There is a question that the age of artificial intelligence makes newly urgent, and it is not the question that most people are asking.
The question most people are asking is: what can AI do?
The answer expands monthly. AI can write, analyze, code, design, compose, translate, diagnose, predict, and generate. It can do most of what knowledge workers have been paid to do, faster and often better. The list of capabilities is long and growing. The question of what AI can do is, for practical purposes, becoming less interesting than the question of what it cannot.
The question that matters is: what remains?
Not what remains useful. Not what remains economically valuable, though that question has its own urgency. What remains irreducibly human — what cannot be extracted, trained on, optimized, or simulated by a system that learns from patterns in data.
The answer, when you look at it clearly, is smaller than most people find comfortable. It is not creativity — AI generates. It is not analysis — AI reasons. It is not even judgment in the broad sense — AI recommends, predicts, and decides within well-specified domains with increasing reliability.
What remains is taste.
Not preference — preference can be modeled. Not opinion — opinion can be generated. Taste, in the specific sense that distinguishes it from both: the accumulated, historically-specific, embodied capacity to recognize what is genuinely good from what merely resembles it. The ability to walk into a room and know. To read a paragraph and feel the difference between writing that is alive and writing that performs aliveness. To meet a person and understand, before any explicit evidence is available, whether they are someone worth knowing.
This capacity cannot be trained. It can only be developed — through the specific, irreplaceable process of living a particular life, accumulating particular experiences, making particular mistakes, and slowly, over years, building the calibrated attention that makes recognition possible.
What Taste Actually Is
Taste is not a preference. Preferences can be aggregated, analyzed, and modeled with reasonable accuracy from behavioral data. Amazon knows what you are likely to buy. Spotify knows what you are likely to listen to. Netflix knows what you are likely to watch next. These are preferences — patterns of choice that emerge from the intersection of your history and the available options, learnable from the record of your behavior.
Taste is what underlies preferences without being reducible to them. It is the evaluative capacity that allows you to recognize, in a work or a person or a space or a decision, something that exceeds or falls short of what your preferences alone would predict. It is the ability to be surprised by what is genuinely good and unsurprised by what is merely popular. It is the capacity to say, with the specific confidence that cannot be fully justified but is rarely wrong: this is the real thing.
The person with genuine taste in literature can distinguish between the sentence that sounds like good writing and the sentence that is good writing — between the surface performance of the qualities that good prose has and the actual presence of those qualities. This distinction cannot be learned from examples alone. It cannot be acquired by reading descriptions of what good writing is. It develops through the slow accumulation of reading — through the specific, bodily experience of encountering language that is alive, and the equally specific experience of encountering language that imitates aliveness without achieving it, until the nervous system has been calibrated, through thousands of such encounters, to feel the difference before the mind has articulated it.
This is what AI cannot have.
Not because AI cannot process language — it can process language with extraordinary sophistication. Not because AI cannot generate text that reads as good writing — it can, increasingly, generate text that passes many of the surface tests of quality. But because the capacity to recognize the difference between the real thing and its simulation requires having encountered both, repeatedly, with the specific attention that only a person with skin in the game can bring.
AI has processed more text than any human being will ever read. But it has not read any of it. It has not been moved by it, bored by it, changed by it, or left cold by it. It has not had the experience of encountering a sentence that made something shift. It has not had the opposite experience — the specific tedium of reading something that performs the motions of significance without generating any. The pattern recognition that AI performs is not the same as the taste that humans develop. Pattern recognition is what taste looks like from the outside. From the inside, taste is something else entirely.
The Accumulation
Taste develops through accumulation. And the accumulation is not primarily accumulation of information.
It is accumulation of experience — the specific, embodied, historically-situated encounters with the world that leave a residue in the nervous system. The books read at specific ages, in specific moods, in specific places, that lodged something in the way language feels when it is right. The spaces encountered before and after knowing what makes a space good, and the slow development of the sensibility that can now walk into a room and know in the first thirty seconds whether it is working or not. The relationships that failed in specific ways, and what those failures taught about what is actually present in a person versus what is performed.
