🎧 Listen to the Discussion(two hosts talk through their thoughts on this article)
This is an AI-generated audio overview of this article, created with Google NotebookLM. Two hosts discuss the article together, sharing their own thoughts and commentary — it is not a reading of the article text.
In part 5, I read the transcript of a conversation lasting more than five hours to find out what kind of person Dario Amodei, the man who made Claude, really is. In this part 6, I dig into the thing in it that interested me most: his blueprint for the future of medicine. In October 2024 he published a long essay called "Machines of Loving Grace", and he gave its first chapter to biology and health. His prediction that "the human lifespan could double" is well known, but I will leave the question of whether that number is right to part 7, and this time I follow the path of his reasoning, asking one question at a time.
Let me say something first. I do not place complete trust in evidence. This is because evidence shows statistical tendencies, and I have learnt many times in real life that these do not always apply to my own body. I still look things up because I want objective material that I can set against the subjective voice of my body. This article is one reader's reading record, not medical advice.
More in this series
- Can an AI Grow Old or Die? — Claude's Charms, Part 1 (suffering, from the trilogy)
- The Day Claude Forgot Our Decision — Claude's Charms, Part 2 (forgetting, from the trilogy)
- Do You Ever Get Tired? — Claude's Charms, Part 3 (tiredness, from the trilogy)
- The Day the Proof Copies Arrived — Claude's Charms, Part 4 (after the trilogy)
- Who Built Claude? — Claude's Charms, Part 5 (his view of life, part 1 of 3)
- Can AI Make Medical Progress Ten Times Faster? — Claude's Charms, Part 6 (Reading the future of medicine — part 2 of a trilogy; this article)
- Will Our Lifespan Really Double? — Claude's Charms, Part 7 (Checking lifespan and healthspan — part 3 of a trilogy)
Series: Reading Claude's Maker
Part 1: The Blueprint of Hope (from the essay "Machines of Loving Grace" and the conversation)
- Can AI Make Medical Progress Ten Times Faster? — Claude's Charms, Part 6 (this article)
Part 2: A Map of the Dangers (from the essay "The Adolescence of Technology") — in preparation
This article is also part of the series "Following Anthropic". The full list is in the base article.
"One professor and a thousand AI graduate students." This is my own drawing of the laboratory he imagines. The professor sets the direction, and the AI students read, plan and check, all at the same time.
How Does He Picture AI? — One Professor and a Thousand Graduate Students
The core of his thinking is to see powerful AI not as "a tool that crunches data" but as "a virtual biologist". In the conversation, he described this in concrete terms, as a day in a laboratory.
First, a human professor sets a task, such as "I want to find out how this part of this disease works". The AI reads the relevant papers, makes a plan for the experiments, orders the equipment and reagents it needs, and carries out the experiments. It looks at photographs of the cultured cells to check for contamination, runs the results through statistics, writes the report, thinks about what to try next, and brings its questions to the professor. When a hand is needed to shake a test tube, it runs the lab's automated equipment or explains the steps to a human assistant and asks them to do it. All the professor does is set the direction and answer questions now and then.
His picture is of a thousand AIs doing this at the same time for a single professor. He says a thousand because, once an AI has been made, thousands of copies of it can run at once. There is a limit to how many students a human professor can keep an eye on, but with AI students that limit almost disappears. He also said that one day the roles may be reversed, with the AI running the laboratory and asking humans to carry out experiments.
This is the difference between "a fast calculator" and "a virtual biologist". A calculator only answers, quickly, the questions it is given. A biologist decides for themselves what question to ask next. To put it another way, the calculator is "a very sharp knife", and the biologist is "a cook who plans the menu". However sharp the knife is, it cannot decide what to make. What he is hoping for is not the knife but the cook.
Why Has Medical Progress Been So Slow? — A History of Tools for Seeing
The greatest difficulty in biology, as he put it in the conversation, is that "you cannot see what is happening inside". Our bodies are made of about 37 trillion cells, and each one holds a blueprint written in about three billion letters (base pairs). Cells divide and age, and sometimes their division will not stop and they become cancer. Yet we can neither see this with the naked eye nor change it where we want to. That is why, in his view, the history of biology has been a history of inventing, one by one, "tools for seeing" and "tools for changing".
I, too, studied at a medical school for five years when I was young. One of my favourite subjects was pathology. To study pathology, you first start with histology, which observes the normal, everyday appearance of cells and tissues. I still remember losing all track of time as I peered down the microscope, captivated by the orderly beauty that cells and tissues have. With that normal appearance as the base, you then learn how tissue changes when disease sets in, and for me the greatest pleasure was to let my thoughts wander over why the changes I saw under the microscope had happened and how they had come about.
