AI Is Not Conscious, but It Is Becoming Our Unconscious

L. M. Sacasas

Jun 16, 2026

Welcome to the Convivial Society, a newsletter about technology, culture, and the moral life. I’ll tell you upfront that this is an odd piece in which I try to write my way through an intuition that the role of AI in society can be helpfully framed, at least in part, by analogy to the unconscious. Whether or not you buy the whole thing, I trust you’ll find at least a few helpfully provocative considerations along the way or at least some helpful questions to think with. However useful or not these reflections might be, I offer them out of the conviction that we need better, more fruitful ways of grasping these strange new technologies in our midst and the nature of our relation to them.

Cheers, Michael

1. According to the early 20th-century English mathematician and philosopher Alfred North Whitehead, “It is a profoundly erroneous truism, repeated by all copy-books and by eminent people when they are making speeches that we should cultivate the habit of thinking of what we are doing.” “The precise opposite is the case,” he argued. In his view, “civilization advances by extending the number of important operations which we can perform without thinking about them.” He concluded with a rhetorical flourish: “Operations of thought are like cavalry charges in a battle — they are strictly limited in number, they require fresh horses, and must only be made at decisive moments.”

There’s a good chance you’ve seen this quoted at some point. It’s fine. It conveys some bit of truth, I’m sure. Nonetheless, it has always struck me as somehow shortsighted or inadequate. Perhaps it’s because I tend to see these lines from Whitehead quoted in defense of indiscriminately outsourcing human cognitive activities to machines without a proper accounting or even awareness of the attendant costs. In defense of Whitehead, who hardly needs my defense, it may also be because those who are quoting him in this manner are almost certainly extending the force of his argument beyond the scope he intended. After all, that paragraph comes from his 1911 An Introduction to Mathematics and what he is actually talking about in that section is the advantage of symbolism and notation in facilitating mathematical calculations, and while he characterizes these as allowing for operations performed without thinking, they are nonetheless learned and deployed by the thinking mind. That said, his talk of “civilization” and that last rhetorical flourish probably invites such a (mis)reading.

The question is not whether such automations of thought, or even externalizations of certain mental processes, can be useful in certain cases. For what it’s worth, I actually think you get a more compelling argument by analogy in the realm of physical rather than mental activities. I’m many years removed from whatever athletic skill I might have once possessed, but the lessons are not lost to me. Much to the dismay of almost every young athlete taking up a sport for the first time, you must spend what always seems like an inordinate and excruciating amount of time executing basic and repetitive drills, over and over again. Wax on, wax off, for those of you of a certain age. But what is, in fact, happening is an ideal example of the dynamic Whitehead is describing. You are in effect automating certain physical movements so that you can perform them without having to think about them. Only then can you play with any kind of creative or exceptional skill. The same holds for learning to dance or to play a musical instrument, etc. Automating basic physical motions is the indispensable foundation of virtuosity.

But does this dynamic apply equally to all realms of human activity? Are there cases in which outsourcing certain forms of activity undermines rather than enables the achievement of the higher goods for the sake of which the activity is pursued? Or might there be goods that attend the “lower order” activities that we would not want to do without? Or is it even always possible, as in mathematics perhaps, to so easily disambiguate distinct sub-routines from given processes or activities? Are there not any irreducibly integral activities that would not survive contact with an attempt to outsource or automate any of their elements? And is there not a difference, as I suggested above, between internally mastered automations of thought and the outright externalization or wholesale outsourcing of cognitive labor? To return to the analogy to physical activity, the would-be athlete that hypothetically employs a machine to do all the “menial” drills for them so that they can get to the really exciting parts of the game will, in fact, never get to them at all. These all strike me as vital and critical questions to explore before we assent to the promises of efficiency and liberation so readily made on behalf of novel technologies.

1. It may be, however, that I’m also rankled by Whitehead’s oft-cited celebration of cognitive automation because it appears to starkly contradict Hannah Arendt’s admonition that we “think what we are doing.” Whitehead was writing nearly half a century before Arendt, and I don’t think Arendt was in any way alluding to Whitehead. It just so happens that these two claims rhyme antagonistically, and I find it useful to explore the tension.

“It is a profoundly erroneous truism … that we should cultivate the habit of thinking of what we are doing.” — Alfred North Whitehead

“What I propose, therefore, is very simple: it is nothing more than to think what we are doing.” — Hannah Arendt

Interestingly, Arendt’s admonition comes from a context that is much closer in its concerns to our present anxieties about technology and human existence. This line comes from the Prologue to The Human Condition, which was first published in 1958. “What I propose in the following,” she explained,

“is a reconsideration of the human condition from the vantage point of our newest experiences and our most recent fears. This, obviously, is a matter of thought, and thoughtlessness—the heedless recklessness of hopeless confusion or complacent repetition of ‘truths’ which have become trivial and empty—seems to me among the outstanding characteristics of our time. What I propose, therefore, is very simple: it is nothing more than to think what we are doing.”

