Is It Beautiful?

The Right Wrong Question About Language

English became the world’s default language by accident. Germanic roots, a Norman French overlay, the Great Vowel Shift, spellings that froze centuries before pronunciation finished moving. The result is a language where you can pronounce a word fluently and still have no idea how to spell it, which is why games like Spelling Bee exist and why the rest of the world pays a tax every time it learns the language.

With AI in the loop, the instinct to fix this gets sharper. If English is now the programming language of the AI era, why not redesign it? Strip out the irregularities. Make it logical. Make it learnable. Esperanto already proved the basic idea works: regular conjugations, phonetic spelling, agglutinative word-building, no exceptions. Lojban pushes the same instinct further, with a grammar based on predicate logic and sentences that parse exactly one way.

The trouble is that the redesign keeps running into a wall that has nothing to do with language design. Esperanto solved the entry problem brilliantly. It did not solve the fluency problem. The basics are easy compared to any natural language, and yet becoming truly fluent requires the same accumulated idiom, register, collocation, and cultural reference that every natural language demands. There is a word in Esperanto, krokodili, meaning to speak your native language at an Esperanto meetup. You cannot derive it from rules. You have to know it. The memorization tax returns at the top of the curve.

Lojban is worse, in a way. It is so unambiguous, so alien, that it never crossed the threshold from designed system to living medium. The very features that made it logical made it unable to carry the things people actually want languages to carry.

The deeper recognition, the one that takes a while to arrive at, is that the irregularities are not the problem. They are part of what makes language an artistic medium at all. A rational language strips out exactly the surface texture, the implication, the room for play that lets language become more than a transmission protocol. The original question, “how do we fix the medium?”, turns out to be the wrong question. But it is the right wrong question. It opens onto a better one, which is: what is the medium actually for?

Two Things We Were Calling Intelligence

The language question opens onto an intelligence question. For about forty years, “knows how to code” worked as a near-perfect proxy for “is smart.” It selected for logical thinking, patience, abstraction, attention to detail, willingness to sit with hard problems until they yielded. Parents wanted their kids to learn it. Liberal arts majors felt vaguely guilty for not knowing it. Programmers became the cultural archetype of the smart person.

The proxy held because the medium was unforgiving enough that you could not fake it. The compiler did not care how articulate you were. Either the program ran or it did not. That filtering effect made code a reliable signal of a certain kind of cognitive discipline.

LLMs broke the proxy. Not because programmers stopped being smart, but because the test stopped being discriminating. The discipline that code used to require is now partially externalized. You can produce working software without the patience that previously gated entry. So the signal stops working, and we have to look more carefully at what was actually being signaled.

What was bundled in the cultural notion of “smart programmer” turns out to be at least two different things. One is what we might call tactical intelligence: the ability to hold precise structures in your head, decompose problems into operations a machine can execute, debug by reasoning about state. It is real, and it is highly transferable to machines because it operates in a domain machines were built for. The other is something different, which we might call generative intelligence: the ability to ask the question behind the question, to notice when an analogy is doing too much work, to hold competing frames simultaneously, to know which direction in idea-space is worth pushing. LLMs can amplify this. They cannot supply it. If you do not bring a real question, you get a polished non-answer.

These were always different intelligences. The culture conflated them because programmers were visible, well-paid, and articulate about their own work. The people doing the deeper generative work, historians making non-obvious connections, essayists noticing patterns, philosophers asking foundational questions, were less visible and usually less rewarded. The market priced tactical intelligence highly because it was scarce and economically productive. Generative intelligence was scarce but not as legibly productive.

AI inverts this. Tactical intelligence becomes abundant. The bottleneck shifts to whoever knows what is worth building, what question is worth asking, what frame is worth questioning. The vibe coder who produces an app from a prompt is using a tool that requires generative intelligence to use well. The tool does not supply taste, judgment, or the sense of what matters.

The Hacker, the Researcher, and the Missing Layer

The cybercriminal is a sharp test case. A skilled black-hat hacker has formidable tactical intelligence: creativity in finding attack surfaces, patience for slow reconnaissance, a clear-eyed read on human and system weaknesses. Some of them have generative intelligence on top of that, the kind that sees an architectural blind spot nobody else has noticed, or that reframes a class of system in a useful new way. We still hesitate to call them wise. We even hesitate, on reflection, to call them fully intelligent in the older sense of the word.

