By some estimates, more than a trillion dollars have already been invested in artificial intelligence. But large tech companies including Meta and OpenAI are still not content with today’s AI, and say they have set their sights on powerful, versatile AI that by some measure would match or even exceed human performance. A remarkable amount of resources is being poured into developing artificial general intelligence (AGI) or even more capable artificial super intelligence (ASI).

Excitement around the potential of such a technology is often accompanied by casual claims of some remarkable capabilities. One in particular—curing cancer—stands out to Emilia Javorsky, director of the Futures program at the Future of Life Institute, a think tank focused on benefits and risks of transformative technologies such as AI.

In March, Javorsky published an essay titled “AI vs Cancer,” which draws on her experience as a doctor, scientist, and entrepreneur. It is a critique of putting our faith and resources into artificial super intelligence as a future solution for disease, particularly when so many factors other than intelligence limit the development of new treatments and access to innovative care. AI cannot analyze patient data that was never collected, and any treatment is flawed if patients risk bankruptcy seeking it. But the essay is also intended, she says, as a source of optimism in the ways that existing forms of AI are already being applied to cancer.

Javorsky spoke with IEEE Spectrum about the essay. The conversation has been edited for length and clarity.

What it means for AI to “cure cancer”

What do you mean when you say “cure cancer”? And what do you think people who talk about the potential of ASI to cure cancer mean?

Emilia Javorsky: “Curing cancer” is how the problem and solution are framed in both the general discourse around AI, but also specifically the promises being made out of the labs developing AGI and ASI. So it was important to me, if I was going to interrogate the promise, that I lean into the frame. But to me, the framing is off.

Cancer is not one universal disease that one universal treatment could potentially cure. It’s a highly individualized co-evolutionary process. In each person, a different set of mutations are driving the cancer. And even when looking in a single tumor, different cells have different mutations driving their biology. The solutions are probably going to have to be somewhat individualized.

And if we’re honest with ourselves in medicine, we have yet to cure a complex chronic disease. We have really good ways to treat and to manage diseases like diabetes, like heart disease, but we’ve yet to actually cure them. So the curing frame is one that I also push back on.

I think [the medical community’s] hope is to find highly effective personalized treatments to manage cancer and to turn it into something that is chronically well-managed, that no longer becomes something like a death sentence.

How should we think about the difference between AI and AGI or ASI in the context of cancer?

Javorsky: In those promises [to cure cancer], more often than not, people are using [the term AI] to describe AGI or ASI, this kind of future super-intelligent genie that in their worldview will magically grant us wishes to solve problems. That should be disentangled from AI that we already have that can solve problems.

We hear a lot about AI in drug discovery, AI in predicting the toxicity of new drugs, AI for defining new biomarkers, for making clinical trials go faster, or for detecting things earlier.

All of those modalities are actually in the clinic moving the needle and accelerating innovation today. There are companies and academics working on all of those. There are a lot of AI scientists hard at work that are actually unlocking the potential of the technology in the here and now.

I think that real progress often gets overshadowed by this kind of looming future AI system promise, when actually, probably the most effective way to solve the problem is with the tools already available to us.

Investing in finding cures

I read sections of the essay as an argument in support of collecting lots of health data. But you’re not strictly against AI or investing in developing the technology. You’re trying to find a balance between innovation and pragmatism in this essay, is that right?

Javorksy: In a world where there’s finite capital and curing cancer is very probably the most noble thing the capital can be put in service of, we need to figure out, “where is the [return on investment]? Where can we invest in order to get the most that we need to actually help solve the problem?”

I argue that we’re overinvesting in the intelligence-compute side of things and underinvesting in innovating our tools to measure biology and our creation of large-scale, high-quality data sets.

We have a healthcare system that is a “sick care” system, fundamentally. We only see people and start to measure them when they become ill. When you start to put the frame of “what data do you need? How do you measure it?” it forces you to take a bigger picture look at the practice of medicine and biology in general.

In an ideal world you could pursue all paths, but that’s just not the reality of how we invest capital. Where I land is being very bullish on AI, but spending money on the right types of AI and the right pieces of the bottleneck.

What AI applications related to cancer are exciting to you right now?

Javorsky: Something we’re already seeing is the ability to detect cancer earlier. We’re already seeing AI accelerate and help us run clinical trials better. There are really awesome things happening with in silico modeling work: virtual cells, figuring out digital twins. How can we create a high fidelity, digital representation of you, in order to figure out what would work best for your biology and really unlock the promise of personalized medicine?

You conclude the essay focused on solutions. Could you explain that roadmap to me in brief?

Javorsky: Part of this essay was to diagnose where we’re getting some things wrong, but with the roadmap, I wanted to offer up my point of view on what we actually need to do to solve this problem. What will it take to cure cancer? Let’s get really serious about what that could look like.

And so I break that down into three buckets. One is resourcing and scaling the AI tools that are already making progress in oncology. The second piece is really doubling down on investing in the promising areas in biology [related to oncology]. And then finally, more broadly, tackling what I would call the institutional and systemic bottlenecks and misalignments to medical progress.

I wanted people to realize that the reality is actually quite hopeful.

From Your Site Articles

Related Articles Around the Web