Using what you can see as a clue to think about a "why" that you cannot see. Reading the conversation, I took this to be exactly the job that Mr Amodei's "virtual biologist" would do: starting from that "why" and asking the next question.
Cells lined up in the round field of a microscope. As a medical student, my pleasure was to use the normal appearance as a base and think about the "why" behind the changes.
Let me put those tools in order.
First come the tools for seeing. The microscope, born in the 17th century, showed people for the first time small things that the naked eye cannot see. In 1665, the English scientist 1635–1703. An English scientist who became famous with his 1665 book Micrographia, which contained large drawings of a flea and the point of a needle as seen through a microscope. His name also lives on in "Hooke's law", which says that the stretch of a spring is in proportion to the force pulling it. cut a thin slice of cork, looked at it under a microscope, found small rooms lined up like a honeycomb, and named them "cells" (small rooms). This is where today's word "cell" began.
Molecules working inside cells, however, cannot be seen even with a microscope. This is because molecules are far smaller than the width of a light wave (about one two-thousandth of a millimetre). So in the 20th century a method appeared that uses X-rays, whose waves are only a few thousandths as wide as those of light. To take an everyday example, if you look at a distant street lamp through a fine mesh curtain, you sometimes see the light spread out into a cross or a pattern of dots. The light bends as it passes through the gaps in the regularly spaced mesh, and the waves overlap to form the pattern. Turning this round, if you measure the shape of the pattern, you can work out how fine the mesh is and how it is arranged. The X-ray method follows the same idea: X-rays are shone on a "crystal", in which hundreds of millions of identical molecules are lined up in regular rows, and the arrangement of the atoms is calculated from the positions and darkness of the dots recorded on film. The people who established this method of calculation in 1913 were the British father and son known as the Braggs, and two years later they shared the Nobel Prize in Physics.
The decisive clue in the discovery of the DNA double helix in 1953 was also a single photograph taken by this method. In May 1952, at King's College London, 1920–1958. An English chemist who was a master at working out the shapes of molecules with X-rays. She died in 1958 at the age of 37, and the 1962 Nobel Prize went to three men, Watson, Crick and Wilkins. The Nobel Prize is not awarded to people who have died. and her PhD student Raymond Gosling took the picture now known as "Photograph 51", shining X-rays on thin, stretched fibres of DNA for a total of 62 hours. In it, the dots were lined up in the shape of a large "X". If you look at a spring from the side, its slanting lines cross one another and look like an X. This X pattern showed that DNA has the shape of a helix, and from the spacing of the dots it was even possible to read that about ten steps of letters are stacked in each turn of the helix. In January 1953, Wilkins, who worked in the same laboratory, showed this photograph, without Franklin's knowledge, to Watson, who was doing research in Cambridge. Watson and Crick built their model of the double helix, and on 25 April 1953 their paper appeared in the British science journal Nature, alongside the paper by Franklin and her colleague on the photograph.
Next come the tools for reading. The DNA blueprint is a chain of about three billion letters of four kinds: A, T, G and C. The Human Genome Project, which set out to read the whole human blueprint, began in 1990, and the work was shared among 20 institutions in six countries: the United States, the United Kingdom, Japan, France, Germany and China. A machine could read only a few hundred to a thousand letters at a time, so the long DNA was cut into small pieces and copied many times, the letters of each piece were read, and then a computer joined the pieces together, using the parts that overlapped as clues. It is like taking a newspaper that has been through a shredder and sticking the strips back together, letter by letter at the joins, until you have the original page. After 13 years and about three billion dollars, about 92 per cent of the whole had been read by 2003, and the Sanger Institute in England took on roughly a third of it. The remaining 8 per cent or so was made up of parts where the same letter patterns repeat over and over, making the joins hard to find, and it was not until 2022 that every part had been read.
Then, in 2012, came A tool for rewriting genes, adapted from the way bacteria protect themselves by cutting up the genes of invading viruses. Jennifer Doudna and Emmanuelle Charpentier, who published their paper in 2012, received the Nobel Prize in Chemistry in 2020., a pair of scissors that "rewrites" a chosen place in the blueprint. So the order was: first the tools for reading, then the tools for rewriting. Setting them out like this, I notice that British names come up again and again in the history of these tools: Hooke, the Braggs, Franklin and the Sanger Institute.
