Learning and Authentic Learning in the Age of AI
Optimization and Orientation in Plato and Augustine
This is a guest post by Zachary Gartenberg, a tutor at St. John's College in Annapolis. If you want to pitch us an article, please send us a suggestion using this form.
Much of the contemporary discourse about AI’s effects on writing, education, and cognition has taken on a declarative tone. We are told what the future will bring, what students will lose, and what machines will replace. But this is not obviously a time for declarations.
It is a time to ask questions – carefully, seriously, and without presuming that we already know what is at stake. What is at risk is not just the authority of teachers or the fate of the humanities, but the intelligibility of learning as something more than adaptive performance.
Among the questions we must ask is whether something essential is being obscured. Not simply by AI’s practical disruptions, but by the model of learning that machine learning makes newly visible and dominant. The paradigm of optimization – of iterative improvement through feedback – has become a background assumption for how we design, measure, and describe learning, even in human contexts. But does this model capture what matters most in our deepest experiences of understanding? Is learning merely a refinement of performance? Or can it also be understood as a mode of orientation: a posture toward the not‑yet‑known, a practice shaped by inward responsibility, difficulty, and directedness toward something beyond utility or output?
These questions became palpable for me during my first year of teaching at St. John’s College, a small liberal arts institution with no departments or majors, where students pursue a common curriculum rooted in the close reading and discussion of foundational texts. The student essays I read were often hesitant, raw, or syntactically uneven. But their awkwardness reminded me that these were projects undertaken at the student’s initiative, for the purpose of working out, however provisionally, something that mattered. I never suspected a student of submitting AI‑generated work. But if someone had, I am not sure it would have mattered in the way most discussions assume. Whatever tools may have played a role, the essays bore the marks of being authored in a deeper sense. The signs of exploratory posture, interruptions of thought, and patterns of struggle that accompany the attempt to learn something for oneself were unmistakable.
Learning as Optimization
Machine learning systems embody a model of learning built from data, computation, and feedback. To take them seriously is to ask what exactly happens when a machine “learns.” At the center stands the artificial neural network: a layered structure of computational units loosely inspired by biological neurons. Such networks do not replicate human cognition, but they exploit a powerful analogy – intelligence arising from structure rather than introspection. Each unit applies weights and biases to its inputs, transforms them, and passes them forward. Learning consists of adjusting those internal parameters in response to error, a process governed by gradient descent where a model minimizes loss between predicted and correct values. One might picture it as a fog‑shrouded descent through a landscape of possibilities, nudging step by step toward lower error without ever seeing the whole. What emerges is a model not of conscious grasp but of refinement through feedback – an astonishingly successful simulation of learning, even as the question of whether it entails understanding remains open.
Deep learning systems echo something true in human experience. We often move from particulars to patterns, from fragments to form. Learning can feel like the slow tuning of attention until the shape of a domain becomes palpable. This picture is compelling in part because it scales: with more data and deeper models, performance improves. On this account, intelligence is not a flash of insight but an ongoing sensitivity to correction and constraint.
Yet as a cultural picture of human learning, optimization is not neutral. It has become a kind of default imagination across domains. In schools, adaptive platforms promise to optimize progress through algorithmically tailored tasks and just‑in‑time feedback. In self‑help culture, habit trackers and productivity apps invite us to engineer ourselves by data: optimize sleep, optimize attention, optimize emotional resilience. On social media, recommendation systems “learn” preferences by reinforcement to retrain attention. The shared promise is frictionlessness. Difficulty – confusion, false starts, frustration – comes to look like mere drag. With the right scaffold or loop, the struggle can be smoothed away – and, it is assumed, with no loss to learning itself.
But our deepest experiences of learning often resist that picture. Sometimes learning improves performance; sometimes it complicates it. A scientist recognizes the limitations of a cherished model. A friend hears a hard truth and must let a long‑held self‑image die. A student discovers that an inherited conviction cannot survive questioning. In these cases, learning is not optimization but reorientation – an interior turning that changes the one who learns. Such learning is not frictionless. It involves exposure to what one would rather not see, and a willingness to bear the costs of seeing it.
The contemporary mood oscillates between enthusiasm for optimization and unease before its implications. Consider the recent suggestion, in public essays and private conversations alike, that the humanities may survive AI not by defending their utility but by retreating into sensibility: a posture of witness, a cultivated vulnerability before loss. There is something admirable in refusing panic or nostalgia. Yet the mood leaves unsettled the very thing at stake. What, precisely, has been lost? Is it the outsourcing of tasks once done by scholars and students? Or something deeper – an erosion of our ability to say what learning is and why it matters? If the latter, then the task is not to mourn but to clarify.
The optimization model gives an answer to the conditions of learning. Specify the inputs, design the architecture, supply the feedback, and the system will learn. The humanist picture asks a different kind of question: how does one become the kind of agent for whom understanding is possible? What posture of inquiry is required? What relation must a learner bear to what is learned such that the learning is not merely an accumulation of correct outputs but a transformation of the knower?
