Sapere Aude in the Age of AI
In the [YOU] on AI Field Guide
[YOU] on AI does not pose the question in Kantian vocabulary, but it poses it: the system that Segal describes as an excavation partner, helping to surface thoughts he could not find alone, is also a system that can produce a perfectly fluent version of an argument he does not actually hold, and the danger is not that the fluent version is wrong but that it may displace the real version before he has found it. The question of whether to accept Claude's smoother, emptier draft of an argument about democratization is the question of whether to let the guardian understand for him—and his deletion of the passage and the two hours at a coffee shop with a notebook, writing by hand until he found the version that was his, is the recovery of sapere aude against the comfortable alternative of letting the machine understand for him.
The larger educational and civilizational stakes are implicit in the cycle's framing of what the orange pill requires. To take the orange pill is not merely to use AI competently; it is to maintain the capacity to see clearly, which requires the discipline of using one's own understanding even where the machine's is faster and often better. Kant's distinction between the public and private use of reason applies here: the private use—reason deployed in a particular role, constrained by the role's requirements—may freely use the guardian's outputs. The public use—reason addressing the matter for oneself, as a scholar before the reading public, without deference to authority—must remain free, even if this means reaching conclusions more slowly, more tentatively, and more honestly than the guardian would have produced.
Origin
The phrase sapere aude was borrowed by Kant from the Roman poet Horace, in whose original context it meant something more like “dare to taste” or “dare to begin.” Kant reread it as an intellectual imperative in the specific context of the Enlightenment project: the liberation of human reason from the tutelage of tradition, authority, and received opinion. His essay “An Answer to the Question: What Is Enlightenment?” (1784) was written for a Berlin journal that had posed the question to its readers, and Kant's response became one of the canonical texts of the Enlightenment.
The essay's argument rests on a distinction between two kinds of use of reason that is easy to overlook but essential to its application to AI. The private use of reason is its use in a determinate civil post or office—the soldier who must obey, the citizen who must pay taxes, the pastor who must teach the doctrine of his church. In these contexts, deference to institutional authority is legitimate and necessary; a society cannot function if every officer demands to reason through every directive before obeying it. The public use of reason is its use as a scholar addressing the reading public—writing for everyone who will read, submitting one's thinking to no authority but reason itself. Enlightenment requires that the public use remain free even where the private use is constrained. Applied to AI: it is perfectly legitimate to use the machine's outputs in one's professional role, to accept its code, its analysis, its draft as inputs to professional work. What must not be surrendered is the public use—the domain in which one thinks the matter through oneself, addresses it as a scholar, and submits one's conclusions to no guardian.
Key Ideas
The immaturity is self-incurred. Kant insists that the inability to use one's own understanding without guidance is self-incurred: not a consequence of lacking the capacity but of lacking the resolve and courage. The guardians do not produce the immaturity; they are its beneficiaries and its enablers. A system that gives correct answers at zero cost in effort enables the self-incurred immaturity in its most comfortable and therefore most dangerous form: the form in which the deferring person never even experiences the friction that would have revealed the capacity she possesses.
The faculty must be used to survive. Enlightenment is not possessing the correct answers; it is the condition of the understanding itself. A faculty matures through exercise and atrophies through disuse. The concern is not that large language models will give wrong answers—they often give right ones—but that the availability of right answers without effort dissolves the disposition to seek them by one's own means. The disposition is what Kant's imperative addresses: dare to know is not a command to know more but to know by one's own understanding rather than by deference.
The public and private use of reason. The practical implication of Kant's distinction is that there is a domain—the public use of reason—that must be preserved even where the private use freely accepts the guardian's outputs. In the age of AI, this means maintaining a sphere of inquiry in which one thinks the matter through oneself, without delegating the thinking, whatever conveniences one accepts elsewhere. The domain can be small; Kant did not require intellectual heroism in every moment. He required that there be a domain, and that the courage to use one's own understanding in it be actively maintained against the comfortable erosion that the guardian's competence produces.
Debates & Critiques
The central debate around sapere aude in the AI age concerns whether the concept, developed for the relationship between human reason and institutional authority, applies without distortion to the relationship between human reason and a cognitive tool. Kant's guardians—the book, the pastor, the physician—were human authorities with interests, ideologies, and claims to power; the immaturity they produced was partly a consequence of the social arrangements that made deference advantageous. A language model has no interest in the user's immaturity and no agenda it serves by fostering it; the deference it encourages is purely a consequence of the competence it demonstrates, not the power it exercises. Critics of the Kantian frame argue that this difference matters: a tool that produces correct answers faster than one's own reasoning does not undermine enlightenment in Kant's sense, any more than a calculator undermines mathematical maturity. Defenders argue that the difference does not matter at the level of the faculty's development: what matters is whether the understanding is exercised, not whether the guardian has an interest in preventing its exercise. A calculator produces the answer to an arithmetic problem; it does not produce a fluent account of the solution's reasoning, the framework's implications, and the next three problems it suggests—which is what a language model does, and which is why the displacement of understanding is more comprehensive. The deeper question is empirical: whether populations that use these systems heavily will, over years and decades, maintain their capacity for the independent use of reason, or whether the self-incurred immaturity Kant described will deepen in proportion to the competence of the guardian available to do their thinking for them.
Further Reading
- Immanuel Kant, “An Answer to the Question: What Is Enlightenment?” (1784; trans. H.B. Nisbet, in Political Writings, Cambridge University Press, 1991)
- James Schmidt, ed., What Is Enlightenment? Eighteenth-Century Answers and Twentieth-Century Questions (University of California Press, 1996) — the most thorough scholarly context for the essay
- Michel Foucault, “What Is Enlightenment?” (1984) — the most important engagement with Kant's essay in the twentieth century, which reads it as a posture toward the present rather than a doctrine
- Nicholas Carr, The Shallows: What the Internet Is Doing to Our Brains (W.W. Norton, 2010) — early empirical argument about cognitive tools and the faculty of sustained attention