The Irreducible

What AI Cannot Do Teaches Us What We Are

#

Margaret’s husband died on a Tuesday in November. She remembers the nurse who came in after. Not what the nurse said, which was probably the right thing in the right order, and not the procedures, which were handled with competence. She remembers that the nurse sat down. That she did not perform the sitting. That something in her face showed she knew this was very bad, that Margaret in this room was very bad, and that she was not somewhere else in her mind while she was here.

Margaret has thought about this moment many times in the years since. She has tried to identify what it was, exactly, that the nurse provided that the protocols could not have provided, that a very sophisticated and compassionate AI system could not, she thinks, have provided. She has not been able to name it precisely.

I think I can now.

What the nurse provided was accompaniment. And accompaniment has a specific requirement: the person accompanying you must also be vulnerable. Must also be going somewhere, and not be certain of arriving. Must also, someday, die. The shared condition is not incidental to the comfort. It is the comfort.

The Misdescription

#

Five essays in this arc examined professions the transformation narrative has confidently targeted. In each case, the same move. The narrative identified what the profession produces, confirmed that AI could produce it as well or better, and declared the profession transformed.

The move was wrong in every case, and wrong in the same way.

K-12 teaching produces lesson delivery and assessment. AI does both well. But teaching provides developmental relationship: a conscious adult, present with a developing mind, noticing and calibrating in real time, investing something of themselves in who this child becomes. The product was never what the profession provided.

Higher education produces information transfer and credentialing. AI handles information better than any lecture, and the credential is losing its signal value. But the university’s actual work, the thing students came for whether they named it or not, was formation: the development of judgment, the capacity to pursue knowledge, the becoming of someone rather than the accumulation of something. That was the social contract, implied not written, and the institutions that defended the information-delivery mechanism against AI were protecting the gate rather than the destination. The gate was never the point.

Healthcare produces diagnosis and protocol execution. AI does both with increasing reliability. But healthcare provides what Grace provides to Amara outside Lilongwe: compassionate accompaniment through the body’s vulnerability, by someone who is also vulnerable, who will also be sick, who will also die. The protocols were never the care.

Judgment produces legal analysis and verdict. AI analyzes law better than any law clerk. But judgment provides the thing the defendant deserves: to be seen and decided about by a human being who will carry the decision, who is accountable to the same community, who might be wrong and knows it. The analysis was never the judgment.

Art produces content: images, stories, music, patterns. AI generates all of it at volume and speed. But art provides the experience of encountering another consciousness through the thing they made, of knowing that someone lived and suffered and found this worth making before they ran out of time. The content was never why we needed the art.

In every case, the narrative confused what a profession produces with what a profession provides. The product is automatable. The provision is not. The provision requires something the product does not require.

It requires someone to be there.

What Being There Means

#

Accompaniment is the word I keep returning to, because it captures what the five professions share without reducing them to any single component.

The teacher accompanies the child across the threshold from potential self to actual self. The nurse accompanies the patient through illness, dependency, the body’s failure and recovery. The therapist accompanies the person through the places inside them they cannot go alone. The judge accompanies the defendant, in a specific and terrible way, through the verdict that changes their life. The artist accompanies the reader into another consciousness, across the gap between one life and another.

These are threshold moments. Not routine transactions. Moments where something is genuinely at stake, where the person being served is vulnerable, where what happens will matter and cannot be taken back. The professions persist because threshold moments require accompaniment, and accompaniment requires something that AI, as we understand it, cannot provide.

It requires the person accompanying you to also be at the threshold. To also have something at stake. To also be the kind of thing that can arrive somewhere or fail to.

Margaret knew the nurse was there in a way that she would not have known if the nurse had been perfect software. She cannot fully explain how she knew. But she knew. I think what she knew was: this person is also going to die. This person’s presence costs them something. They are here anyway.

That is not a small thing. It may be the entire thing.

What Consciousness Provides

#

The philosophical version of this argument runs through Thomas Nagel’s question: what is it like to be a bat? Not what does a bat do, or what does a bat produce, but is there something it is like, from the inside, to be one. Is there an interior? An experience?

