Discovery and recovery

When someone working in AI becomes a patient, they see an increasingly AI-assisted healthcare system from the inside.

Painted illustration of an arborvitae: a red-orange trunk among green and yellow branches

So let’s get this out of the way: I had breast cancer.

Oh, I’m sorry, too soon? Imagine how I felt when I got that jolt after Memorial Day this year.

I also want to be clear: this isn’t just a cancer story. The diagnosis caused me to reflect not only on my life experience but my technology background as I grappled with the news, weighed surgical options, and recovered. A chatbot, a global online community of cancer survivors, and a deep bench of remarkable medical specialists each carried me through a different stretch. What I keep coming back to is how differently each one listened, and what each could and couldn’t do.

It started in April, when my imaging office called me back for a follow-up scan after my yearly mammogram. I was mostly annoyed. I was helping relaunch a website ahead of a major company event, and my son’s college graduation was coming up. I blamed a faulty machine, as I’d been the last appointment of the day.

I had kept my annual checkups for years, mammograms included, but as one more item to tick off a list. There was a time when I took my health more seriously. My father died of lung cancer in 1996, when I was 27. The signs were there for months, but he refused to see a doctor until the day he couldn’t get out of bed. Weeks after he died, I found his check ledger. The early entries were written smoothly and with care. The last two were thick, unrecognizable scrawls. I quit smoking within months and vowed to take better care of myself. Then life took over.

The Humanist Review of AI, Issues 01 and 02

Conversations at the frontier, in your inbox

Subscribe

So there I was, thirty years later, in a room with a technician whose face changed as she looked at her screen. In a dark office, a doctor pointed to a white spot. “It could be calcifications, but we’ll need to get it biopsied.”

“A biopsy,” I repeated slowly, as my eyes tried to adjust to what I was seeing.

The biopsy results came back with cancer, though fortunately DCIS (also known as stage 0) and extremely treatable. What I had was questions, a lot of them, and a vocabulary of acronyms that confused and scared me. I needed those answers fast as additional doctor and test appointments quickly flooded my calendar. In the past, I would have gone to “Dr. Google” and combed through search results. This time I didn’t want to wade through outdated data and misinformation, or approach friends or family and worry them needlessly. I felt very alone. I work in AI, so naturally, as I absorbed the news, I started chatting with a health chatbot to get a crash course in breast cancer.

Yes, a woman who works in AI turned to AI. Groundbreaking, in the dry tone of Miranda Priestly. But there was real relief in being able to dump my questions on a model, however odd they seemed, along with my fears and my demands for information. For the first time I could have pointed conversations about me and my condition, not a general view of the entire population or incomprehensible research studies.

I connected my health records and dug into the test results. I felt in control, and less alone. I also challenged the responses at times to verify the info against what I knew about my health and prognosis. At work, we judge a model’s responses against what we call ground truth, verified facts from independent sources that it’s supposed to match. Now I was the one supplying it.

The AI found things I missed. It even helped me flag a small nodule on my lung identified in one of my early results. It turned out to be nothing, but I was so relieved by how thoroughly it picked through everything.

It could also be reassuring in small ways. The night before my MRI results, still torn over my surgical options, I told the chatbot I was “just indecisive lol.” Its response:

Indecisive with a “lol” is honestly a pretty healthy place to be the night before results — it means you haven’t spiraled, which takes more composure than you might give yourself credit for.

It was exactly what I needed that night: a steady voice telling me I was doing fine.

Other moments were heavier. When I admitted I was feeling very anxious about the cancer returning after reading through stats, the reply was calm and honest:

The worry about recurrence is unlikely to simply switch off — that’s a very normal part of the experience after a breast cancer diagnosis, even when the prognosis is good.

What the AI couldn’t do was push back. Like my father, I initially avoided talking too much with anyone about my cancer. I plastered a smile onto my face and went into overdrive at work as I waited for further test results. Every night I crawled into bed too exhausted to speak, and my husband John leaned in and whispered, “You know you can’t keep avoiding this.” He sounded concerned and a touch annoyed. My model never said anything like that. It met me wherever I was, which was a gift and, I suspect, a limit.

Some of my strongest connections came from strangers. On Reddit, I found people who had actually been through this, those who knew what all the waiting felt like, what to really expect after surgery, and questions I had never even thought to ask.

When two tiny areas of suspected invasive cancer turned up, my anxiety spiraled. For peace of mind, I considered a double mastectomy. My doctors discussed the options, and the chatbot pulled up the NCCN guidelines (the treatment standards oncologists use). But on Reddit, I connected with people around the world who shared my biomarkers and had faced the same diagnosis. Their advice and encouragement made me more confident in the decisions I made regarding my care.