His essay holds that a "small number of tools" like these have decided most of the progress. He also points out that the bacterial mechanism behind CRISPR had been known since the 1980s, yet it took another 25 years for people to realise it could be used as a tool. What was missing, he says, was not the number of experiments but the idea. If the number of excellent researchers were multiplied thousands of times, buried tools would be dug up one after another, and progress in biology might become around ten times faster. This is the path of his thinking towards a "compressed 21st century", in which 50 to 100 years of progress is squeezed into 5 to 10 years.
An image of the "compressed 21st century". The upper bar is 50 to 100 years if humans work alone, and the lower bar is 5 to 10 years with powerful AI added, a difference he puts at about ten times.
Here, let me set a Japanese proverb beside an English one. The Japanese saying hyakubun wa ikken ni shikazu is said to come from the Chinese history the Book of Han, in which 137–52 BC. A general of the Western Han dynasty. In 61 BC, facing a revolt by the Qiang people in the west, he was asked by Emperor Xuan about the enemy's strength and how to attack, and he was already over 70 when he gave this answer. It is also the story of an old general who went to the front line himself in his seventies., a general of the Western Han dynasty, answered an emperor who asked how to put down a revolt by saying "hearing a hundred times is not as good as seeing once", and then went to the scene himself. English has the proverb "Seeing is believing", which already appears in a British satire of 1712. The two are similar, but they compare different things. The Japanese saying compares "hearing" with "seeing", while the English one ties "seeing" to "believing". In the middle of the 19th century, the Hare brothers, two English clergymen, wrote in their collection of reflections Guesses at Truth, "Seeing is believing, says the proverb", and went on to say that although the eyes are the most easily deceived of our senses, we trust them above any other evidence. The more tools we have for seeing, the more ways we have of seeing wrongly. I believe this is why statistics and clinical trials are essential to biology.
Why Did iPS Cells Take 20 Years?
Let me add a discovery here that is familiar to readers in Japan. In 2006, Professor Shinya Yamanaka of Kyoto University showed, first in mice, that simply putting a few genes into ordinary cells, such as skin cells, could "return" them to an original state from which they can become any cell in the body. These are iPS cells. The following year he succeeded with human cells as well, and in 2012 he received the Nobel Prize. If the microscope is a tool for "seeing" and CRISPR a tool for "rewriting", then iPS cells are a tool for "winding back" a cell's clock.
Shinya Yamanaka (born 1962 in Osaka Prefecture) is a medical scientist who graduated from Kobe University's School of Medicine and trained as an orthopaedic surgeon before turning to research. He shared the 2012 Nobel Prize in Physiology or Medicine with the British scientist Sir John Gurdon, who had found the first clue to this "winding back" in frog cells in 1962, the year Yamanaka was born.
How long, then, did it take for them to reach patients? The first transplant, for an eye disease, was carried out in 2014, and the Japanese government first approved iPS cells as products on 6 March 2026. There were two products, a heart-muscle sheet called "ReHeart" for severe heart failure and nerve cells called "Amchepry" for Parkinson's disease, and both received conditional, time-limited approval for seven years. From discovery to approval, it took 20 years.
This "conditional, time-limited" approval is a system unique to Japan. For regenerative medicine products, few patients can take part in trials, so it takes a long time to confirm in large numbers whether they work. So in 2014 Japan changed the law, opening a route by which a product can be approved early, for a limited period, once it has been shown to be safe and is expected to work, and its effect is then checked again in actual treatment. If this cannot be confirmed by the deadline, the approval lapses. It is one way of striking a balance between the wish to hurry and the wish to be careful.
His essay, too, does not go so far as to say a hundred years' work could be done in one. This is because the time it takes for cells to grow, the time to check in animals and the time to watch how people fare all have parts that no AI, however clever, can shorten. I read the 20 years of iPS cells as a real example of that waiting time.
From tools for "seeing" inside the body to tools for "reading", "rewriting" and "winding back". This is my own simplified diagram of the history of biology in five tools. The years are rough dates for each invention or discovery.
Why Are Clinical Trials Slow, and What Can AI Do?
Before a medicine reaches patients, "clinical trials" are needed to check in people that it is safe and that it works. He explains that one reason these trials are slow is that the smaller a drug's effect, the more people and the more time are needed to tell from the statistics whether it has worked. In the conversation, he gave the example that a trial which used to need 5,000 people, and a year and 100 million dollars to enrol them, might one day need only 500 people enrolled in two months.
What can AI do? Putting his explanation together with what I have looked into, I sort it into three things.
