>> In a 42-minute IPO roadshow video the company released on Tuesday, Amodei made the audacious prediction that "cancer, Alzheimer's, heart disease, mental illness — these will be relics of the past in, I think, less than a decade."
Mental illness? I'm sure that will be worked out exactly a day after AI-induced psychosis stops being a thing.
Question: Without heart disease, cancer and Alzheimer's, what will kill us? It cannot all be shark attacks and avalanches. What new horrible diseases will step up to be big killers in the future?
Probably really just brain decline, if you can grow new cells and organs without having to care about cancer. And even there, technology may be able to replace senescent cells at some point. But nobody really knows as even (multicellular, animal) model organisms cannot be kept alive forever today.
Lots of things in the body just stops function correctly at some point. Some old people today just die peacefully in their sleep without it being cancer or heart dicease.
Total organ failure. Even if you can cure any individual illness, eventually your body would accumulate enough imperfections due to cellular replication that a unstoppable collapse becomes more and more likely.
accumulated errors in DNA replication. telomere shortening. whatever other things cause aging.
we could eliminate all disease, but without solving senescence itself I don't think it's possible to achieve even 200 year lifespans. by 70, even the most blessed of us begin to rapidly fall apart.
I've yet to see anyone with expertise in medical research express optimism that AI will cure major diseases like cancer in the next several years. Help us in parts of the research? Sure. But the bottlenecks to medical research are the speed at which the human body does things, and AI isn't going to move that.
That also aligns with my (non-medical-expert) intuition.
Dario et al are smart enough to know that as well, and they have not presented any compelling counterargument to the speed of biology, which makes me they that they are full of shit and know it.
I kinda agree with you. One counter argument I could imagine though is that we make really rapid progress in figuring out low level biology in a first principles predictive way that might make us able to zero in on a solution directly.
> I've yet to see anyone with expertise in medical research express optimism that AI will cure major diseases like cancer in the next several years.
Depending on "several years" -> it is almost by definition impossible. The trial periods are too long.
A lot of major breakthroughs have happened recently and are already in the pipeline, looking promising.
There's a lot of progress on this stuff already. Compared to 10-20-30 years ago, a lot of diseases already look "cured", even if they aren't completely.
Near immediate death sentences have become 5-10+ years of life expectations, etc.
I don't see any reason why AI wouldn't help speed things up in the future.
But we won't see any of that for "several years" regardless.
The trials argument is a good one, but if a good cure is found, trials can be sped up. A lot of medication today has an impact that's hard to measure, so large and long trials are necessary. If a cure is so good that improvements can be seen immediately, I can imagine that they are greenlighted faster, especially if the alternative for patients is death, since what side effect could be worse than death.
> I've yet to see anyone with expertise in medical research
I mean... that's literally Amodei. But anywho.
We just went through this with coding: We are setting up a silly strawman, that will be goal posted to the end of the world. I am relatively certain Amodei does not literally mean "Will you flip AI on, go to sleep, and get a "cancer" pill at your door some day x in the future, all just AI?". Yes, it will involve humans.
It's what AI enables, in combination what exists. This might also include changes in regulation (which to me seems like the bottleneck, much more so than speed of the human body).
I remember 5 years ago when you couldn't find any programmer that would agree that AI would replace them. Fast forward to today and most programmers haven't opened an IDE in a year or so and hiring freezes all over the place.
I'm not saying both cases are equivalent, but A LOT can change in 10 years, specially since the acceleration is compounding over time.
Today we're using the absolute worst models that we're going to use in the next 10 years.
Ah yes, AI is better than we expected at some things so we should expect it to be good at everything. That's basically the "we're building a god" argument.
AI, it turns out, is very, very good at anything that can be verified on computers. We should expect it to continue to improve until it effectively saturates that (i.e., if a problem can be verified on computers in any reasonable amount of time, it's effectively solved).
The human body is not something that can be verified on computers.
> The human body is not something that can be verified on computers.
...yet.
