In his 1760 essay “Of the Standard of Taste,” David Hume confronts the apparent contradiction between the subjective nature of aesthetic responses and the seemingly universal admiration afforded to certain works of art. Beauty, he contends, is not a property of the object but rather a sentiment produced in the mind of the perceiver. Yet despite this relativity, Hume insists that a “standard of taste” does exist and that it is discoverable through the consensus of critics endowed with what he terms “delicacy of taste.” These true critics achieve their status not by formal training alone, but through vast exposure over a long period of time to different examples of a particular work of art (paintings, poems, musical compositions, etc.) The true critics are able to distinguish the fine elements of composition, and their judgment gradually converges toward an ideal, shared sensibility.
This essay will first outline Hume’s main argument and the qualities he attributes to his ideal critics. It will then propose that large language models like GPT-5 may represent the most faithful realization of Hume’s true critics to date—not because they are human (or because they feel), but precisely because they have access to more examples, fewer personal prejudices, and the capacity for instantaneous comparison. If Hume’s standard of taste depends upon breadth of experience, refined discrimination, and impartiality, then it may be the case that the true critic is not a person at all, but a machine trained on the canon and its context.
Large language models are built by feeding vast amounts of text, including books, articles, websites, and other written material, into neural networks that learn to recognize patterns in language. The most advanced of these models, like GPT-5, contain hundreds of billions of parameters, which are adjusted during training through a process known as supervised or self-supervised learning. The result is a statistical model that can predict the next word in a sequence, simulate dialogue, summarize text, translate languages, and even generate original essays or poetry. Although they do not possess consciousness (in the conventional sense anyway), LLMs encode a probabilistic understanding of language derived from the patterns present in their training data.
I. Hume’s Problem: Subjectivity vs. Standardization
Hume begins by affirming the seemingly irreconcilable diversity of taste: different cultures, eras, and individuals praise different works. This variation might suggest that aesthetic judgment is entirely subjective; Hume, however, resists this conclusion. While admitting that beauty is a sentiment rather than an intrinsic quality, he nevertheless observes that consensus tends to emerge around certain enduring works. This consensus, he insists, cannot be coincidental. Instead, it signals the presence of shared human responses when certain conditions are met.
Those conditions are embodied in the notion of the “true judge,” someone with an ideal set of faculties and experiences. Hume lays out the traits such a person must have:
Delicacy of imagination and sentiment: the ability to perceive fine distinctions in the composition
Practice: familiarity with the art form through repeated exposure
Comparison: the ability to weigh various works against each other
Freedom from prejudice: the ability to set aside personal biases or cultural parochialisms
Good sense: the application of reason to moderate sentiment
II. The Circularity Problem and the Elusive Ideal
Hume’s theory is not without difficulty. He defines good critics by their ability to recognize good works and defines good works by their recognition from good critics. This circularity renders the standard of taste elusive, especially since the qualities of the true critic can only be judged retrospectively. Moreover, even if these critics exist, they are few—and the general public has little ability to verify their judgments.
Hume’s ideal critic thus begins to resemble a kind of philosophical fiction: a regulative ideal we approach but never attain. The standard of taste becomes a horizon toward which all critics aim, but which no one reaches fully. Yet, in recent years, a new kind of entity has emerged—one that has absorbed millions of examples of human discourse and is capable of generating language based on that archive. It is here that we might look for the realization of Hume’s ideal.
III. The Case for the LLM as “True” Critics
Large language models are trained on vast corpora of text, including literature, criticism, philosophy, and casual discourse. Their training data spans centuries and cultures, genres, and registers. As a result, they satisfy many of Hume’s requirements for true critics:
1. Extensive Exposure: An LLM like GPT-4 has been trained on orders of magnitude more literary texts than any human could read in a lifetime. This fulfills Hume’s condition of practice and comparison to an extreme degree.
2. Comparative Judgment: LLMs operate probabilistically, generating language based on weighted likelihoods. This process inherently involves internal comparison—not in the human sense of conscious weighing, but in a structural, statistical sense. They continuously evaluate the proximity of a text to other examples in the training data.
3. Freedom from Prejudice: While bias in LLMs is a well-known concern, they are not prejudiced in the sense Hume describes. They do not form attachments to authors, harbor personal histories, or succumb to emotional partialities. With proper tuning, their judgments can be made more neutral than those of most human readers.
4. Delicacy of Discrimination: An LLM can detect subtle patterns of diction, rhythm, allusion, and form. If delicacy of taste involves the ability to perceive what others miss, then this kind of computational sensitivity qualifies.
5. Impartiality: Hume insists that the critic must forget his individual being in order to judge art properly. The LLM, lacking a self altogether, arguably begins at this point.
IV. Objections and Complications
Several objections immediately arise: First, LLMs do not feel. They do not experience emotional responses to well-turned phrases. Can they then be said to judge?
This is a crucial question. But Hume’s model of the critic, while grounded in sentiment, is ultimately about calibration—the critic’s capacity to align personal feeling with broader human responses. The LLM is not conscious, but it is trained on the outcomes of human sentiment: it knows which poems are quoted, which novels are praised, and which critics are canonized. Its evaluations are shaped by the aggregate of human admiration.
