The Texture of Memory
On how insights accumulate, how the past inhabits the present, and what AI can (and cannot) remember for us.
When you grasp something, really grasp it, not merely hear or register it… wait, isn’t that a funny concept to try to bring across? What are tests for? What are interviews for? What do you know/understand/grok? Have you internalized something? Has something shifted in you that cannot be undone?
Remember the moment you first understood why the night sky is dark, or why a thrown ball arcs rather than flies straight, or why that friend acted the way she did all those years?
That shift, once it occurs, tends to remain. What was laborious becomes easy, what was opaque becomes obvious. Your mind is not the same mind it was a moment before; it now carries within itself a small piece of structured understanding that will color everything that follows.
This is the curious fact about insight: it passes into the habitual texture of one’s mind. All of us who have spent time attending to the movements of intelligence might put it this way: “Once one has understood, one has crossed a divide. What a moment ago was an insoluble problem now becomes incredibly simple and obvious.” The divide cannot be uncrossed. You cannot un-understand the punch line of a joke or the proof of a theorem. The insight remains, woven into the fabric of who you are and how you see.
Memory, in its deepest sense, is not the passive storage of sensory snapshots. It is the accumulation of such crossings—a vast, layered, living architecture of insights that have settled into your being. Each new understanding complements and combines with what came before, modifying the whole structure in ways both subtle and profound. The child who has learned to count does not simply possess a new fact; she now inhabits a different cognitive world, one in which new realities can be tracked, compared, and discovered. And that world will expand again when she grasps multiplication, and again when algebra opens a higher viewpoint, and again when calculus reveals the continuous behind the discrete. Each stage presupposes and transforms the last.
This is what we could call the self-correcting process of learning. Insights do not arrive fully formed and final. Each one reveals, by its very exercise, its own incompleteness. You act on what you understand, or speak it aloud, or think it through—and the gaps show themselves. A question arises that the insight cannot answer. A situation arises in which the insight does not quite fit. These failures are invitations: they prompt further questions that lead to complementary insights, which, in turn, reveal new gaps, which prompt new questions, in an ever-spiraling ascent. Memory is what holds this spiral together. Without the accumulation of past insights, there would be no platform from which to launch the next inquiry, no context in which a new insight could be recognized as progress or error.
Consider what this means for ordinary life. You walk into a meeting, and within moments you grasp the mood of the room—tense, perhaps, or expectant. You do not deduce this from explicit premises. You simply “see” it, as immediately as you see the color of the walls. But that immediate seeing is possible only because you have, over years, accumulated a vast store of insights into human faces, tones of voice, postures, and social situations. Those insights are not consciously recalled; they are present and operative, but operate behind the scenes. They govern the direction of your attention, evaluate the significance of small cues, and guide your own responses. The past is not behind you; it is within you, shaping the very texture of your present perception.
We might speak of a habitual orientation—a perpetual alertness formed by past inquiry. The physicist sees the world differently from the poet, not because their eyes are different, but because each has spent years asking different questions and accumulating different constellations of insights. What each notices, what leaps out as significant, what recedes into background noise—all this is determined by the accumulated structure of understanding. The habitual orientation is a kind of perceptual stance, an angle from which reality is approached. It is memory present to spring into readiness.
This is why learning is not merely the adding of facts to a mental warehouse. It is a transformation of the self who perceives and understands. Each insight, as it enters the habitual texture of the mind, changes what will count as data for future insights. The trained physician observes a slight asymmetry in a patient’s gait and suspects a neurological condition; the untrained observer sees only a person walking. The data are the same; the habitually structured minds are different. This is also why learning has direction and a strategy. One cannot simply leap to higher viewpoints; one must pass through the lower ones, accumulating the insights that will serve as the images for the next level of understanding. The acorn and the oak are both alive, but there are vast differences in what they can do. Similarly, the child and the master both think, but the master’s thinking has a reach and a suppleness that presuppose decades of accumulated insight.
There is, however, a shadow side to this architecture of memory. Not all that we have understood is fully integrated. Not all past insights are mutually compatible or consistent with our current self-image. Some experiences and some recognitions are too threatening to be admitted to consciousness and are therefore repressed, pushed below the threshold of awareness. What results is not the absence of memory but its distortion. The repressed insight continues to exert influence, manifesting in slips of the tongue, in odd aversions, and in dreams charged with displaced affect. Lonergan, drawing on Freud, speaks of screening memories—vivid but fictitious recollections that cover over actions too disturbing to recall directly. We remember a false scene precisely because the true one would demand an understanding of ourselves that we are not yet prepared to accept.
This is the censor at work: the preconscious mechanism that selects which materials from neural processes will be allowed to emerge into consciousness. In healthy development, this selection serves the dramatic pattern of experience—the ongoing project of making one’s life a coherent, livable narrative. But when the censor operates under the pressure of fear or shame, it can block not just painful memories but the very insights that would allow one to understand oneself more fully. The result is what Lonergan calls scotosis: a blind spot, an inability to see what is plainly there. Memory, in this pathological mode, becomes not the servant of understanding but its obstacle. The past persists, but in distorted form, shaping the present in ways the conscious mind cannot recognize or correct.
