, allowing us to get from point A to point B on battery power. The mobility scooter replaces muscles that we have already allowed to atrophy, enabling us to accomplish things on its power that we once accomplished on our own.

The Exercise Bike:

These four might seem to have exhausted our typology of locomotion technologies, but I think that there is one more category worth considering: the exercise bike. It resembles the bicycle in every way except that, well, it doesn’t actually go anywhere. It is a locomotion technology that doesn’t locomote. Rather, it exists purely to help us grow (or regrow) our capacity for locomotion—if we are recovering from injury, perhaps, or seeking to find our way back from obesity to fitness. It offers a simulation of the benefits of a bicycle, but in a controlled environment, without the risk of falling. It is a crutch, to be sure, but one that, if used properly, we will come to depend on less and less. The exercise bike leverages weak existing muscles to gradually build more and more muscle mass, so that we may again be able to accomplish things on our own power.

As you were reading this, I suspect, plenty of applications to digital technology may have already suggested themselves to you. Perhaps it is already obvious to you, indeed, that digital technology may function in any of these five ways, depending on the tool, the use, and the user. This is especially so for AI. This essay has already run quite long, however, so I will conclude with the briefest sketch of how generative AI that might fall into each of these boxes. I will go in reverse order so as to conclude on a high note, suggesting ways from my own personal experience in how AI can serve as a “bicycle for the mind.”

AI in Service of Human Competence

Exercise Bike for the Mind:

One of the most talked-about applications of AI these days is in the context of education. It is probably no secret by now that I do not share the strangely universal enthusiasm on this score, for reasons I have written about most recently two weeks ago. However, even in my lengthiest critique of AI EdTech last fall, I ended by suggesting a number of ways in which AI could serve as an excellent educational tool. It is entirely impossible to imagine thoughtfully-designed AI educational applications that functioned much as an exercise bike does: allowing a young learner with few cognitive skills a controlled, low-risk environment in which to begin to develop them. Indeed, this is in some measure what the “social technologies” of our previous education systems have already sought to do, and there is no reason why thoughtfully-designed software could not improve on them in certain use cases, forcing a student to practice a skill and solve problems at gradually increasing degrees of difficulty, like a good workout plan on an exercise bike.

Mobility Scooter for the Mind:

While education is one area where AI could function as an exercise bike, thus far it has been the domain where mobility scooters have run rampant. Whether you need to solve an equation, read a novel, or write an essay, there’s no reason anymore to sweat on it for more than a minute. Just ask ChatGPT, and it will do it for you. The temptation to prompt your way through high school or college, moreover, comes so naturally because so many young people have already fried their brains on Instagram and TikTok. We have spent nearly twenty years systematically depriving them of the ability to do anything that requires focused concentration, and then offered them a patent medicine that relieves them of the need to try. No wonder then that “mobility scooter” usage has quickly broadened out into other domains, as people use AI to ask a girl out on a date or to write a funeral eulogy. In a society of the mentally obese, a mobility scooter is going to be in high demand.

Automobile for the Mind:

Many uses of AI at this point are probably not quite so dystopian, but fall squarely in this middle category. People use AI to more efficiently get a task from Point A to Point B—not in order to save them the trouble of ever thinking or working again, but, ideally at least, so that they can free up more time and energy for “the important stuff.” This can be a reasonable trade-off and one I’ve made a number of times myself. For instance, although I’m not a data guy, my job involves a fair number of spreadsheets—membership databases, event registrants, email recipients. While I know how to edit a spreadsheet and even know how to run a few commands that might speed up the process, merging or collating spreadsheets can be quite tedious work, and isn’t the primary skill set I was hired for. So it makes plenty of sense to ask Claude to do it, now that the bots have gotten good enough to rarely need manual review. Of course, one must be careful; it might make sense to use my car to commute to work, but if so, I’d better set aside time for hiking (or biking!). Just so, if I’m going to use AI to replace certain secondary tasks, I’d better make sure I’m leaning into the use of my primary mental muscles all the more.

