The essays · 4 September 2026
On what the algorithm reduces and what medicine must expand.
Some years ago, I took care of a petite woman in her early seventies admitted to the intensive care unit after her surgery was unexpectedly complicated with severe bleeding and shock. She was on life support, with an array of drips pouring into her small body, like creeks from melted snow in early spring. The early evening was sinking over the Florida sky and hard rain drummed on the windows, steadily drowning the sound of alarms. Her family had brought a small brown teddy bear and set it on the bedside table beside the ventilator alarm.
The bear was old and worn in places but still sitting upright on the table. “It survived so many hugs and kisses from our son since he was three,” her husband told Marsha, an ICU nurse who was planning to retire at the end of the year. Their son was now forty-three and working overseas, trying to figure out what to do, whether to come or wait. His teddy bear was guarding his mother in the ICU, bringing his silence as a shield against the hum of the machines.
When you can no longer bring words into your mother’s ICU room, you bring your love by any means possible.
The electronic medical record did not have a field for the teddy bear. The score calculated to measure her severity of illness did not adjust for it. No algorithm in the room registered it as clinically relevant.
Yet, the teddy bear was the most important thing in that room. No artificial intelligence deployed at any bedside in the foreseeable future will register it as such, and that failure of registration is not a bug we can eventually fix. The algorithm was never designed for it, and likely never will be, because it simply cannot see it.
Here at Nonalgorithmic I write about the human work of medicine in the AI era. Every week I rewrite in different keys an argument that some parts of a person cannot be reduced to computation, and that the practice of medicine exists to attend to the parts that cannot be reduced to 1 and 0.
Nearly forty years ago, nineteen degrees east and forty-five degrees north, Đorđe Balašević wrote the song D-moll and three lines I have carried in my body since I was young enough to still believe I would go home.
Three lines, in a language older and truer than any I can use in a policy paper.
Line one. Opening.
Balašević refuses the reduction. The song describes a state that some might call sadness, and he interrupts to remind us that sadness is too small a word for what this is. The state has its own vocabulary. Do not smooth it down.
This is the temptation medicine has been sliding into for two decades and that AI is about to accelerate to a rate we have never had to think through before. Depression is not just a PHQ-9 score of 12. Grief is not just an adjustment reaction that will resolve in six months. Dying is not a goals-of-care conversation with a time limit.
The nearest measurable proxy is never the thing itself: not the pain the family is carrying at the bedside, not the suffering her body is carrying. The proxy is what we can put in the field of the digital medical record. What is in the field is not what is in the patient.
The algorithm cannot know this because reduction is what an algorithm is. It compresses, matches, produces the nearest available label. It cannot say nije to, it is not that, because that is the only vocabulary the algorithm has. The physician in the room can and must.
The voice of the non-algorithmic self begins with a physician who is willing to say to the machine, what you have just labeled is not what this person is, and to keep saying it until the machine or the profession or both learn to hear it.
Line two. The bridge.
Balašević’s D-moll arrives when he is alone in the room. He does not resist. He knows the road, and he has been carried down it before.
The physician at that bedside, a woman attending another woman, is doing the same walking. The pain is already in the room. She is guiding them to it, by the shortest and cleanest path. The sentence she uses when she sits down and says your mother is dying and I want to help you understand what that will look like is the shortcut, delivered quietly by someone who has walked this path many times.
An artificial intelligence system can describe the pain. It can predict, from the trajectory, that the family will grieve, but it cannot walk the path with them.
Walking is not modeling, accompaniment is not description, presence is not information transfer. It is the guided delivery of a person to a place they were already going, by the shortest route their body can bear.
Line three. The coda.
Balašević ends with an inventory of the small little things the loved one left behind, letting them carry the whole weight of the absence.
Dubravka Ugrešić, six degrees west and seven degrees north of Novi Sad, opens The Museum of Unconditional Surrender with Roland the walrus, who died at the Berlin Zoo in August 1961. In a case beside his pool the zoo displayed what was found in his stomach: a cigarette lighter, lollipop sticks, a beer-bottle opener. The book is built the same way: fragments of a life in exile, set side by side because there is no longer a whole to set them in. Item by item, the list is junk. Together, it is a life.
The teddy bear on the bedside table of the woman in the ICU is that coda. The family photographs propped on the wooden edge of the pull-out sofa. The bathrobe hung on the closet door. The plastic bag with the watch and the ring tucked in the closet beside it.
The physician who attends to them is attending to the whole weight of what the patient is losing. The physician who does not attend to them is treating something that is not the person.
An artificial intelligence system will not register the teddy bear because it was not trained to see clinical weight in the small little things we call life. The training data does not have a label called son’s bear that meant everything, waiting on another continent for directions to the shortcut to pain.
The small little things that make a human life are invisible to the model, and if we build clinical workflows entirely around the model, they will become invisible to the profession.
I do see an important place for AI in medicine, hoping it will make many of us better at parts of our work we have never enjoyed being bad at. I welcome this, and I have spent the last decade of my career building the framework and the governance that will let it happen responsibly.
Two weeks ago in JAMA, Ezekiel Emanuel, Vinod Khosla and their colleagues argued that AI alone will soon exceed both physicians and AI-aided physicians at five medical tasks, and that humans in the loop degrade AI performance. They may or may not be right. Every one of the five is a cognitive task, and the authors say so plainly. Whether AI surpasses us at those tasks in the real world, with real patients, has not been rigorously tested. And if one day it does, the harder question is still waiting: what did the patient gain?
Because nothing in that paper is about the teddy bear.
I am writing to argue something very precise: that the parts of medicine AI will never do are the parts medicine is all about:
The refusal to reduce life to numbers, likelihoods and probabilities.
Delivering an unbearable knowledge.
Being present in another’s D-moll.
Witnessing the small little things that hold the entire absence.
Đole of our Balkan youth, our sad jazzman, sang all of this nearly forty years ago, about a wanderer who kept leaving.
The algorithm reduces and the physician expands. The refusal and the expansion are what medicine is.
Balašević died on 19 February 2021, of Covid, in Novi Sad. For a week the whole region of the former Yugoslavia sang him back. He had written his own coda thirty-seven years earlier: the records, the unread pages, the fine little blue powder box.
The sad jazzman came for him. He knew the way.
He always did.
Azra
The photographs are mine: my son’s room, my cat Gracie. The bear in the essay belonged to someone else’s son.
The book. Muzej bezuvjetne predaje (The Museum of Unconditional Surrender), Dubravka Ugrešić, Fabrika knjiga, Belgrade. In English, translated by Celia Hawkesworth.
A novel in fragments, written in exile after Ugrešić left Croatia in 1993, about memory, photographs and the things a life leaves behind.
The song. D-moll, Đorđe Balašević, live at Sava Centar, Belgrade, 1994.
Balašević was the singer of Novi Sad; he wrote of the Pannonian plain as if it were a person.
This essay first appeared in Nonalgorithmic on Substack, 4 September 2026.