Plato’s Meno: From True Belief to Knowledge
Plato’s Meno begins with a question about whether virtue can be taught. But the dialogue soon becomes a meditation on what learning entails. Its most famous puzzle – Meno’s paradox – puts the point sharply. If you know what you are looking for, you need not inquire. If you do not, you cannot: you would not recognize it even if you found it. The implication is that learning appears either trivial or impossible. Socrates answers, not with a definition, but with an enactment. Learning, he suggests, is not a transaction of facts but a posture of inquiry, a willingness to let questioning change you.
This becomes vivid in the exchange with Meno’s uneducated slave. Socrates poses a geometrical problem: how to construct a square with twice the area of a given square. The boy answers first by intuition: double the side. That, of course, yields a square four times as large. Under further questioning, with lines drawn and possibilities tested, the boy notices a new path: the diagonal of the original square, used as the side of a new square, yields the desired doubling. What matters here is not that the boy reaches the “right answer” without being told. It is that his earlier picture of the situation is exposed as inadequate, and that he learns to stand differently in relation to the problem – to look again, to be corrected, and to sustain the tension of not knowing long enough for a better grasp to emerge.
Plato calls this movement “recollection” – not because the slave recovers a previously stored fact, but because genuine learning feels like remembering something you somehow already knew how to see. Recollection names a regained capacity for recognition. The boy’s mind is awakened to the structure that was available all along but occluded by a hasty picture. The point is not mystical; it is phenomenological. Insight feels like remembering because it restores us to ourselves as knowers.
The dialogue ends with a different, equally important distinction. Socrates compares true beliefs to the marvellously lifelike statues of Daedalus: they are valuable, but they run away unless tied down. Knowledge, by contrast, is true belief “tied down by an account of the reason why” - aitias logismos. The difference is not primarily in accuracy but in ownership. A belief becomes knowledge when it is held non‑accidentally, when the knower can give a reason that renders its truth intelligible to herself. Tying down does not mean discovering a mechanical cause; it means understanding what makes the claim hold and being able to stand behind it.
This “tying down” is both epistemic and ethical. To know, in Plato’s sense, is to bear responsibility for what one asserts. Untethered true beliefs may guide us well enough, but their guidance is accidental; we do not possess them so much as they possess us. Knowledge is claimed rather than merely had. It requires the learner to become, in some sense, the cause of her belief – to be able to say not only that it is so, but why it must be so.
Placed alongside modern machine learning, the distinction presses uncomfortably. A neural network can converge on highly accurate predictions. It can be probed, sometimes, for post‑hoc explanations; and techniques in interpretability aim to recover reasons from models. But the system does not appear to own its outputs. It does not stand in a reason‑giving relation to what it says. Its correctness is not the achievement of an agent who can be addressed, challenged, and held to account. If Plato is right, then what machines do – even at superhuman levels of performance – remains at the level of true belief. That is not an insult. It is a way of marking the difference between prediction and understanding.
The pedagogical implications are equally sharp. If knowledge is tied‑down true belief, then the task of a teacher is not to deposit right answers but to cultivate a poise in the learner: the readiness to inquire well, to live with perplexity, to give and demand reasons. Socrates models this by asking questions that do not humiliate or coerce, but invite the student into responsibility. The goal is not speed of convergence but the formation of a knower for whom convergence, when it comes, is not accidental. In an optimization culture obsessed with outcomes, this is a reminder that the interior stance of the learner is part of what learning is.
Augustine’s The Teacher: Signs and Illumination
If Plato dramatizes learning as a public inquiry, Augustine turns our attention to the interior conditions that make understanding possible. The Teacher is framed as a dialogue with his son Adeodatus about whether teaching is even possible. Augustine’s starting claim is deliberately paradoxical: words do not teach; at most, they remind. Language offers signs, but the grasp of what those signs signify must occur within the learner. The teacher points; the student sees – or does not.
Consider Augustine’s small but telling example. Suppose I try to teach you what “walking” means by performing it. You watch; I walk faster to make the action clear. But you might then think walking is hurrying, and be misled. The demonstration is not self‑interpreting. Signs are always liable to confusion. This is not merely a problem of ambiguous words. It is a structural fact about communication: the meaning of a sign is not in the sign. It is in the act of recognition by which a mind sees what is meant.
From here Augustine draws a radical conclusion about the locus of learning. When we “learn from words,” we either already know what they signify – in which case we are being reminded – or we do not, in which case we cannot be taught by them alone. Understanding is an event of inward illumination. For Augustine theologically, the inner teacher is Christ – the Truth that instructs every mind from within. For readers who do not share his theology, the point can be expressed more generally: genuine learning involves an encounter with intelligibility that is not reducible to external transmission. There is a moment in understanding that cannot be handed over; it must occur as recognition.
This interiority is not privatism. Augustine is equally insistent that truth is public. When two people recognize the same truth, they do not do so by peering into one another’s minds, but by turning toward a reality available to both. Hence the striking claim elsewhere in the Confessions: if we both see that something is true, we see it not in me nor in you, but in a light above our minds. The language is metaphysical, but the experience is common. Think of the moment in a seminar when a difficult passage clicks for several readers at once. The recognition is intensely personal and unmistakably shared.