In the end we're all made of genes, DNA sequences and proteins that can move around due to electric gradients, which are all governed by the laws of physics and can be verified mathematically.
My daughter is an expert on the novel non-invasive diagnostic methods side. She states that the bottleneck is that most diagnostic methods and treatments don’t work. And that’s because we have an incomplete understanding of the human body. AI make some small gains but most of the work is a lot of slow, laborious experimentation and measurement which can’t be automated in any reasonable way.
>I've yet to see anyone with expertise in medical research express optimism that AI will cure major diseases like cancer in the next several years
It's crazy how AI has completely reversed who is technology optimistic and who is not.
A 60-year-old immigrant medical researcher thinks we live in the most exciting time in history and that all major diseases will be cured within a decade.
A 29-year-old American zillennial genderfluid medical researcher thinks Dario and Sam Altman are demonic creatures, worse than the devil himself.
Whatever your opinion on whether we're there right now, there is nothing contradictory about this. The entire point of AI is that it is/can be/will be (at the limit) incredibly powerful. That power can be used for good or bad things.
Which one? All of them? The mechanisms of neoplastic development and growth are protean. If his prediction is based on aspirations of finding a common pathway, I’m afraid it will be unrealized.
I share his enthusiasm for assisted drug development and so forth; but let’s be realistic. Besides, doesn’t his company have an upcoming IPO? I guess it’s a fine time for unbridled optimism then, isn’t it?
I don't actually think that he's overselling that much. Of course there is no guarantee that all these illnesses will be fixed, but imagine how much more progress can be made if we set up one million automated wetlabs performing research on these sicknesses. I had an agent (irony) estimate that there are around 100k such labs around right now, and these labs are not used around the clock. So 1 Mio automated wetlabs count more like 3 Mio conventional wetlabs. If we now factor in that current models already have much better capabilities to access existing knowledge, and that they may in fact be(come) smarter than the average researcher, that factor increases.
I really expected better arguments to be put forth by the community, however that hasn't really happened yet. Across the few HN threads this has popped up, the general stance is:
"In other fields [coding, math] the experts said Amodei was wrong. He proved to be right. In this field experts are saying Amodei is wrong, therefore he will be proven to be right."
That's as simple of a summary as I can given without getting into the details, which are irrelevant and devolve into nit-picking regarding specific wording used (can vs. will), who said and what was really said (or was not said), and so on.
To me, that argument is not good enough. I would like to hear something a bit more step-by-step as to why people think he is right or wrong that's beyond mere "because he said so" takes.
It is very hard to appreciate what exponentials look like in reality. We're on track to have a GPT-3-to-Mythos sized jump in a single year in 2027 or 2028, and its not as if it's going to stop after that year.
People have a hard time believing that's going to happen, so they don't believe we will soon "solve" robotics, and so they don't believe robotics can really run an entire biology lab, and so they don't believe we'll have industrical scale data-center sized labs running at high speed 24/7, and so they don't believe that the lab bottleneck can be broken.
IMO the problem now is that the human body is an extremely complex system and we only understand some aspects of it. The way we tend to make progress with drugs is to understand one or two aspects that are relevant to a disease, and then make a molecule that affects those, and then try it out.
When we try it out though we very often are surprised that there is often some other regulatory system that nullifies the effect, or there is some other unintended consequence of the intervention, because we do not have a complete understanding of the system going in.
The trying out part is what takes so much time and money, whether that’s in vitro, in animals, or in a human clinical trial.
If we had a far better understanding of biology, we could predict much better what effects a certain molecule would have. Also, if we found a way to try things out much more quickly and cheaply, we could accelerate progress.
I don’t really know what to believe anymore, nor do I feel like I have the time to think through what I should believe, but yesterday I was reflecting on how deeply crazy it sounded when Dario called “AI will write all code in 6 months” early last year. Granted, Opus 4.5 was released 8 months later and his “90% in 6 months, 100% in 1 year” ended up being more like “90% in 1 year, 100% in 18 months”, but it still feels crazy to look back on this claim from March 2025 and feel like it was an entirely different world.