Among the “newest experiences” eliciting Arendt’s investigations were the launch of Sputnik and the advent of automation. While it may have until recently been judged that Arendt’s fears regarding automation’s impact on labor were misplaced, it might turn out that they are better judged to have simply been premature.

1. As I consider the difference between Whitehead and Arendt on the question of whether we need more thinking or less, I find it useful to distinguish between, on the one hand, building a repertoire of non-conscious routines that allow you to develop higher capacities and, on the other, being in thrall to unconscious forces which erratically drive your behavior in a potentially destructive manner.

Is there a threshold across which Whitehead’s principle flips or reverses? “Civilization advances,” Whitehead claimed, “by extending the number of important operations which we can perform without thinking about them.” But what if there is a tipping point? Past a certain threshold, either of quantity or quality, does civilization degrade by extending the number of important operations humans can perform without thinking about them? Or at the very least, do we get a very different sort of civilization than the one we have known? We may soon learn the answer to these questions because the adoption of AI often amounts to the automation of our thinking and the creation of a vast realm of non-conscious action in the world that arises from the externalized storehouses of our personal and collective memory.

1. Arendt worried that we might lose the ability “to think and speak about the things which nevertheless we are able to do” as our technological capabilities, explicable chiefly in the language of mathematics, outstripped our capacity to comprehend them in ordinary language. (This suggests that a form of the problem of interpretability predates the advent of LLMs and so-called blackbox algorithms.) In a passage that has renewed relevance and urgency, she went on to say that “it would be as though our brain, which constitutes the physical, material condition of our thoughts, were unable to follow what we do, so that from now on we would indeed need artificial machines to do our thinking and speaking.” This was a political problem of the first order. Under such conditions, we would no longer be, in any meaningful sense, governing ourselves. Politics as a realm of human action, and constituted by speech, would cease to exist.

1. On Christmas Day, 1958, the same year that Arendt published The Human Condition, W. H. Auden, published “Friday’s Child,” which included these thoroughly Arendtian lines:

The self-observed observing Mind We meet when we observe at all Is not alarming or unkind But utterly banal.

Though instruments at Its command Make wish and counterwish come true, It clearly cannot understand What It can clearly do.

Auden and Arendt were friends and Auden favorably reviewed The Human Condition, but I am uncertain as to the direction of influence.

1. While Arendt offers a straightforward account of how we arrive at the place where we clearly cannot understand what we can clearly do, that is to say, when our technologically enabled action cannot be fully comprehended by our ordinary language, there is another intriguing way of framing the matter.

Actions that are impenetrable to language and thus to conscious thought can also be explained by reference to the unconscious. I am not one to instinctively appeal to the language of the unconscious, but it may be a useful analogy in the service of understanding our relation, individually and collectively, to the artificially intelligent apparatus that is increasingly mediating our experience of the world.

To put it plainly, I can think of two ways of analogizing certain instances of consumer AI to the unconscious. The first is relatively straightforward: as we outsource more and more tasks at the personal, organizational, and institution levels of society to agentic AI, we generate a layer of activity in the world that is functionally sundered from active human judgment and oversight.1 And to the degree that this layer of society structures and informs our experience, the ratio of conscious to unconscious human action shrinks.

Obviously, since at least the dawn of the industrial age if not earlier, there have been relatively opaque systems at work in the human lifeworld. The difference, as I see it, is that these systems were relatively sequestered from the course of ordinary human activity. Artificially intelligent systems on the other hand are increasingly woven far more intimately into our experience. In many cases, these are not only processes that are brought to bear on us by external forces, they are also processes that we unleash and initiate and which act for us and back on us. It is this proximate intermingling of human and machine that leads me to reach for the analogy to the unconscious.

Erik Hoel recently made a similar case more eloquently and at greater length. Here is his conclusion:

The traditional danger of AI is usually thought to be superintelligence acting as an existential threat. Yet, this may miss the true and more subtle danger: the AI revolution is a mechanism for transferring the processes of our civilization from under the supervision of consciousness to unconsciousness. But as AI removes consciousness from the workings of the world, it renders the world increasingly uninterpretable, ever more strange and unintelligible. So far, the great ensloppification of the commons has supported this as the major risk of the LLM revolution. And as

[truncated for AI cost control]