That hesitation is information. It reveals that our concept of intelligence has thinned over time. The word used to mean something closer to “perceives reality clearly and acts well in it.” Somewhere along the way we narrowed it to “executes complex cognitive operations.” A hacker without conscience executes complex operations. In the older sense, though, they are missing the part of intelligence that sees other people as real. That is not a separate moral failing layered onto an otherwise complete intelligence. It is a failure of intelligence itself, properly understood.

Now turn the same lens on AI researchers, the people we currently treat as geniuses. Someone like Andrej Karpathy is widely admired, and when he says that LLM-generated code is not yet beautiful, the comment lands. But what is he actually doing when he says it?

He is making an aesthetic claim. Beauty in code is a real thing, and it usually means something specific: the solution reveals the structure of the problem rather than obscuring it. Beautiful code is short not because shortness is virtuous but because the author saw the problem clearly enough to remove what was not essential. That kind of seeing is closer to what good essayists do when they cut a paragraph that almost worked. It is taste. Taste is not tactical.

Three layers come into view once you look carefully. There is tactical execution, which is making the machine do the thing. There is tactical taste, which is making the machine do the thing in a way that reveals rather than obscures. And there is wisdom, which is the capacity to tell what is worth doing at all, including across moral and human dimensions. The hacker has the first, sometimes the second, almost never the third. The mediocre researcher who happens to be well-positioned in the industry often has the first and gets credit they have not earned for the other two. The genuinely good researcher has all three. The culture rewards the first, occasionally celebrates the second, and rarely notices the third.

The AI industry has a values vacuum that its tactical brilliance is poorly equipped to fill. The same field that produces Karpathy-quality teaching also produces people who genuinely believe that having built impressive things gives them authority on questions of meaning, ethics, governance, and human flourishing. It usually does not. Being good at making a transformer learn faster does not make you good at thinking about what humans should do with their attention.

The Dimension That Doesn’t Care What You’re Doing

What the hacker is missing, what the merely tactical researcher is missing, what the cultural model of intelligence has been missing, is something the Greeks already had words for. They distinguished technē, which was skill or craft, the reliable know-how of producing something, from a higher register that sat closer to what Aristotle called phronēsis, practical wisdom, the judgment of what to do in the particular case. Technē makes the thing. The higher register makes the thing and means it.

This higher register is what we usually call art, though the word has been worn thin by overuse. It is not art as a profession. It is not art as a product hanging on a wall. It is a dimension available within any activity, the layer at which what you are doing becomes capable of bearing meaning.

The clearest examples are ordinary ones. Consider the waiter at a quiet restaurant on a weekday night who notices that the elderly couple at table six are celebrating something privately, an anniversary perhaps, and adjusts the pace of the meal without being asked. Nothing in the job description covers that. No training program reliably produces it. Everyone who has been served by such a person knows the difference, and the person doing it knows they are doing something that exceeds the transaction. The food is the same food. The service has become something else.

The Japanese shokunin tradition names this directly. The sushi chef who has made the same cuts for forty years is not doing it for efficiency. The repetition is the medium through which something else gets cultivated. Western culture has analogues, in the monastic traditions, in certain strands of Quaker thought, in the slow-craft movement, but they sit at the margins of an economy that mostly wants to optimize the technē layer and ignore the rest.

This dimension is more democratic than wisdom. Wisdom, in the classical sense, tends to attach itself to elevated activities: philosophy, governance, contemplation. The artistic dimension attaches to anything. It is available to the line cook, the bus driver, the person folding laundry. It does not require the activity to be important. It requires the attention to be real.

There is a relationship here to ethics that the cybercriminal example was reaching toward. Iris Murdoch argued that the moral life is mostly a matter of attention, of seeing what is actually there rather than what your ego projects. The artistic dimension and the ethical dimension converge at the level of attention. The cybercriminal’s failure is not really a failure of ethics imposed from outside. It is a failure to see other people as real, which is also, in this sense, a failure of art. The waiter who notices the elderly couple is not just exhibiting taste. They are exhibiting a kind of seeing other humans as real that is the root of ethics. The two dimensions, art and ethics, turn out to share a substrate.

Available Now, Not Later

A misconception is easy to acquire from the previous section, and it is worth pulling out explicitly. The artistic dimension can sound like a reward for mastery, something you graduate into after ten thousand hours of practice. It is not.

The beginner cook who genuinely attends to what they are doing is already in the dimension, even if their food is not yet very good. The student programmer who pauses to ask whether their solution reveals or obscures the problem is in the dimension, even if their code still has bugs. Mastery is the slow accumulation of skill. Art is the quality of relationship to whatever skill you currently have. They are separable.