This accumulation cannot be shortcut. It cannot be downloaded. It cannot be acquired by reading about it, talking about it, or training on data about it. It develops through the specific, time-consuming, often uncomfortable process of being a particular person in a particular life, paying attention to what actually matters, making mistakes, and slowly becoming someone whose attention is calibrated by experience rather than instruction.
This is why taste correlates with age in a specific way. Not with age as such — there are young people with extraordinary taste and old people with none. But with the specific kind of age that comes from having lived deliberately, paid attention to what you were encountering, and allowed experience to change the calibration of your attention rather than simply accumulating as stored information.
The person of genuine taste has been changed by what they have encountered. Their nervous system is different from what it would have been if they had encountered different things. The taste is not a set of opinions they hold. It is a property of who they have become.
Why AI Cannot Simulate This
The simulation problem is not a technical problem. It is a structural one.
AI systems learn from patterns in data. They become very good at recognizing the patterns that correlate with human evaluations of quality — the features that tend to appear in things that people with taste judge as good. They can generate outputs that match these patterns with increasing fidelity. They can, in many cases, produce work that passes the surface tests of quality well enough to deceive.
But the simulation of taste is not taste. And the person of genuine taste can, in most cases, still tell the difference — not always immediately, and not always with articulable reasons, but with a reliability that the simulation cannot quite achieve. The difference is subtle. It is in the sentence that almost works but doesn’t quite. The design that has all the right elements but doesn’t cohere. The argument that is technically correct but misses what actually matters. The performance that hits all the notes but generates no feeling.
These are the failures of simulation. They are invisible to systems that learn from patterns, because the pattern is present. They are visible to taste, because taste is not pattern recognition. It is the recognition of the real thing — which is different from the recognition of patterns that correlate with the real thing.
This gap will not close entirely. It will narrow, in some domains, as AI becomes more sophisticated. But the gap between pattern recognition and taste is structural, not technical. It is the difference between a system that has learned what good things look like from the outside and a person who knows what they feel like from the inside. These are not the same thing. They cannot be made the same thing. The inside knowledge requires an inside — a perspective, a history, a nervous system changed by experience.
Taste as the Last Advantage
The economic and professional implications of this are significant.
As AI increasingly automates the production of competent output across more and more domains — competent writing, competent analysis, competent design, competent code — the value of competence relative to taste will shift. Competence is the baseline. AI can produce it reliably, cheaply, and at scale. Taste is the differentiator. AI cannot produce it, because it cannot have it.
The person who can tell the difference between the AI output that is good enough and the AI output that is genuinely excellent — and who can produce, direct, or curate toward the latter — has the one capability that the automation of competence makes more valuable, not less. They are the editor who knows which sentences to cut. The designer who knows which version actually works. The investor who knows which company is the real thing. The leader who knows which person to trust.
These are all instances of taste in operation. They are exercises of the specific, accumulated, embodied evaluative capacity that develops through a life lived with attention, that cannot be trained on data, and that becomes more valuable as the competence it evaluates becomes cheaper to produce.
The age of AI does not eliminate the advantage of taste. It clarifies it. When competent output is abundant and cheap, the scarcity is in the judgment that can distinguish the excellent from the merely competent. That judgment is taste. And taste is yours in a way that nothing AI produces ever will be.
How Taste Is Developed — And Why It Is Being Threatened
The development of taste requires specific conditions that the current information environment is systematically undermining.
Taste develops through sustained, attentive encounter with things that are genuinely good — and the equally sustained encounter with things that are not, so that the nervous system can be calibrated by the comparison. It requires time — the time to read slowly, to look carefully, to sit with an experience long enough for it to leave a mark. It requires the friction of difficulty — the experience of encountering something that resists easy understanding and rewards the effort of staying with it.