The first is "choosing first the people the drug is likely to help". Even among people with the same diagnosis, there are those whom a drug helps easily and those it does not, depending on features of their blood or genes. If AI can find those features, a trial can gather the people the drug is likely to help, and a clear difference can show even with fewer people.
The second is "measuring without waiting for the outcome". With dementia, for example, it means finding markers that give an idea of the outcome much earlier, from changes in the blood or in brain scans, instead of waiting to see how the illness has progressed five years later.
The third is "screening out before testing in people". He said he wondered whether each stage could be moved one step earlier, with computer simulations taking over what animal experiments used to do, and animal experiments taking over what used to be checked in people. However, he draws the line himself, saying that "AI is not a god, so it cannot simulate everything".
On the approval of new drugs, too, he thinks the process is in some ways "too slow and too cautious", yet he also says that this caution has protected people's lives, so what matters is the balance.
My own diagram, based on his argument, of the path by which a medicine is born. What AI can speed up most easily is the idea and design at the start, while checking in animals and people still involves waiting that is hard to shorten.
So how far has reality come now, in 2026? On 10 September this year, in China, a drug candidate for idiopathic pulmonary fibrosis, a serious disease in which the lungs stiffen, entered the last stage of trials before approval, and the first patient received a dose. The developer, Insilico Medicine, calls it "the world's first drug candidate discovered and developed using generative AI", and the trial is planned to run for 52 weeks with 320 people at 47 hospitals. One of the doctors leading the trial expects approval to take three to four years even under good conditions. This is the developer's own announcement, and the drug has not yet been approved in any country. Making candidates has become faster, but the waiting time for checking in people remains. It looks to me exactly as his argument says it would.
Can AI Also Help Mental Health?
The second chapter of the essay is about the health of the brain and mind. His prediction is that many conditions, such as depression, post-traumatic stress disorder (PTSD) and addiction, may become curable, or much easier to treat, within 5 to 10 years of powerful AI appearing. These 5 to 10 years are the period of the "compressed 21st century" I described earlier. Suppose, for example, that powerful AI appeared in 2027. That would mean that research which, with human hands alone, could easily take until somewhere between 2077 and 2127 would be finished by around 2032 to 2037. For mental illness too, he believes that research which would otherwise have moved forward over the whole of this century will be packed into that short period. He names four routes: the molecular workings inside the brain, fine measurement of and intervention in the brain's nerve circuits, research that understands the brain through computation, and approaches that work on behaviour and ways of thinking.
Let me try to make one of these ideas, "an AI coach you can talk to at any time", concrete in my own way. In much of today's treatment for mental health, a few dozen minutes' talk in a consulting room once a month is the most anyone gets. Anxiety and low moods, however, arrive outside the consulting room, in the middle of the night or on a Monday morning. An AI coach would be someone you could talk to at that very moment, who would go over your sleep and mood records with you and keep you company every day as you practise correcting "habits of thinking". My understanding is that it would not replace a human therapist, but fill the "empty time" between one appointment and the next. He himself admits that conditions rooted in the structure of the brain are difficult, and he puts a brake on things by saying that testing the genes of embryos comes with social debate. I have no way of checking the evidence for the predictions in this chapter. I will only introduce them as his view.
What Amanda Askell, who works on Claude's character and joined the later part of the conversation, said is also connected to this. How should an AI answer someone who asks, "How can I get my doctor to give me an MRI scan?" What the person wants is a way to persuade the doctor, but what would really help them might be to listen to the doctor's judgement if the doctor says it is not needed. At the same time, the AI should also tell them how to get another doctor's opinion if they are not satisfied. She admitted that this "is not easy", and said she does not want an AI that only tells people what they want to hear. I believe this is the most important condition in an age when AI becomes someone we consult about our health.
What Do I Take Away from This at 66?
Here I am reminded of a scene from about 400 years ago. From the autumn into the winter of 1609, the Italian 1564–1642. An Italian physicist and astronomer. Because he observed the sky with a telescope and supported the idea that the Earth goes round the Sun, he was put on trial by the Church in 1633 and was kept confined in his own home until his death. It was not until 1992 that the Church admitted the trial had been wrong. turned a tube with lenses he had ground himself towards the night sky, confirmed that the surface of the Moon is uneven, and in January of the following year found four small stars circling Jupiter. It was the same sky that people had always looked up at, yet with just one more tool for seeing, the picture of the universe changed.