Motorcycle for the Mind:

This is an interesting one. Like many people over the past couple months, I’ve discovered the magic of “vibe-coding.” I never knew the first thing about coding, and have spent my entire adult life at the mercy of software engineers to make programs that did what I needed them to; usually they did, but when they didn’t, I would Google for a bit then give up in frustration. No longer. Now I can say, “hey, I wish there were a program/plugin that would do such-and-such,” and Claude will either tell me, “Well yes, there is” and walk me through a step-by-step of how to set it up, or Claude will say, “There isn’t, but I can build it for you!” Thanks to Claude, I now use an awesome app named Obsidian for all my research and writing, and I have a couple of custom plugins that Claude has created for it, as well as some larger products-in-progress. In the process, I’ve not merely been able to get from intellectual Point A to Point B faster, but I’ve actually acquired at least the rudiments of a new skill set. I actually know at least a tiny bit about coding and setting up software now, since I have to copy Claude’s code into Github, set up API keys, and the like. Claude is doing the work for me, but as it’s work I never knew how to do in the first place, I’m actually gaining some new agency and competence in the process.

Bicycle for the Mind:

Now, for the Holy Grail: can AI serve as a bicycle for the mind, leveraging your existing mental muscles so as to both make you more productive and make you stronger? It’s a tricky tightrope to walk, but in principle I think the answer is yes. Here again, I will share personal experience. Have you noticed how much more productive my writing output has been for the past few weeks here on this Substack? It’s almost as if I were using AI to help me! Well, I have—sorta. While I will never use AI to write so much as a sentence for me, or even to propose an outline, it can serve as a very effective mental sparring partner. I could be wrong, but I don’t think there is anything perverse about this, for it is in the nature of human thought to be dialogical. Think about it! You form your best and clearest thoughts when you are hashing things out in conversation with another person; or if not, when you are pretending to, perhaps by talking aloud to yourself or carrying on a mental dialogue. The greatest philosophers have developed their own private reflections in the form of dialogue, because we think best through such interactive point and counterpoint. I have thus experimented with coming to Claude with a set of half-formed hypotheses and intuitions (although not while writing this post, for what it’s worth), and saying, “Does this make sense? Are these ideas actually related or am I making something out of nothing?” The chatbot’s answers are not necessarily right—for my last essay, I ended up entirely disregarding Claude’s structural advice—but even having something to push back against helps me think harder and more clearly. Used carefully, it seems that AI can help you get further and get smarter in the process.

Do Not Go Gentle into that Good Night

That said, as things currently stand, there is little or nothing in the design of these products that encourages you to use them that way. To be sure, if you ask them, they will offer valuable suggestions; once, unnerved by just how much connecting-the-dots Claude seemed to be doing for me, I said, “No, it’s important that I do this myself; how do I avoid using you as a crutch?” and it proposed some very sound principles and habits. But I have an almost obsessive “do it myself” mentality; in college, I would refuse to read the Editor’s introduction to Kierkegaard or Kant, wanting instead to throw myself into the deep end and sink or swim; in grad school, I insisted on doing all my own footnotes when everyone else was using software. Most people, I suspect, are a bit more relaxed.

For the past year, there has been a raging debate over just how much AI was set to disrupt or destroy segments of the job market. Skeptics have always made a very powerful argument: however many tasks may be automatable, the most important parts of any job are those that can’t, those that require a human being to make a human decision between incommensurable goods, or require the “human touch” for customer service. Perhaps, but I’m not convinced. The question is not how many jobs AI can replace without losing something important, but how many we are willing to let it replace, if it saves enough effort. It seems quite clear that AI cannot fully replace a friend or lover; that hasn’t stopped many from being satisfied with ‘’close enough.” AI may not be able to write perfect code, but if it’s 80% as good with 20% the work, most coders seem happy handing off their unique competency to a machine.

A couple of weeks ago, there was a huge online sensation about the emergence of “Moltbook,” a Reddit-style “social media” site for AI agents. The AIs, it briefly looked, had become powerful enough that they really were plotting to rebel, sci-fi style. Soon it became clear that most of the craziest screenshots were fake, or prompted by human provocateurs who had hacked the agents. But this led to an equally troubling realization: many very smart, very competent people, it turned out, had handed off huge amounts of their private data and secure logins to extremely insecure open-source chatbots. Why would they do this? Because it was convenient.

It turns out that we are all much more willing to engage in self-erasure than the sage labor market analysts might believe. Rather than fighting hard to keep our jobs and demonstrate our uniquely human value-add, raging against the dying of the light, many of us seem to be not just going gently into that good night of human obsolescence, but tripping over one another in a race to be the first to fully outsource ourselves. The bicycle is still there to be ridden, but how many will mount it and head for the hills? That mobility scooter looks awfully inviting.