The current information environment optimizes for the opposite of all of these conditions. It delivers content at a rate that precludes the sustained attention that taste development requires. It removes friction — making access to every option equally immediate, which removes the comparative encounters through which calibration develops. It optimizes for engagement rather than quality — which means it systematically surfaces the things that capture attention most efficiently, which are not necessarily the things that develop taste most effectively.
The child who grows up primarily in the algorithmically optimized information environment — whose cultural diet is determined largely by recommendation systems designed to maximize engagement — is a child whose taste development is being systematically impeded. Not because the content is bad, necessarily. But because the conditions required for taste development — sustained attention, comparative encounter, the friction of difficulty, the time required for experience to leave a residue — are absent from the experience of consumption that algorithmic optimization produces.
This is the threat to taste that the age of AI represents. Not that AI will develop taste of its own — it cannot. But that the conditions required to develop taste in humans are being eroded by the systems that optimize for engagement rather than depth, for immediate reward rather than the slow accumulation that taste requires.
What This Means for a Long Life
The cultivation of taste is a long-game investment. It pays returns across a lifetime in ways that are difficult to quantify but impossible to ignore.
The person of genuine taste makes better decisions — not because they have access to better information, but because their evaluative capacity is calibrated by experience in a way that allows them to recognize what actually matters in a situation. They hire better, choose better, create better, and navigate better — because taste is the capacity to recognize the real thing, and the real thing is what matters in every domain.
The cultivation of taste requires the specific behaviors that the current information environment makes more difficult and therefore more valuable: reading slowly, engaging with difficulty, choosing depth over breadth, allowing time for experience to accumulate and calibrate rather than consuming at the rate that maximizes the quantity of input.
These are not sacrifices. They are investments in the one capacity that AI cannot replicate, the current environment is eroding, and the coming decades will need most urgently.
Taste is what remains when everything that can be automated has been automated.
It is what remains yours.
For those who intend to last.
Frequently Asked Questions
What is the difference between taste and preference?
Preference is a pattern in behavior — what you tend to choose from the available options, learnable from the record of your choices. Taste is the evaluative capacity that underlies preference without being reducible to it. It is the ability to recognize what is genuinely good — to distinguish the real thing from its simulation — based on the accumulated, embodied calibration that develops through sustained encounter with both. AI can model preferences. It cannot have taste, because taste requires having been changed by experience in a way that data processing does not produce.
Why can’t AI develop taste?
Taste develops through the specific, embodied, historically-situated encounters with the world that leave a residue in the nervous system — encounters that change the person having them. AI processes data. It has not been moved by a sentence, bored by a performance, or changed by an experience. The pattern recognition it performs is not the same as taste. Taste is not the recognition of patterns that correlate with quality. It is the recognition of quality itself — which requires an inside perspective that data processing cannot provide.
Is taste the same as expertise?
They overlap but are not identical. Expertise is domain-specific competence — the ability to perform at a high level within a defined area. Taste is the evaluative capacity to distinguish the excellent from the merely competent within that domain — and often across domains. Many experts lack taste; many people with taste lack technical expertise. The combination of both is rare and extraordinarily valuable.
How is taste developed?
Through sustained, attentive encounter with things that are genuinely good, paired with the equally sustained encounter with things that are not — so that the nervous system is calibrated by comparison. This requires time, attention, the friction of difficulty, and the willingness to allow experience to change your evaluative capacity rather than simply accumulating as stored information. It cannot be shortcut, downloaded, or acquired by reading descriptions of what good taste is.
What threatens the development of taste in the current environment?
The algorithmically optimized information environment systematically undermines the conditions taste development requires. It delivers content at a rate that precludes sustained attention. It removes friction by making all options equally accessible. It optimizes for engagement rather than quality. And it surfaces what captures attention most efficiently, which is not necessarily what develops taste most effectively. The development of taste requires deliberate resistance to these conditions — the choice to read slowly, engage with difficulty, and allow experience to accumulate over time.
SIGNAL tracks the recurring patterns of human experience across history, philosophy, and science — for people living long enough to encounter them more than once.
For those who intend to last.