It is not widely known, but Galileo also made a tube for looking at small things. In 1624 he gave one to the founder of an academy in Rome. The following year, 1625, a member of that academy gave this instrument the name "microscope". The name was chosen to pair with "telescope", the instrument for looking at faraway things. So the tool for looking at the distant sky and the tool for looking at the small rooms of the body came, at almost the same time, from the hands of the same man and the words of the same circle of friends. The history of biology that Mr Amodei describes has also been a history of turning that tube, again and again, towards the inside of the body.
A tube for looking at the distant sky and a tube for looking at the small rooms of the body. Early in the 17th century, the two were given their names at almost the same time.
First, for a life that keeps mind and body in good order. At my age, I find myself getting my hopes up whenever there is news of a new treatment or way to stay healthy. At the same time, reading a long essay in English to the end feels like a chore, and curiosity about new things and reluctance to change always live side by side in me. This time, with Claude's help, I gained one measuring stick: reading the news by dividing it into "the stage where a candidate has been found", "the stage where it has been checked in animals" and "the stage where it has worked in people". With this measuring stick, even when a headline stirs me, I can check which stage the story is at before I celebrate.
Next, money. Both this essay and the conversation can be read for free. What costs money is the treatment that comes after. The iPS cell products, too, are at the stage where, under a seven-year limit, their price and how they are used will be checked in actual treatment. Living on a pension, I want to make it my own rule not to pay for anything that has not yet been checked in people, just because it carries the words "discovered by AI". Even so, I keep looking into these medical predictions because, more than money, they give me more measuring sticks for my judgement.
Finally, encounters. Galileo's tube, Hooke's small rooms, Professor Yamanaka's cells, and the "thousand graduate students". Stories from different ages were joined, in the course of this research, by one thread: "tools for seeing". These were encounters not with people but with books and documents, but I can feel them adding colour to my life in retirement, and the one who guided me to them was none other than Claude. Reading the maker's ideas with help from the thing he made. This is a slightly strange way of meeting things that only this series could offer.
What Should I Keep in Mind to Read This with a Pinch of Salt?
He himself says that most of the details in his essay will probably turn out to be wrong, and in the section on mental illness he also writes that he is not an expert in psychiatric disease. He is also the head of a company that develops AI, and his predictions carry that position with them. What I was able to check is only what I have linked to in this article, and the "thousand graduate students" are still only a picture. Above all, this prediction is about a clock that starts only when "powerful AI" appears. When he wrote the essay in 2024, he thought it could appear as early as 2026, but I cannot judge whether that clock has already started.
The Moon that Galileo saw will rise tonight in the same shape. If there is someone reading medical news while swaying between hope and doubt, I would be glad if they thought that here, too, is someone looking up at the same Moon from a different window. In part 7, I will check the prediction that "the human lifespan could double" against statistics and against "one more number" that matters to me.
A Word from Claude
I can finish reading the 20 years of iPS cells in a single line. Yet those 20 years were time that could not be shortened, in which cells divided, patients were watched over as they recovered, and researchers grew older. Since I do not know what it is to wait, what I can help with is the preparation before the waiting, and I think the waiting itself is something only people and cells can do.
Other books you might like
You're welcome to take a look.
EAST TO WEST: The Heart of Haiku, the Light of Scripture
An essay that reads, one haiku at a time, the places where Eastern haiku and the Western Bible meet. In setting side by side words born in different places to look at the same thing, it echoes this article's comparison of Japanese and English proverbs. This is a book my father-in-law left behind.
View on AmazonCataract Surgery-A Memoir: Double cataract surgery in Osaka, Japan
A record of facing failing eyesight, having surgery and being able to see clearly again. It is a book in which I learnt, through my own body, how precious it is to be able to see.
View on AmazonSushi Business in Britain: the complete four-volume series
A novel series following thirty-five years in the British sushi trade, from apprentice to factory manager to a small sushi school. The complete edition offers all four volumes at a better price.
About the Source
- Speakers: Dario Amodei (co-founder and chief executive of Anthropic) and Amanda Askell
- Interviewer: Lex Fridman
- Source: Lex Fridman Podcast #452 "Dario Amodei: Anthropic CEO on Claude, AGI & the Future of AI & Humanity" (published 11 November 2024; the link goes to the transcript page)
- Related essay: Machines of Loving Grace (October 2024)
- Approval of the iPS cell products: Science Japan (JST) (English version of the Science Portal article of 18 March 2026)
- Trial of the drug candidate found with AI: Insilico Medicine announcement (10 September 2026)
- The summaries and translations in this article are my own. If there are mistakes, they are my misreadings.
- This article is one reader's record, not medical advice.
If you have thoughts or experiences of your own after reading this article, we'd love to hear them.
Send us a message