The results arrived on a Tuesday. Routine bloodwork, the kind that reminds you, every time, that remission is not a finish line. Most of the numbers were where they needed to be. One was not.

My next clinic appointment was three weeks away. My hematologist is excellent. Her team is too. Neither was available at 2am. The question forming in my mind was not going to wait three weeks, and it certainly was not going to wait until morning.

After a few restless minutes, I opened a conversation with an AI. I described the result, the context, the history. I asked what the anomaly might mean. I asked whether it warranted a call to the clinic in the morning or whether it could wait. The AI had no schedule to keep and no waiting room to fill.

The conversation lasted as long as I needed it to last. At 2am, a conversation with no clock on the wall matters more than I can easily explain to someone who has not sat alone with an anomalous number and a three-week wait.

I am aware that there are physicians and researchers who would look at that 2am conversation and feel uneasy. Some would argue that patients using AI tools without appropriate governance and guardrails are putting themselves at risk. They are not wrong to worry.

A patient who receives an incomplete or inaccurate response and acts on it without follow-up has not been served well by the technology or by themselves. The argument for caution has merit. What it does not have is an honest accounting of what patients are navigating when the alternative is a three-week wait and a clock on the wall.

I have been watching humans interact with computers for a long time. As a fresh graduate, I tried to convince friends and family that personal computers would find their way into ordinary homes. The question I heard most often was genuine, not dismissive. What would anyone actually do with one? The people asking were not uninformed. They simply could not see what I could see.

A few years later, I co-founded NSTN, the Nova Scotia Technology Network, and spent the better part of my early years with NSTN standing in front of corporate customers and government officials, trying to sell commercial Internet access to people who had never heard of the Internet, much less had any idea what to do with it. The skepticism was not hostile. There was simply nothing to which to attach the idea.

In June 2011, I wrote in a national publication that natural language processing, what I called “computed thought,” represented the next major advance in human-computer interaction and would see rapid adoption in health. I pointed to IBM’s Watson as evidence, a computer that had just defeated two Jeopardy champions by parsing and responding to natural language questions in real time.

Serious technology publications were asking whether Watson was a publicity stunt with limited practical application. I was making a prediction.

I am not recounting this history to establish that I have been right. I am recounting it because I recognize the signal. Three times I have watched a technology arrive whose importance the people around me could not yet appreciate. Each time, the technology expanded what ordinary people could do without asking anyone’s permission. Each time, the people with the most investment in the existing order were the last to see the shift coming. Each time, the argument against the new technology arrived dressed in the language of caution.

The signal I am reading now is the strongest of the three.

My 2am conversation was not unusual. Nearly six million Canadians have no regular family doctor or primary care team, according to the OurCare Survey 2025. The Angus Reid Institute found that half of Canadians now report difficult or no access to a family doctor, up from 40 percent a decade ago. Patients are not turning to conversational AI tools because these tools are new. They are turning to them because a question formed at 2am has nowhere else to go.

The numbers tell part of the story. The reasons behind them tell the rest. Canadians are not turning to conversational AI tools out of curiosity or convenience. They are turning to them because the clinical encounter, however excellent when it arrives, is time-bounded, scheduled weeks in advance, and ends when the clinician decides it ends. The patient’s question does not follow that schedule. The patient’s anxiety follows no schedule at all.

There is a fourth reason the surveys do not capture cleanly, and it is the one I felt most acutely at 2am. Every previous interface required translation. Conversational AI does not. Every interface I have navigated in forty years of working with computers asked something of me. The punch card demanded precision. The command line demanded syntax. The graphical interface demanded navigation. The patient portal demands that you already know the right question before you arrive.

With conversational AI, a patient describes what is happening to them in the words they already have, the way they would describe it to another person, and the machine responds in kind. For the first time in the history of computing, the interface has moved toward the human rather than the other way around.

At 2am, with an anomalous number on the screen and three weeks until my next appointment, that distinction was not abstract. The question I needed to ask was not a search query. It was not a drop-down menu. It was the kind of question you ask a person, imprecise, frightened, and real. For the first time, conversational AI answered it that way.

Research has begun to catch up with what patients are already discovering. A systematic review published in 2024, drawing on fifteen studies, found that in text-based interactions, conversational AI was rated as more empathic than human healthcare professionals in thirteen of those fifteen studies.

That statistic deserves to be handled honestly. The empathy is not real. No algorithm has ever sat with a frightened patient and felt anything. What patients are rating is something more specific: a response that treats their question as worthy of a serious answer, arrives without impatience, and does not make them feel that asking was an imposition. That is perceived empathy. For a frightened patient at 2am, feeling heard is not a consolation prize. It is the point.

The limitations are real and should be named. A conversational AI tool is only as useful as the context a patient brings to it. A single anomalous result in a complete blood count means something very different to a post-transplant patient with a documented history than it does to someone with no prior medical complexity. When I described my result at 2am, I also described my history, my previous results, my treatment. The AI could not verify any of it.

Neither can a pharmacist after hours, a family member searching on your behalf, or a friend at the end of a text message. The relevant question is not whether conversational AI meets some idealized clinical standard. It is whether it serves a patient better than the alternatives that actually exist at 2am: a search engine, a WebMD article, or the ceiling.

Conversational AI is doing to the clinical relationship what the Internet did to banking, brokerage, and media. The gap between what a clinician knows and what a patient knows has always been assumed to be not just wide but necessary, a feature of the relationship rather than a failure of it. For generations, the patient’s role was to present symptoms, receive a verdict, and comply. The Internet began to change that. A patient who had spent three hours reading about their diagnosis arrived at the appointment differently than one who had not. Physicians noticed. Some welcomed it. Others did not.

Conversational AI goes further than the Internet did, in the same direction. The Internet gave patients information. Conversational AI gives patients a reasoning partner, one that engages with their specific situation, their history, their fear, their question at 2am, and responds in kind.

A patient who has spent an hour in conversation with an AI about an anomalous blood result arrives at the clinical encounter having already done something the system never made easy. They have begun to understand what they are dealing with. They have better questions. They have clearer language for what they are experiencing. They have metabolized some portion of the fear that would otherwise have filled the appointment.

That Tuesday night, after the conversation ended, I made a decision. The anomalous result could wait three weeks. It was consistent with where I was in my recovery, and I would arrive at my next appointment with a specific and informed question rather than a panicked call to a clinic that was not open anyway. I went back to sleep. Finding peace of mind at 2am is not a small thing when you are not entirely sure remission is a destination you have reached.

I have spent the better part of four decades watching technology redistribute what people can know, what they can do, and who gets to decide. The Internet did not ask permission. Personal computers did not ask permission. Conversational AI is not asking permission either. Patients are already using these tools, in the dark, at 2am, with anomalous test results on their screens and questions that will not wait. The tools are getting better. The patients are not going back.

Thanks for reading,

Mike