Dario Amodei, head of AI company Anthropic, told investors that cancer, Alzheimer’s disease, heart disease, and mental disorders would become “relics of the past” in less than ten years
Dario Amodei, head of AI company Anthropic, told investors that cancer, Alzheimer’s disease, heart disease, and mental disorders would become “relics of the past” in less than ten years
On November 10, Anthropic released a 42-minute video for investors ahead of its stock market listing. That same day, cardiologist Eric Topol responded that curing these diseases within ten years was impossible, while biomedical researcher Avi Roy explained why testing drugs in humans takes years.
At the forty-minute mark in the video, Dario Amodei says:
“If we look at things like cancer, Alzheimer’s disease, heart disease, and mental disorders today, I think they will become relics of the past in less than ten years.”
He made a similar prediction in an essay. In “Machines of Loving Grace”, he defined “powerful AI” as a system smarter than a Nobel laureate in most fields and suggested that biology and medicine could make 50–100 years of progress in the five to ten years after its arrival. In his June essay, he wrote that such a system would probably emerge if AI capabilities continued to improve for another year or two.
In the first essay, he attributes much of the time spent on clinical trials to drugs with small effects: detecting a survival benefit of a few months requires enormous patient groups, whereas a drug with a large effect could reach approval in about a year. In the June essay, he puts the typical approval process through the FDA and the European regulator at 7–8 years. He writes that AI will overwhelm this system unless it is reformed and proposes establishing rules in advance for regulators to accept AI simulations in place of some experiments and surrogate endpoints in place of disease outcomes such as heart attacks.
Greg Meyers of pharmaceutical company Bristol Myers Squibb, an Anthropic client, says in the same video that AI will help shorten clinical trials by 20–30%. He describes bringing a drug to market two to three years earlier as a gain.
Topol responded on X to a post that paraphrased the promise as curing these diseases within ten years:
“This sets an unattainable goal. I wish it could happen, but curing all these diseases in the next ten years is impossible. And I am very optimistic about AI’s potential to improve health.”
Roy divides drug development into two halves. AI is already accelerating the first, finding a molecule: it predicts the structures of most proteins, a task that used to take laboratories months or years. The second, testing in humans, depends on the disease: two drugs that slow Alzheimer’s disease were tested for about 18 months because the disease progresses slowly, while a Novartis heart drug was tested for almost five years. “No computer can make a disease progress faster,” Roy writes.
The Novartis drug is pelacarsen. It lowers lipoprotein(a), a particle in the blood associated with the risk of heart attack and stroke, by 78% over a year. This level is a surrogate endpoint: it changes within a year, while disease outcomes take years to become apparent. The trial results were published on November 7: over almost five years, a heart attack, stroke, urgent procedure on the coronary arteries, or cardiovascular death occurred in 12,3% of those receiving pelacarsen and 12,1% of those receiving placebo, almost identical proportions.
Roy estimates that it takes about ten years from the first human dose to approval, with nine out of ten drugs failing along the way. If AI helps select targets (the molecules a drug acts on) more accurately, and two out of ten drugs succeed instead of one, the same number of attempts would yield twice as many medicines. His prediction for 2036 is that there will be more drugs for these diseases, each tested in humans; the diseases will become “relics” later.
- https://www.youtube.com/watch?v=eynv5VR9qos&t=2437s
- https://darioamodei.com/essay/machines-of-loving-grace
- https://darioamodei.com/post/policy-on-the-ai-exponential
- https://www.youtube.com/watch?v=eynv5VR9qos&t=804s
- https://x.com/EricTopol/status/2120302576398958633
- https://x.com/agingroy/status/2120282861634802086
- https://newsroom.heart.org/news/lp-a-horizon-trial-lp-a-fell-by-78-but-cardiovascular-outcomes-unexpectedly-neutral