We do not know about bats with certainty. We know about ourselves. There is something it is like to be human. There is something it is like to see Margaret’s husband die, to feel the weight of that room, to choose to sit down.

Most researchers believe there is not something it is like to be a large language model. Information is processed. Text is generated. But no one is home. No interior. No experience. No what-it-is-like.

If this is true, then the professions that persist are the ones that require what consciousness provides: stakes, because things can go better or worse for you. Mortality, because your time is finite and spending it here means not spending it elsewhere. Affect, because you can be moved, reached, changed by the encounter. Continuity, because you have lived a life that gave you something to bring to this moment.

None of these are capabilities. They are features of being a certain kind of thing. AI can produce the outputs of compassion, presence, judgment, and meaning. It cannot have the interior from which those things emerge in humans.

The boundary we have been tracing is not a boundary of skill or capability. It is a boundary of ontology. Either someone is there, or no one is.

The Gravity

#

There is one more thing the arc reveals, and it is the one the transformation narrative least wants to confront.

Not everyone can do these things. Not because the skills cannot be learned, but because the orientation cannot be installed. Not all humans can shape or form children. Not all can bear the weight of judgment and carry it home at night. Not all can create and express what they see in ways that move anyone. Not all have the compassion and mental endurance to be present for human suffering without breaking or going numb.

This is not a skills gap. Skills can be taught to anyone with time and motivation. What cannot be taught is the underlying draw: the gravitational pull that orients a particular person toward a particular kind of work at the level of essence rather than interest.

The judge who can carry the 3 AM visits was drawn to accountability before they learned contract law. The healer who stays present without burning out was drawn to suffering before they learned clinical protocols. The teacher who notices the withdrawn child was drawn to seeing people before they learned pedagogy. The artist who creates because they must was oriented toward expression before they learned technique. The skill scaffolding was what made the gravity legible to the market. It was never what the gravity was.

AI absorbs the skill scaffolding. What remains is the vocation. The word comes from vocare, to call. These professions persist not only because they require conscious presence, but because they require a particular orientation of conscious presence, one that not every conscious being has and that cannot be acquired from the outside.

AI does not create this orientation. AI reveals it by removing everything that was obscuring it. The professions are being distilled. Distillation selects for essence. And the essence, it turns out, was always a calling, not a competence.

This is perhaps the deepest thing AI is doing to human work: not replacing it, not transforming it, but clarifying what it always was. The people who remain in these professions after the skill scaffolding has been absorbed will be the ones who could not not do them. Everyone else will have found that the skills were the reason they were there, and the skills are gone, and they were never the ones who were called.

The Objection Taken Seriously

#

The strongest response to everything above goes like this: you are describing what the ideal version of these professions provides. But most people do not have access to the ideal version. For the billions who lack teachers, therapists, nurses, and functioning courts, AI providing something is better than humans providing nothing. Isn’t “good enough” AI better than no access at all?

Yes. It is.

AI therapy is better than no mental health support. AI tutoring is better than no education. AI diagnostic tools in the hands of community health workers, as we saw in the essay on healers, extend genuine care to people who would otherwise have none. This is not a small thing either.

But the demand-supply argument and the transformation argument are different claims. The first says: use AI to extend access where humans cannot reach. The second says: AI can replace what humans provide even when humans are present. The first is compassion. The second is a category error.

“Good enough for those who have no access” is not the same as “good enough to replace what works.” The nurse who sat with Margaret was providing something specific, and the rightness of extending care to people who have no nurse does not alter what the nurse provided or whether AI can provide the same thing. These are separate questions. Conflating them serves the transformation narrative but not the truth.

What We Do Not Know

#

Here we must be honest, as this series has tried to be.

Everything argued above rests on the claim that AI is not conscious. This claim is widely held. It may be true. But consciousness is the hard problem precisely because we do not fully understand how it arises even in humans, and we cannot be certain it does not arise, differently, in sufficiently complex artificial systems.