Conversely, a healthy memory supports the self-correcting process. It allows past insights to be recalled, reexamined, and revised in light of new experience. It does not cling to its own constructions but remains open to the further questions that would transform them. Such a memory is porous—willing to let go of what no longer fits, willing to integrate what once seemed threatening. This porosity requires courage, the willingness to face one’s own errors and blind spots. But it is also the condition for growth. The mind that cannot revise its habitual orientations is a mind that has stopped learning, a mind whose spiral of insight has frozen into a fixed picture.
Memory is also crucially social. We are born into a community that possesses a common fund of tested answers—language, custom, technique, story. From that fund, we draw our initial stock of insights, measured by our capacity, our interests, and our energy. The self-correcting process of learning that unfolds in each individual consciousness is simultaneously a communal development, effected through speech and example, disseminated, tested, and improved across generations. What one person discovers passes into the possession of many, to be checked against their experience and confronted with their further questions. The achievements of each generation serve as the starting point for the next.
This social memory is a mixed blessing. On one hand, it accelerates learning enormously: what took Archimedes years to discover can be taught to a schoolchild in an afternoon. On the other hand, shared biases and shared blind spots can be transmitted as easily as genuine insights. Each tribe and nation, each group and class, tends to develop its own brand of common sense—and to strengthen its convictions by pouring ridicule on the common nonsense of others. The social memory is contested terrain, shaped by power and prejudice as much as by disinterested inquiry.
Now consider what happens when memory is externalized in technology. Writing, libraries, and databases all extend the reach of memory far beyond what any individual brain can hold. They allow insights to be preserved across centuries and transmitted across continents. They multiply the data available for future understanding. And now, with artificial intelligence, we have systems that can store and retrieve patterns with a speed and scale no human community could match. An AI can hold in its circuits the entire history of a scientific field, the complete works of a literary tradition, the billions of data points generated by modern sensors and networks. It can, when prompted, reproduce those patterns, recombine them, and present them in novel arrangements.
However, as we work to create useful human experiences with AI, memory remains a fundamental problem. For memory, in the sense that matters for us, is not the storage of data but the accumulation of insights that have passed into the habitual texture of a knowing subject. It is the spiral of understanding, building on understanding, each crossing of the divide leaving behind a transformed mind. Our AI tools have difficulty crossing divides. It does not experience the tension of inquiry released in a flash of comprehension. It has no habitual orientation, no perceptual stance formed by years of accumulated understanding, no sense of what matters and what can be ignored. It has patterns—immense, intricate, statistically powerful patterns—but lacks the insight that explains why those patterns hold, the judgment that affirms their truth, and the lived experience of the past inhabiting the present.
This is not to diminish what AI can offer. On the contrary, it clarifies the distinct and valuable role such systems can play. Because they are tireless repositories of pattern, they can present to human minds data that would otherwise be lost or inaccessible. They can flag the anomaly, retrieve the precedent, and suggest the connection that a human inquirer might never have noticed. They can, in a sense, extend our external memory: holding materials that our limited brains cannot retain and making them available at the moment of need. This is no small thing. A well-designed AI might function as a kind of prosthetic memory, freeing human minds to do what they alone can do—grasp the point, reach the insight, make the judgment.
But we must be careful not to mistake the prosthetic for the organ. The danger with AI-mediated memory is not that it remembers too much but that we remember too little—that we become dependent on external pattern retrieval and allow our own habitual orientations to atrophy. If the self-correcting process of learning requires the accumulation of insights within a subject, then anything that short-circuits that accumulation threatens the process itself. A student who looks up every answer without struggling to understand it may pass the test, but will not have crossed the divide. An expert who relies entirely on AI recommendations may lose the hard-won feel for a field, the sense of which questions to ask, and the judgment to recognize when patterns are misleading.
The opportunity, then, is to use AI as a support for memory without substituting for it. The spiral of understanding can continue, even accelerate—not because the machine understands but because it enriches the field of experience from which human understanding can arise.
Memory is what makes us historical beings. It is the presence of the past within the present, the accumulated insights that shape perception, the habitual orientations that guide inquiry, the screening and distortion that reveal our resistances, the social fund of tested answers that gives us a place to stand. We carry time within us, folded into the texture of our minds.
The promise of AI for our memory is not that it will remember for us. It is that it will help us remember better by holding what we cannot hold, by presenting what we might overlook, by freeing our attention for the insights that only we can achieve. In this service, it becomes one more instrument in the long human project of accumulating understanding, one more layer in the already rich texture of what is given to consciousness and community. The ascent of insight remains our own. I’m excited for the collaborative possibility ahead.
“For teaching is the communication of insight. It throws out the clues, the pointed hints, that lead to insight. It cajoles attention to drive away the distracting images that stand in insight’s way.”
— Bernard Lonergan, Insight