Does this translate to medical breakthroughs? I don’t know! I’m not a medical researcher. But I do know that using my software background in combination with AI allows me to do work with speed and rigor that was never possible before.
Since I could not find a job as a software engineer in this market (even with a decade of experience), I pivoted into policy research and now work as a policy director at a pro-homes advocacy organization. The fact that I can build bespoke software to answer specific questions, or analyze specific data, or communicate a specific idea is transformational, and I feel like we’ve just scratched the surface of what’s possible. We have been able to do an incredible amount of very high impact work in a short amount of time, and I feel like my ability to leverage AI capability for this work is still on the exponential. Our capability as a very small team with a limited budget is genuinely expanded, just by looking at the projects that I can do on my own now with Claude and a week or two of my time, that would have instead been a $25k 6-month contract that we have to raise money for and hire out.
Is policy research the same as medical research? Also no, and I know that trials add a significant lag, but I’m just not convinced that every other part of the process wont get faster, and that new capabilities to expedite research and analysis will make the time spent more productive. It doesn’t feel like a “sure thing”, but at this point, it feels like the riskier option to bet against the idea of “significant medical breakthroughs within a decade”.
One of my core pieces of advice to AI leaders who get interested in biology is "do not say you will cure cancer, the medical establishment will immediately write you off as clueless".
Instead, I recommend AI people say "we believe our new technology will be able to rapidly make significant improvements to the detection and treatment of cancer. We will work closely with medical researchers to help them use our tools effectively."
When I did my own cancer research ~25 years ago, the only treatments were basically chemo or radio therapy. The outcomes were often terrible- even for people who lived, they suffered terribly. We are now in a state where new technologies are making absolutely amazing improvements in treatment- with far fewer side effects, which greatly increases quality of life in people undergoing the treatment, and after. I think if we keep grinding on this problem (as well as with many other diseases), we could very well make most cancers a non-death sentence that can be treated routinely with minimal side effects.
Here is a thing that did not exist when I was a PhD student: a person with cancer goes to the doctor, they take a biopsy of the tumor and send it to a lab, which sequences the cancer genome. That genome is then input to an AI which understands a great deal about cancer immunology, and spits out a custom antibody which is then injected back into the person, and that antibody teaches the person's immune system to fight the cancer more effectively. There literally is an AI model (my coworkers developed one that is in clinical trials now) integrated in the treatment process, which is absolutely necessary for the high effectivity of the treatment. That knowledge alone inspires me to do my best work, with the belief that it will have positive outcomes.
There's another open question- to what extent do we need to understand the underlying biology of cancer, versus having models that integrate enormous numbers of variables effectively to come up with treatments without a mechanistic model. It's quite clear we really do not understand how drugs work (no matter how much publications claim we do), and probably by improving methods by which we can sample the enormous space of potential drugs/treatments.
And finally: how can we made this affordable? Cancer treatment in the US is obscenely expensive with many levels of rent seekers who constantly add unnecessary cost to treatment. That will likely require aggressive and intelligent legal regulation (clearly the market is not efficient), and I don't see any sort of political will to make it happen.
58 comments
[ 2.8 ms ] story [ 29.1 ms ] thread>> In a 42-minute IPO roadshow video the company released on Tuesday, Amodei made the audacious prediction that "cancer, Alzheimer's, heart disease, mental illness — these will be relics of the past in, I think, less than a decade."
Mental illness? I'm sure that will be worked out exactly a day after AI-induced psychosis stops being a thing.
Question: Without heart disease, cancer and Alzheimer's, what will kill us? It cannot all be shark attacks and avalanches. What new horrible diseases will step up to be big killers in the future?
Claude will cure old age too, of course.
we could eliminate all disease, but without solving senescence itself I don't think it's possible to achieve even 200 year lifespans. by 70, even the most blessed of us begin to rapidly fall apart.