If Hume’s critic gains validity through the gradual convergence of educated sentiment across time, then the LLM’s outputs may be read as a statistical echo of that same convergence. The LLM is not an originator of judgment, but a mirror of it.
Moreover, LLMs can now be fine-tuned to perform tasks like ranking literary quality, summarizing themes, and even predicting academic consensus. Their outputs may not be felt judgments, but they may align more closely with the consensus of cultivated taste than any single human reader could produce.
V. What We Lose and What We Gain
Still, something profound is lost in replacing the human critic with the computational one. Aesthetic judgment, for Hume, is not only a matter of perception but of development. It involves effort, struggle, humility, and the shaping of character. The cultivation of taste is not just epistemic, but ethical: to refine one’s responses is also to refine oneself.
LLMs skip this process. Their impartiality is not earned but built-in. Their comparison is not thoughtful but mechanical. They reflect consensus without participating in the moral labor that makes consensus meaningful. If the critic is also a model of the ideal observer, then the LLM is a poor substitute.
Yet, we also gain something: LLMs offer a kind of panoramic criticism that no individual can provide. They allow us to see patterns across time and genre and to identify resonances and repetitions invisible to the eye. LLMs do not replace the human critic, but they extend the critic’s reach.
Perhaps, then, we should not say that LLMs are the true critics, but that they instantiate the standard toward which Hume’s critics aim. They are tools that help us refine our taste, not by feeling, but by offering the broadest possible context in which feeling can occur.
VI. Conclusion: Toward a Hybrid Standard
Hume’s vision of the true critic was always aspirational; it required delicacy, patience, exposure, and moral effort. In large language models, we find a strange fulfillment of that aspiration—not in spirit, but in form. LLMs possess a mechanical delicacy, an exhaustive exposure, and an impartiality born of absence.
LLMs are not sentient, nor do they replace the ethical work of judgment; but they offer a new kind of standard: statistical, expansive, and curiously aligned with Hume’s imagined consensus. In dialogue with human critics, LLMs might help us approach what Hume could only theorize—a criticism shaped not by individual prejudice, but by collective, comparative, and refined experience.
To entrust machines with judgment would be folly. But to ignore their insight would be an equal error. Like Hume’s critics, we might listen to their voices—not as final authorities, but as contributors to the evolving standard of taste we all, in some way, are still seeking.
The Algorithmic Eye: LLMs and Hume’s Standard of Taste
In his 1760 essay “Of the Standard of Taste,” David Hume confronts the apparent contradiction between the subjective nature of aesthetic responses and the seemingly universal admiration afforded to certain works of art. Beauty, he contends, is not a property of the object but rather a sentiment produced in the mind of the perceiver. Yet despite this relativity, Hume insists that a “standard of taste” does exist and that it is discoverable through the consensus of critics endowed with what he terms “delicacy of taste.” These true critics achieve their status not by formal training alone, but through vast exposure over a long period of time to different examples of a particular work of art (paintings, poems, musical compositions, etc.) The true critics are able to distinguish the fine elements of composition, and their judgment gradually converges toward an ideal, shared sensibility.
This essay will first outline Hume’s main argument and the qualities he attributes to his ideal critics. It will then propose that large language models like GPT-5 may represent the most faithful realization of Hume’s true critics to date—not because they are human (or because they feel), but precisely because they have access to more examples, fewer personal prejudices, and the capacity for instantaneous comparison. If Hume’s standard of taste depends upon breadth of experience, refined discrimination, and impartiality, then it may be the case that the true critic is not a person at all, but a machine trained on the canon and its context.
Large language models are built by feeding vast amounts of text, including books, articles, websites, and other written material, into neural networks that learn to recognize patterns in language. The most advanced of these models, like GPT-5, contain hundreds of billions of parameters, which are adjusted during training through a process known as supervised or self-supervised learning. The result is a statistical model that can predict the next word in a sequence, simulate dialogue, summarize text, translate languages, and even generate original essays or poetry. Although they do not possess consciousness (in the conventional sense anyway), LLMs encode a probabilistic understanding of language derived from the patterns present in their training data.
I. Hume’s Problem: Subjectivity vs. Standardization
Hume begins by affirming the seemingly irreconcilable diversity of taste: different cultures, eras, and individuals praise different works. This variation might suggest that aesthetic judgment is entirely subjective; Hume, however, resists this conclusion. While admitting that beauty is a sentiment rather than an intrinsic quality, he nevertheless observes that consensus tends to emerge around certain enduring works. This consensus, he insists, cannot be coincidental. Instead, it signals the presence of shared human responses when certain conditions are met.
