The cart is a tablet on a wheeled pole, about five feet high, with a camera at the top and a handle for pushing it between rooms. It connects in under a minute to an interpreter in one of about two hundred and forty languages. There is a log. The log records, per encounter, the language requested and that an interpreter was provided.
That log is what the hospital reports, and by it we are compliant, and I am the person who signs some of those encounters off.
Here is the claim, and it is narrower than the one usually made about this equipment, which is that it is impersonal. Impersonal I could live with. The claim is that the thing a medical interpreter actually does is not translation, and the cart does translation excellently and cannot do the other thing at all — and because the log records only that an interpreter was provided, the loss is invisible in every document the institution produces about itself.
What does an interpreter do? I worked for four years in a clinic that had two on staff, in the building, and I can tell you what those two did that a screen does not.
They watched the patient while I was talking, because they were not the ones talking. Then they told me things. Not translations — observations. She said yes and she does not understand. He is answering the question he thinks you are asking. That word does not exist in her dialect and the one I used means something slightly harsher, so soften it when you repeat it. His daughter is in the room and he will not say this in front of her, ask her to step out.
None of that is on the transcript. All of it is the job. An interpreter with a line of sight to a patient's face and a relationship with the clinician is running a second, parallel channel about how the conversation is going, and that channel is where the clinically important failures get caught.
A camera on a pole has one field of view and it is pointed at whoever positioned it. The remote interpreter cannot see the daughter in the corner. They have ninety seconds of context, no relationship with me, and a queue behind them, and they are — correctly, professionally, exactly as trained — rendering utterances.
Now the case against everything I have just said. It carried enough weight with me that I changed my practice without changing my opinion, which is an uncomfortable place to have ended up.
On-demand remote interpreting made two hundred and forty languages available at three in the morning in hospitals that have never had a single staff interpreter and never will. Before the cart, the realistic alternatives in a small hospital at 3am were: a bilingual nurse pulled off her own patients, a family member, or the patient's twelve-year-old daughter. I have watched a child interpret a cancer diagnosis to her mother. Anyone who wants to be nostalgic about the era of in-person interpreting is describing a handful of large urban hospitals and ignoring the entire rest of the country, and the cart is one of the genuinely large improvements in access I have seen in my career.
So the cart should not go. What I am arguing about is that we stopped there, and we stopped there because the measurement said we had arrived.
The second case is where this stops being about quality and becomes about consent.
A consent conversation is not an exchange of information. It is a process with a required output: a clinician's judgment that this person understands what is being proposed, what happens if they decline, and that they are free to decline. Every part of that judgment is made from things that are not words — how long the pause was, whether the yes came too fast, whether the eyes went to the relative in the corner before the answer.
I have signed consents obtained through the cart. Under the regulation they are valid. Under my own understanding of what I was certifying, some of them I am not sure about, and the two I am least sure about were both procedures that went fine, which is not the same as the consent having been real.
That is the part I have not resolved and I do not want to smooth it over. The instrument that made access possible is the same instrument that makes the assessment of understanding harder, and both of those are true at once, and pretending otherwise in either direction is how this conversation usually goes.
What I want is not a return to staff interpreters, which is not on offer and would not scale. It is that the log record two more fields it could record today: the modality, and the duration. A four-minute remote session and a twenty-five-minute in-person one currently produce the same row. With those two fields, an institution could see for the first time whether its consent conversations in Spanish take as long as its consent conversations in English, which is a question nobody in my building can currently answer and which every clinician in it could guess.
She was seventy-four and the cart was between us at an angle where I could see the interpreter and she could see mostly the back of it. At the end I asked, through the tablet, whether she had any questions, and she said no. Then, after I had stood up, she said something else, and the interpreter said: she says thank you for coming.
She had said something longer than that. I could hear that it was longer. I asked, and the interpreter said it was a blessing and did not translate directly, which was probably correct and which I have thought about every week since, because I have no idea whether it was a blessing or a question and there is no field on the form where that fits.
Modality and duration. Two fields, already collectable, and they would answer whether consent conversations in one language take as long as in another. That question being currently unanswerable is the finding.