Placed alongside AI, Augustine’s account reframes the worry. Artificial systems generate signs – text, images, speech. They can be powerful engines of reminder, redescription, and juxtaposition. But they do not recognize truth. They do not awaken to meaning. The question, then, is not whether machines will replace teachers, but whether we will remember what teaching asks of us: to arrange signs so that recognition can occur, while never confusing the sign with what is signified. On this view, AI can serve learning – but only if the learner remains the agent of understanding. The danger is not that we will be fooled by synthetic language. It is that we will forget the difference between words and what they are for.
Augustine also gives us a richer account of memory than the mere storage of data. In his reflections, memory is the living archive in which our past is not simply kept but interpreted; it is the site where impressions become parts of a life. To learn authentically is to re‑pattern memory: to reorder what we recall and how it matters. Machines store and retrieve. They do not remember in this sense, because they are not selves for whom anything can matter. This is not an argument for human uniqueness so much as a reminder of the sort of being a learner is.
A Constellation Called “Authenticity”
But what is “authentic learning”? The phrase is awkward because it tempts a binary – real versus fake – precisely the polarity this inquiry tries to resist. Perhaps authenticity names not an essence but a family resemblance, a constellation of experiences that differ in form and context yet share certain features. Among these are orientation (the posture of inquiry that refuses to collapse the question prematurely), responsibility (the willingness to stand behind what one claims), and interior disruption (the way understanding dislodges us before it settles us). These are not outcomes but stances. They describe the relation a learner bears to what is learned and to the sources that make learning possible.
Seen in this light, the optimization model is not false but partial. It describes how a system can become more accurate at a task by iterative adjustment. It does not describe what it is like to come to know in such a way that the knowledge becomes yours. Plato’s language of tying down and Augustine’s language of illumination point to that interior achievement. They name a movement from correctness to possession, from output to understanding. The point is not to romanticize struggle or to reject efficiency. It is to insist that the struggle is often where responsibility is acquired, and that efficiency can be a vice when it insulates us from the risks on which insight depends.
If this is right, then the presence of powerful artificial systems is a double-edged sword. Precisely because machines can so fluently simulate certain performances, they force us to articulate what, in those performances, we care about. If a model can produce an essay indistinguishable in surface features from a competent student’s, then the work of teaching cannot be reduced to enforcing prohibitions or detecting anomalies. It must become the work that Socrates and Augustine model: cultivating poise, demanding reasons, arranging signs for recognition, and inviting students to tie down what they say.
These reflections are not merely theoretical. They suggest concrete practices for classrooms, institutions, and personal study:
Design for reasons, not just results. Ask for explanations, comparisons, and counterexamples that require aitias logismos – an account of the reason why. Invite students to identify where a claim could be wrong and what would change their minds.
Make room for productive difficulty. Build assignments that include stages of confusion and revision: interviews with primary texts, problem notebooks, seminar minutes that track the evolution of a question rather than its answer. Treat perplexity as part of the method.
Stage recognition. Use AI as a tool for juxtaposition – summaries to dispute, alternative framings to compare, synthetic outlines to reorganize. But let the decisive act be the student’s: the moment of saying what follows and why it matters.
Attend to voice. Encourage writing that shows the marks of ownership: hesitations that are earned, commitments that are justified, references that are integrated into a living question. Evaluate for the quality of relation to sources, not only for correctness or fluency.
Develop interpretive memory. Ask students to curate a portfolio of passages, problems, and questions that continue to inspire them, with periodic returns. The aim is to cultivate the Augustinian work of recollection, not the mechanical work of storage.
Cultivate shared seeing. In seminar, pause when something seems clear and ask: what exactly did we just see? Where do we see it in the text or the problem? How could it be otherwise? Make public the interior act of recognition.
These are more than just protections against technology; they are practices of freedom in the presence of it. They treat students as potential knowers capable of responsibility.
What We Keep in View
The optimization paradigm will continue to transform our tools and our institutions. It will also tempt us to take performance as the whole of learning. Plato and Augustine do not deliver a counter‑paradigm so much as they restore what a paradigm cannot hold: the posture of inquiry, the bearing of reasons, the interior act of recognition. They remind us that learning is not only a matter of getting things right, but of being transformed into the kind of person for whom getting them right is not accidental.
If “authenticity” remains useful, it is because it can name this quality of relation – a fidelity to difficulty, responsibility, and shared truth. Authentic learning is not a property of an output. It is a way a knower stands. Machines will continue to refine the outputs. They may even help us see more quickly what we could not see before. But they cannot relieve us of the work of tying down our beliefs, or of turning toward the light by which we come to see. That work remains ours, and it is, still, worth learning how to do.
Cosmos Institute is the Academy for Philosopher-Builders, technologists building AI for human flourishing. We run fellowships, fund fast prototypes, and host seminars with institutions like Oxford, Aspen Institute, and Liberty Fund.
What a great piece! thanks!
I'll admit that in struggling to focus while reading this piece I asked ChatGPT for help, which provided some surface level notes but left me with neither an authentic understanding of the material nor an appreciation for the time and attention required to offer my own insights.