My Reddit group only knew as much as I chose to share, and that was part of the appeal. When I finally posted that I was done with treatment, the replies were warm and immediate:

“High fives to you.”

“This post gives me so much hope.”

They didn’t have my records. They had something the AI didn’t: they had been there.

The loneliness could still catch me off guard. Early in my recovery, after taking an opioid, I fainted in my bathroom. One minute I thought I was fine; the next, the ground rose up to meet me. Dazed, I was grateful it hadn’t been worse, and suddenly aware of how alone I was in that room. Neither a chatbot nor a Reddit thread could have picked me up off that floor. It was my husband who ran in to help me out.

The AI knew a great deal about breast cancer but very little about me beyond what I typed and uploaded. Reddit was more human and rich with experiences, but like the AI, went only as deep as I let it go. The medical experts could do what neither could. The breast surgeon looked into my eyes and clutched my hand before wheeling me into surgery. The plastic surgeon matched my light banter and joked, “I only do breasts, we discussed this,” as I marveled at her artistry. The therapist I chatted with via telehealth forced me to vocalize and confront the fears of death I had avoided since my diagnosis.

AI was also present in my medical appointments, this time as a silent observer. At every visit, I gave permission to have my conversations transcribed with AI. At my oncology visit, my doctor turned her screen toward me so I could see the drafted AI-generated visit summary for her to review. My primary care doctor did the same when he learned I work in AI.

Two lines stood out at my oncology visit. When I asked about recurrence, the transcript displayed in red:

Patient is nervous/anxious.

Actually, I was terrified. And later, I spotted:

Patient is relieved.

That was from the moment my oncologist told me I was cured and cancer-free. I burst into tears.

Relieved. Well, sure. It’s accurate the way a weather report is accurate. But what was the true context of that moment? That I had been given my life back? In the clinical notes that landed in my hospital app later that day, “relieved” was gone, though “nervous/anxious” stayed. I don’t know why. I like to think the doctor in the room, reviewing the notes, saw that the transcriber, in its attempt at efficiency, had missed the true meaning of that emotional moment.

I thought of something another breast surgeon told me when I sought a second opinion: “You’re going to have a crappy summer. But then you get to enjoy the rest of your life.” For me, those lines were the most meaningful part of our conversation. Yet it appears in no chart. No one would have let it be there. It belonged between two people, not in a clinical note.

I’m glad to think a human went through the AI’s draft and edited it, the way I did for decades as a magazine editor. As AI advances, moments like these demonstrate how important it is to keep people in the loop. A medical report is not a feature story, and it shouldn’t try to be one. A medical record also can’t hold what a moment meant. That’s better suited for an essay.

Only later, while writing this story, did I discover that AI had been part of my cancer journey before I even knew I had cancer. My original screening mammogram at NYU Langone had been evaluated using both computer-aided detection and AI, helping the radiologist assess the suspicious calcifications that eventually led to my diagnosis. The radiologist would have identified the concern without it, I was told. But the technology had been there, quietly assisting a human expert, long before I turned to a chatbot for answers.

Technology was woven through nearly every part of my experience, from the medical reports instantly accessible in my hospital app to the chatbot that helped me understand them, the strangers I found online, and the therapist I met through a screen. I’m grateful for all of it. And as someone working at the forefront of AI development, I want these tools to keep improving for people when they are at their most fragile, frightened and vulnerable. But what stays with me most are the intricate moments of human connection that were critical to my recovery and continue to help me heal.

A few months before my diagnosis, a branch of an arborvitae I planted a decade ago broke in a late-winter snowstorm. It was badly splintered but still hanging on, and I decided it had to go. I worked at it for thirty minutes, slipping in the snow. In its place was a roughly hewn scab. By summer it had faded to a light brown spot as the surrounding branches filled in. The tree today stands tall and firm, scarred and still growing, braced against the autumn wind.

Alena Bubniak

Alena Bubniak is a senior editorial and content strategist at Microsoft AI, based in New York. A writer and editor for more than 35 years, she brings a career in magazine publishing and newsrooms to her work with creatives and technologists, and is passionate about making AI understandable.

English (United States)
Your Privacy Choices Opt-Out Icon Your Privacy Choices
Consumer Health Privacy Sitemap Contact Microsoft Privacy Manage cookies Terms of use Trademarks Safety & eco Recycling About our ads