Those conditions are embodied in the notion of the “true judge,” someone with an ideal set of faculties and experiences. Hume lays out the traits such a person must have:
Delicacy of imagination and sentiment: the ability to perceive fine distinctions in the composition
Practice: familiarity with the art form through repeated exposure
Comparison: the ability to weigh various works against each other
Freedom from prejudice: the ability to set aside personal biases or cultural parochialisms
Good sense: the application of reason to moderate sentiment
II. The Circularity Problem and the Elusive Ideal
Hume’s theory is not without difficulty. He defines good critics by their ability to recognize good works and defines good works by their recognition from good critics. This circularity renders the standard of taste elusive, especially since the qualities of the true critic can only be judged retrospectively. Moreover, even if these critics exist, they are few—and the general public has little ability to verify their judgments.
Hume’s ideal critic thus begins to resemble a kind of philosophical fiction: a regulative ideal we approach but never attain. The standard of taste becomes a horizon toward which all critics aim, but which no one reaches fully. Yet, in recent years, a new kind of entity has emerged—one that has absorbed millions of examples of human discourse and is capable of generating language based on that archive. It is here that we might look for the realization of Hume’s ideal.
III. The Case for the LLM as “True” Critics
Large language models are trained on vast corpora of text, including literature, criticism, philosophy, and casual discourse. Their training data spans centuries and cultures, genres, and registers. As a result, they satisfy many of Hume’s requirements for true critics:
1. Extensive Exposure: An LLM like GPT-4 has been trained on orders of magnitude more literary texts than any human could read in a lifetime. This fulfills Hume’s condition of practice and comparison to an extreme degree.
2. Comparative Judgment: LLMs operate probabilistically, generating language based on weighted likelihoods. This process inherently involves internal comparison—not in the human sense of conscious weighing, but in a structural, statistical sense. They continuously evaluate the proximity of a text to other examples in the training data.
3. Freedom from Prejudice: While bias in LLMs is a well-known concern, they are not prejudiced in the sense Hume describes. They do not form attachments to authors, harbor personal histories, or succumb to emotional partialities. With proper tuning, their judgments can be made more neutral than those of most human readers.
4. Delicacy of Discrimination: An LLM can detect subtle patterns of diction, rhythm, allusion, and form. If delicacy of taste involves the ability to perceive what others miss, then this kind of computational sensitivity qualifies.
5. Impartiality: Hume insists that the critic must forget his individual being in order to judge art properly. The LLM, lacking a self altogether, arguably begins at this point.
IV. Objections and Complications
Several objections immediately arise: First, LLMs do not feel. They do not experience emotional responses to well-turned phrases. Can they then be said to judge?
This is a crucial question. But Hume’s model of the critic, while grounded in sentiment, is ultimately about calibration—the critic’s capacity to align personal feeling with broader human responses. The LLM is not conscious, but it is trained on the outcomes of human sentiment: it knows which poems are quoted, which novels are praised, and which critics are canonized. Its evaluations are shaped by the aggregate of human admiration.
If Hume’s critic gains validity through the gradual convergence of educated sentiment across time, then the LLM’s outputs may be read as a statistical echo of that same convergence. The LLM is not an originator of judgment, but a mirror of it.
Moreover, LLMs can now be fine-tuned to perform tasks like ranking literary quality, summarizing themes, and even predicting academic consensus. Their outputs may not be felt judgments, but they may align more closely with the consensus of cultivated taste than any single human reader could produce.
V. What We Lose and What We Gain
Still, something profound is lost in replacing the human critic with the computational one. Aesthetic judgment, for Hume, is not only a matter of perception but of development. It involves effort, struggle, humility, and the shaping of character. The cultivation of taste is not just epistemic, but ethical: to refine one’s responses is also to refine oneself.
LLMs skip this process. Their impartiality is not earned but built-in. Their comparison is not thoughtful but mechanical. They reflect consensus without participating in the moral labor that makes consensus meaningful. If the critic is also a model of the ideal observer, then the LLM is a poor substitute.
Yet, we also gain something: LLMs offer a kind of panoramic criticism that no individual can provide. They allow us to see patterns across time and genre and to identify resonances and repetitions invisible to the eye. LLMs do not replace the human critic, but they extend the critic’s reach.
Perhaps, then, we should not say that LLMs are the true critics, but that they instantiate the standard toward which Hume’s critics aim. They are tools that help us refine our taste, not by feeling, but by offering the broadest possible context in which feeling can occur.
VI. Conclusion: Toward a Hybrid Standard
Hume’s vision of the true critic was always aspirational; it required delicacy, patience, exposure, and moral effort. In large language models, we find a strange fulfillment of that aspiration—not in spirit, but in form. LLMs possess a mechanical delicacy, an exhaustive exposure, and an impartiality born of absence.
LLMs are not sentient, nor do they replace the ethical work of judgment; but they offer a new kind of standard: statistical, expansive, and curiously aligned with Hume’s imagined consensus. In dialogue with human critics, LLMs might help us approach what Hume could only theorize—a criticism shaped not by individual prejudice, but by collective, comparative, and refined experience.
To entrust machines with judgment would be folly. But to ignore their insight would be an equal error. Like Hume’s critics, we might listen to their voices—not as final authorities, but as contributors to the evolving standard of taste we all, in some way, are still seeking.