On a summer afternoon 15 years ago, I met a privileged researcher outside a Stanford coffee shop to discuss our shared dream: using AI to diagnose cancer. She had soft hair, liked to talk with her hands, and a reputation for beauty. He worked in a research lab that developed early cancer screens; I, at the age of 20, had just learned that I carried a mutation that gave me an increased risk of breast, ovarian, and other cancers. Over the following years, he provided guidance on how to enter his profession, preparing me to apply for scholarships that would fund a Ph.D. mine, and warned me to stay away from cancer screening companies that made exaggerated claims.
But from there our paths diverged. I became a professor of AI. He founded Anthropic. My mentor was Dario Amodei, the man who leads one of the most powerful AI companies in the world. In the dream 2024 essay dubbed “Grace-loving Machines,” he predicted that a humanoid AI—smarter than Nobel Prize winners, using computers autonomously, and collaborating with millions of copies of itself—could soon supersede a century of scientific progress in a decade, and potentially reduce cancer deaths by 95 percent.
Which should sound good to me. At 35, my cancer risks are catching up with me. A few weeks ago, surgeons removed my ovaries, causing immediate menopause and ruining my ability to bear children. By 40, the risk of breast cancer for carriers of my mutation increases to one out of fourdouble the risk of life for the average woman. My mother, who also carries the mutation, was diagnosed with breast cancer at the age of 45. Now would be a great time in my life for super-intelligent AI to treat cancer.
Why, then, do I find myself rooting for delays in the development of this AI—hoping, in my heart of hearts, that GPT-6 will be a disappointment?
Part of the answer is that, despite the incredible pace of AI development, I don’t believe AI can cure cancer anytime soon—certainly not enough to bet my life on it. These doubts are shared by many AI experts in a investigation I recently advised, who generally expect slower progress than AI lab leaders. AI systems are most powerful in settings like chess, where they can generate unlimited data (for repeated play), experiment freely, and observe exactly what is happening. Many important systems, including math and coding, share these characteristics, and AI does it brought remarkable progress there. But cancer is different. Cancer data is limited and comes from biological experiments and clinical trials that cannot run at the speed of silicon. Freely experimenting on cancer patients would be unethical. And the cancer data only highlights the complex processes that our cells betray us. There are, in short, more obstacles to curing cancer than lack of intelligence.
Our minds existing ones The AI systems provide are also already terrible and underutilized. We have not yet fully utilized systems such as the Nobel Prize winner AlphaFoldwhich predicts protein structures with astonishing accuracy but has not done so not yet has been released revolution in drug development; or AI algorithms that match or call radiologists in many types of image analysis; or chatbots that are now helping scientists conduct research. My Ph.D. students used to write rules to analyze medical data; now they express their thoughts in plain English and let the AI do the rest. They work essentially like professors, limited only by their own ideas. A student of mine recently came to me full of excitement about AI-assisted medical discovery.
So as scary as the cure for cancer remains, I am sure that AI will contribute to it. And if curing cancer were the only result of building more powerful AI systems, I would be happy for their arrival. But the problem is that their effects are much wider, and we are moving very quickly to ensure that these effects are positive.
The recent chaotic release of Anthropic’s latest build, Fable 5, shows how unprepared we are to address the broader implications of these builds. Anthropic, fearing that the model could be misused to make biological weapons, at first he knelt his power to answer many basic biological questions, which the company said is a temporary step. This made the model, surprisingly, much less effective for cancer research than its less powerful predecessors. A few days later, the US government issued a national security order prohibiting foreign nationals from using the model, possibly out of concern that it could be used for cyber attacks. In response, Anthropic closed the fashion altogether. Reasonable people disagree about how dangerous this trend is and whether Anthropic or the government is overreacting. But clearly, our institutions are not remotely ready to respond to urgent deployments. (Anthropic did not respond to a request for comment about the publication of Fable 5, nor other questions.)
Many developers of these models, including Dario Amodei, agree that AI is developing faster than society is changing. The solution they propose is for society to speed it up, not for AI to slow it down, which they see as impossible; the very title of Amodei’s latest essay, “Policy on the Definition of AI,” focuses on the development of AI as an iron column on which society must bend. But moving forward will undoubtedly mean more of the kind of chaos that surrounded the release of Fable 5. More fundamentally, it will shorten our time to deal with the many social challenges that powerful AI can raise, including mass unemployment, rising inequality, oppressive surveillance, and wars of freedom. Each of these – and many others corresponding to their scope – are serious problems, no less important than curing cancer, for which we lack good solutions. It is not entirely clear that creating a global response to all these issues at breakneck speed is easier than reducing AI to the ground.
I myself am impatient with severity; since the day i learned i carried my change, i have lived with the constant awareness that life has an end. But I will wait a long time for a cure—even if it means losing my fertility and living under the shadow of danger—if it allows us to navigate this new world more carefully, and to make sure that, in curing cancer, we don’t lose the things that make cancer worth curing.
Of all the things we stand to lose, I worry perhaps most about how we will find meaning if we escape our own minds. Amodei struggles constantly with this question his essaycalling it “harder than the others.” I admire his attempt to deal with the question but find his answer unconvincing. “I spend a lot of time playing video games, swimming, walking outside, and talking with friends,” he writes in “Machines of Loving Grace.” But I doubt he would want to spend the rest of his life doing that only those activities – I certainly wouldn’t do it. He suggests that humans will still find meaning in deep mental activities, such as doing research, even if AI can do them better. For my own part, I wouldn’t spend months struggling with a research problem that I knew AI could solve instantly or take great pleasure in the answers it provided. I don’t want to be a mere observer of the world, whatever wonders AI may reveal.
Or take this essay. I will be heartbroken when a chatbot can extract my innermost feelings and, after indulging in the words of a million artists, imitate Fitzgerald-worthy prose that I cannot match. For me, writing is a process tied to self-discovery and human connection. My sister suggested the idea of this essay; my wife, seeing me suddenly and very sad when I thought about it, touched my cheek, giving comfort that no AI expert could. Later, I wrote late into the night at the handmade dining room table I inherited from my grandparents. I imagined how my family would gather for a long dinner around this table—adults relaxed with wine, children happy to be a part of it all, everyone laughing and talking about each other and discussing physics and philosophy—trying, slowly, in a small, human way, to figure things out.
