Please Leave the Patient in the Sentence

I have a pencil between my teeth and an imaginary victory speech on the table. The speech is imaginary because I would rather invent one honestly than pretend every fool in this business has said exactly the same thing.

The promises underneath it are real. Dario Amodei has suggested that cancer and other diseases could be cured within a decade. Sam Altman has floated the possibility that enough computing power could help figure out a cancer cure. Meanwhile, the people working on actual tumors are still trying to determine what the damned things are doing.

I am going to edit this speech. I cannot improve a pancreas. Perhaps I can improve what gets said in front of one.

1. Delete the delivery date.

I would strike out within ten years before touching the punctuation. A forecast about medicine can arrive in a sick person’s house sounding like an appointment. The speaker may mean a research ambition. The person listening may start doing arithmetic with birthdays.

I do not know when these treatments will arrive. Neither does a paragraph with a confident ending. Some will work. Some will fail. Some will help one kind of patient and do nothing for another. I want that uncertainty left where people can see it, not swept into a footnote by somebody whose job is to make the future sound punctual.

An honest estimate has uses. A doctor discussing a particular treatment has evidence to explain and a patient to answer. A man announcing that a whole family of diseases will soon be history has given himself a much easier audience. He can finish speaking before anybody asks about their mother.

2. Change cancer to the thing being studied.

I would need more paper here. Cancer is not one disease waiting for one sufficiently expensive machine to guess its password. The location, the cells, the mutations, the treatment already tried: the particulars keep multiplying while the slogan stays beautifully short.

Felix Sahm, a neuropathologist in Heidelberg, describes more than a hundred types of brain tumors, not all cancerous, and unresolved questions throughout their care. Sherene Loi, a breast-cancer specialist in Melbourne, is hopeful about better diagnostics and treatment. She is not handing out a date for the end of all cancer.

I prefer these untidy sentences. They have room for somebody to be wrong about a specific thing and then learn something. The grand sentence has room mainly for applause.

I would also cross out AI wherever the speech quietly changes what it means. A model examining a scan is not interchangeable with a chatbot producing a paragraph. Success at one job does not certify a machine for every other job under the same two letters. I have worked with men who could sort mail fast and should not have been allowed to explain a parking sign.

3. Keep the part about the scan.

Here I would put the pencil down long enough to listen. At the Mayo Clinic, Ajit Goenka has been investigating whether machine learning can find early signs of pancreatic cancer in scans that were read as normal. His team’s proof-of-concept work was published in 2022. The question is whether a signal can be found earlier, and whether the method holds up across different settings.

I would not inflate that into a cure. I would not dismiss it because it isn’t one. Finding a dangerous change sooner is a serious ambition. It does not need to be promoted into the abolition of death before I am willing to care.

The speech ought to spend time here. What did the tool find? What did it miss? Who gets tested next? Instead, this is usually where I would expect the speaker to leap over the unfinished work and land in a world without disease. The jump gets easier if nobody onstage has to carry a patient.

4. Put the experiment back in.

I would underline Loi’s question: “Someone actually has to go back and do the experiment, right?”

There is a whole working day inside that actually. Probably a great many. A proposed drug must meet a living body, not merely another pleasing sentence about the proposed drug. Somebody has to establish what helps, what harms, and what happens when the conditions stop resembling the ones that made the idea look good.

I don’t want the people doing this treated as slow clerks standing between intelligence and its destiny. Nor do I want every useless delay defended as wisdom. I want to know which work protects a patient and which work merely protects a desk. Calling the entire interval red tape saves the speaker from making that distinction.

If a machine supplies more plausible things to test, I want to see the staffing bill. More lab hours, more people to run the tests and examine the failures, more capacity to find out which promising idea does nothing. Unless the machine also makes that work easier, a faster supply of hypotheses could mean a longer queue of untested ones. I cannot tell from a triumphant prediction whether anybody has paid for the other end.

The victory speech has made the computer sound like a replacement for the slow people. It may have invented a reason to hire more of them. I would enjoy watching the speaker explain that part to the investors: the wonderful machine works, and now we need to pay the people we had hoped to stop mentioning.

5. Do not replace waiting less with living forever.

Sahm talks about reducing the time patients spend in uncertainty. His collaboration is exploring AI-assisted ways to connect stages of brain-tumor diagnosis and treatment. Some of the work is in trials, not routine care. I would keep that last sentence in the same type size as the exciting part.

A shorter wait sounds unimpressive beside immortality. I suspect that comparison tells us something ugly about the people choosing the scale. Waiting for an answer can occupy a life without filling a single hour with anything useful. The person still has to eat, answer messages, decide whether to make plans.

If the work delivers that shorter wait, it gets an entire paragraph. No merely. No apology for its size. And the credit stays with everyone who made it possible, including the patients who took part in the trials. I will not edit a building full of people down to a brand name.

6. Leave the last paragraph unfinished.

The speech will want a promise here. I would refuse to supply one. I would write the questions the promise keeps displacing: whether a tool works for this disease, whether it helps this patient, whether the clinic can use it, whether the person who needs it can get there and afford the care.

These are not reasons to stop trying. They are the work. I can be pleased about a useful discovery without signing over my judgment to the company announcing it. I can want these machines to help and still resent a sales pitch that borrows its urgency from people who cannot afford to wait.

The last sentence still has a blank where the delivery date used to be. I have no replacement. In the margin I have asked whether the money for testing is included in the announcement or merely expected to turn up after everyone has applauded. Until there is an answer, the speech remains marked for revision. I expect this will be described as a problem with my attitude.

Source: Big tech says AI can find a cure for cancer. So where is it?