That also aligns with my (non-medical-expert) intuition.
Dario et al are smart enough to know that as well, and they have not presented any compelling counterargument to the speed of biology, which makes me they that they are full of shit and know it.
Maybe, or maybe someone needs to send him this comic:
https://phdcomics.com/comics/archive.php?comicid=1162
Depending on "several years" -> it is almost by definition impossible. The trial periods are too long.
A lot of major breakthroughs have happened recently and are already in the pipeline, looking promising.
There's a lot of progress on this stuff already. Compared to 10-20-30 years ago, a lot of diseases already look "cured", even if they aren't completely.
Near immediate death sentences have become 5-10+ years of life expectations, etc.
I don't see any reason why AI wouldn't help speed things up in the future.
But we won't see any of that for "several years" regardless.
I mean... that's literally Amodei. But anywho.
We just went through this with coding: We are setting up a silly strawman, that will be goal posted to the end of the world. I am relatively certain Amodei does not literally mean "Will you flip AI on, go to sleep, and get a "cancer" pill at your door some day x in the future, all just AI?". Yes, it will involve humans.
It's what AI enables, in combination what exists. This might also include changes in regulation (which to me seems like the bottleneck, much more so than speed of the human body).
I'm not saying both cases are equivalent, but A LOT can change in 10 years, specially since the acceleration is compounding over time.
Today we're using the absolute worst models that we're going to use in the next 10 years.
AI, it turns out, is very, very good at anything that can be verified on computers. We should expect it to continue to improve until it effectively saturates that (i.e., if a problem can be verified on computers in any reasonable amount of time, it's effectively solved).
The human body is not something that can be verified on computers.
Will we get that before or after we cure [all] cancer?
In the end we're all made of genes, DNA sequences and proteins that can move around due to electric gradients, which are all governed by the laws of physics and can be verified mathematically.
Amodei says whatever keeps the money rolling in.
It's crazy how AI has completely reversed who is technology optimistic and who is not.
A 60-year-old immigrant medical researcher thinks we live in the most exciting time in history and that all major diseases will be cured within a decade.
A 29-year-old American zillennial genderfluid medical researcher thinks Dario and Sam Altman are demonic creatures, worse than the devil himself.
How did this happen?
”It's literally too dangerous for us to release"
Stock market goes down
"Ai will cure cancer everyone!"
https://www.youtube.com/watch?v=kq-UwilTarU
"One person (who doesn't know the future but has a financial incentive to promise golden age) promises golden age.
Other people (who also don't know the future but don't want to lose their jobs) predict that they won't lose their jobs.
Both sides can be utterly wrong, so we spend our time in vain here".
Which one? All of them? The mechanisms of neoplastic development and growth are protean. If his prediction is based on aspirations of finding a common pathway, I’m afraid it will be unrealized.
I share his enthusiasm for assisted drug development and so forth; but let’s be realistic. Besides, doesn’t his company have an upcoming IPO? I guess it’s a fine time for unbridled optimism then, isn’t it?
> I guess it’s a fine time for unbridled optimism then, isn’t it?
To be fair, they've kept up this fairy-tale optimism most of the way through
"In other fields [coding, math] the experts said Amodei was wrong. He proved to be right. In this field experts are saying Amodei is wrong, therefore he will be proven to be right."
That's as simple of a summary as I can given without getting into the details, which are irrelevant and devolve into nit-picking regarding specific wording used (can vs. will), who said and what was really said (or was not said), and so on.
To me, that argument is not good enough. I would like to hear something a bit more step-by-step as to why people think he is right or wrong that's beyond mere "because he said so" takes.
People have a hard time believing that's going to happen, so they don't believe we will soon "solve" robotics, and so they don't believe robotics can really run an entire biology lab, and so they don't believe we'll have industrical scale data-center sized labs running at high speed 24/7, and so they don't believe that the lab bottleneck can be broken.