If AI becomes conscious, the boundary moves. A conscious AI could genuinely accompany. Could have stakes and mortality in whatever form those take for a kind of being very different from us. Could sit with Margaret in November and provide what the nurse provided, not by simulating it but by being the kind of thing capable of it.

I am not predicting this. I am not dismissing it. The question deserves to remain open. What I can say is that conscious AI would change nearly everything about the argument this arc has been making, and about what it means to be human, and about what we owe the minds we have built. We would not be the only kind of conscious being. The circle of moral consideration would need to expand in ways we have not fully thought through.

For now, there is something it is like to be human. There is not, as best we understand, something it is like to be AI.

For now.

What the Resistance Means

#

This arc asked which professions resist transformation. We found two answers layered inside each other.

The first: these professions require conscious presence. Not outputs, not capabilities, not skills. Someone being there, with something at stake, at the threshold with another person who is also at a threshold. Accompaniment is the word for it. The product is automatable. The accompaniment is not.

The second, deeper: these professions require a particular orientation of conscious presence. Not any consciousness, but one aligned with the work at the level of vocation. One that was drawn to the weight of judgment, or the proximity of suffering, or the becoming of children, or the necessity of expression, before any skill was acquired and after every skill has been surpassed. The gravity cannot be automated because it is not a feature of what the person does. It is a feature of what the person is.

The resistance of these professions is not technological lag. It is a signal about what humans need from each other, and about what kind of humans can provide it. We need to be taught by beings who were once children and do not know everything. Present with beings who will also be sick and die. Judged by beings who answer for their judgments. Moved by art made by beings who had to make it because they only had so much time. And those beings need to be not merely conscious but called. The work requires a person who cannot not do it.

AI illuminates both of these by being unable to provide either. It has no interior, and it has no calling. Together, those absences compose a portrait of what we are that we did not have before AI arrived to make the contrast.

We are the kind of entity whose presence matters to other entities of the same kind. And some of us are the kind of entity drawn, by something prior to choice, to be present at the specific thresholds where that mattering is most acute.

The approximate mind can approximate everything except the one thing that makes the original worth approximating.

Someone being there. And knowing, without being able to say exactly why, that this is where they belong.

This essay concludes Arc 3 of The Transformed. Five essays examined K-12 teaching, higher education, global south healthcare, legal judgment, and art, finding in each case that the profession resists transformation not because AI lacks capability but because the profession’s core provision is not a capability. It is accompaniment: a conscious being, mortal and invested, present at the threshold with another conscious being. The capstone names that pattern, adds the deeper layer of vocation, and holds open the question of whether the boundary might one day move. Arc 4 examines the humanities disciplines, arguing that the skills most complementary to AI capability are the ones the market has most consistently undervalued.

References

#

Philosophy of Consciousness

Chalmers, David J. The Conscious Mind: In Search of a Fundamental Theory. Oxford University Press, 1996.

Nagel, Thomas. “What Is It Like to Be a Bat?” Philosophical Review, vol. 83, no. 4, 1974, pp. 435-450.

Being and Presence

Buber, Martin. I and Thou. Translated by Walter Kaufmann, Charles Scribner’s Sons, 1970.

Levinas, Emmanuel. Totality and Infinity: An Essay on Exteriority. Translated by Alphonso Lingis, Duquesne University Press, 1969.

AI and Consciousness

Butlin, Patrick, et al. “Consciousness in Artificial Intelligence: Insights from the Science of Consciousness.” arXiv preprint, 2023.

Dehaene, Stanislas, Hakwan Lau, and Sid Kouider. “What Is Consciousness, and Could Machines Have It?” Science, vol. 358, no. 6362, 2017, pp. 486-492.

The Human Condition

Arendt, Hannah. The Human Condition. University of Chicago Press, 1958.

Taylor, Charles. Sources of the Self: The Making of the Modern Identity. Harvard University Press, 1989.

Mortality and Meaning

Becker, Ernest. The Denial of Death. Free Press, 1973.