This matters more than it first appears. It means the artistic dimension is not reserved for the Karpathys or the great chefs. It is available to anyone, at any stage, in any task, the moment they bring real presence to it. What accumulates over a lifetime is not access to the dimension. It is depth within it.

The practical implication is one that any reader can act on directly. You do not need to wait until you are good at something to find meaning in it. You do not need credentials, recognition, or audience. You need the question, asked honestly while you are doing whatever you are doing: am I actually here, or am I just executing? The first time you ask the question in the middle of an ordinary task, something shifts. The task does not change. Your relationship to it does. That is the dimension opening.

Attention Is All You Need

The architectural insight that powers modern LLMs is named, almost too neatly, by the title of the 2017 paper that introduced the Transformer. Attention Is All You Need. In the technical context, attention is a mechanism for weighting which parts of the input matter when processing each part of the output. The authors picked the word because, even in the machine learning context, it gestured at something humans recognize from their own experience: that focused selective awareness is what makes meaning out of noise.

The coincidence runs deeper than wordplay. Simone Weil, writing decades before any of this, called attention “the rarest and purest form of generosity.” The contemplative traditions across cultures have circled the same recognition for millennia. The Transformer needs attention to make coherent output from a sea of tokens. The human needs attention to make a meaningful life from a sea of moments. The same word names both because the same principle operates at both levels.

This reframes the AI question in a way that defuses most of the anxiety around it. The dominant version of the question has been “what can humans still do that machines cannot?” That framing already concedes that meaning comes from output, from being the one who can perform the task. If meaning comes instead from the quality of attention brought to whatever you are doing, the question dissolves.

What AI actually threatens is not meaning. It threatens the economic arrangement that paid people to do certain kinds of work. That is a real problem, a serious one, and it deserves serious attention in its own right. It is a political and economic problem, though, not a metaphysical one. Conflating the two has made the conversation about AI more confused than it needs to be. People feel that something deep is being taken from them, and they reach for arguments about consciousness, creativity, or irreplaceability to defend it. The deeper thing is not actually under threat. The shallower thing, the paycheck, is, and it deserves its own honest defense.

The tools amplify whatever you bring to them. If you bring attention, they extend what attention can reach. If you bring distraction, they extend the distraction. The same instrument, in different hands, produces different lives.

Is It Beautiful?

The question that compresses everything is the simplest one. Whatever you are doing, in whatever medium, with whatever level of skill, the question worth asking is: is it beautiful?

The word has to be heard properly. Not beautiful as in pleasing to look at. Not beautiful as a judgment passed on finished work by an outside observer. Beautiful as something that emerges from the inside of the activity, a quality that appears when attention meets medium in the right way, and that the person doing the work recognizes before anyone else can.

It is a different test than the ones we usually apply. We ask: is it useful, is it correct, is it impressive, is it successful, is it valuable. Those are all external measures, and they all admit of comparison and ranking. Whether something is beautiful, in this sense, does not work that way. It cannot be benchmarked. It can only be perceived from inside the act by someone present enough to perceive it.

Which makes it the truest measure. It is the one measure that cannot be faked, outsourced, or optimized for. Either you find it beautiful or you do not, and only you can know. A machine can produce technically superior output in every domain you care about, and the question of whether your own engagement with the work has the quality of beauty remains entirely yours. The question is not threatened by capability. It is only threatened by your own inattention, or by an economy that does not give you the space to attend.

This is why the question survives every technological shift. The tools change. The economic arrangements come and go. The thing the tools are tools for, the life lived from inside one’s own activity with real presence to it, was never about the tools. It was about the person. People have been asking this question, or living it without naming it, in every culture and every era, in every kind of work, for as long as there have been people. AI will compress more and more of the technē layer. What it cannot compress is the layer where the question lives.

So the answer to where we started, the question about a more rational language, turns out to fold into something larger. We do not need a more rational language. We have languages enough. We do not need a more efficient kind of intelligence. We have intelligence enough. What we need, and what has always been the harder thing, is the attention that turns whatever we are doing into something we can stand inside and recognize as our own.

If you can ask the question honestly while you are doing the thing, you are already living the answer. If yes, the life you are inside of, whatever its surface, is already a blessed one. If no, that is information too, about the activity, or about your relationship to it, or about whether you are really there.

The question outlasts the tools. It outlasts the language. It outlasts the era. It will be asked by someone, somewhere, today, while making something ordinary with their hands or their words or their attention. And when they ask it, and when the answer comes back yes, they will know what every person who has ever lived well has known: that this, exactly this, is what a meaningful life feels like from the inside.

Photo by Kate Townsend on Unsplash