Eli Lilly CEO pushes back on Dario Amodei's ludicrous medical claims
https://news.ycombinator.com/item?id=50483480
When we try it out though we very often are surprised that there is often some other regulatory system that nullifies the effect, or there is some other unintended consequence of the intervention, because we do not have a complete understanding of the system going in.
The trying out part is what takes so much time and money, whether that’s in vitro, in animals, or in a human clinical trial.
If we had a far better understanding of biology, we could predict much better what effects a certain molecule would have. Also, if we found a way to try things out much more quickly and cheaply, we could accelerate progress.
I don’t really know what to believe anymore, nor do I feel like I have the time to think through what I should believe, but yesterday I was reflecting on how deeply crazy it sounded when Dario called “AI will write all code in 6 months” early last year. Granted, Opus 4.5 was released 8 months later and his “90% in 6 months, 100% in 1 year” ended up being more like “90% in 1 year, 100% in 18 months”, but it still feels crazy to look back on this claim from March 2025 and feel like it was an entirely different world.
Does this translate to medical breakthroughs? I don’t know! I’m not a medical researcher. But I do know that using my software background in combination with AI allows me to do work with speed and rigor that was never possible before.
Since I could not find a job as a software engineer in this market (even with a decade of experience), I pivoted into policy research and now work as a policy director at a pro-homes advocacy organization. The fact that I can build bespoke software to answer specific questions, or analyze specific data, or communicate a specific idea is transformational, and I feel like we’ve just scratched the surface of what’s possible. We have been able to do an incredible amount of very high impact work in a short amount of time, and I feel like my ability to leverage AI capability for this work is still on the exponential. Our capability as a very small team with a limited budget is genuinely expanded, just by looking at the projects that I can do on my own now with Claude and a week or two of my time, that would have instead been a $25k 6-month contract that we have to raise money for and hire out.
Is policy research the same as medical research? Also no, and I know that trials add a significant lag, but I’m just not convinced that every other part of the process wont get faster, and that new capabilities to expedite research and analysis will make the time spent more productive. It doesn’t feel like a “sure thing”, but at this point, it feels like the riskier option to bet against the idea of “significant medical breakthroughs within a decade”.
Instead, I recommend AI people say "we believe our new technology will be able to rapidly make significant improvements to the detection and treatment of cancer. We will work closely with medical researchers to help them use our tools effectively."
When I did my own cancer research ~25 years ago, the only treatments were basically chemo or radio therapy. The outcomes were often terrible- even for people who lived, they suffered terribly. We are now in a state where new technologies are making absolutely amazing improvements in treatment- with far fewer side effects, which greatly increases quality of life in people undergoing the treatment, and after. I think if we keep grinding on this problem (as well as with many other diseases), we could very well make most cancers a non-death sentence that can be treated routinely with minimal side effects.
Here is a thing that did not exist when I was a PhD student: a person with cancer goes to the doctor, they take a biopsy of the tumor and send it to a lab, which sequences the cancer genome. That genome is then input to an AI which understands a great deal about cancer immunology, and spits out a custom antibody which is then injected back into the person, and that antibody teaches the person's immune system to fight the cancer more effectively. There literally is an AI model (my coworkers developed one that is in clinical trials now) integrated in the treatment process, which is absolutely necessary for the high effectivity of the treatment. That knowledge alone inspires me to do my best work, with the belief that it will have positive outcomes.
There's another open question- to what extent do we need to understand the underlying biology of cancer, versus having models that integrate enormous numbers of variables effectively to come up with treatments without a mechanistic model. It's quite clear we really do not understand how drugs work (no matter how much publications claim we do), and probably by improving methods by which we can sample the enormous space of potential drugs/treatments.
And finally: how can we made this affordable? Cancer treatment in the US is obscenely expensive with many levels of rent seekers who constantly add unnecessary cost to treatment. That will likely require aggressive and intelligent legal regulation (clearly the market is not efficient), and I don't see any sort of political will to make it happen.
We might understand cancer better but like food, practical reality means once again sales moves past that pesky social reality.