Could AI really cure cancer?

This podcast contains audio advertising. To hear it without ads, head to the ABC Listen app now. Everyone's got a hard question about AI. Your job, your kids, where it's going. Anthropic was built to surface those questions and share what it finds along the way, because there's hope in hard questions. Ask yours at clod.ai/spotify and keep thinking. That's clod.ai/spotify. ABC Listen, podcasts, radio, news, music and more. Hi, I'm Patricia Carvellas, the host of the Politics Now podcast, and navigating what our rapidly changing world means can be tricky. I do think you're right to point to the ability of the Backbench in Britain to agitate, be upset with Keir Starmer. That's one of the reasons he's gone. That's why I'm joined by the sharpest political minds of the ABC to break down what each development means for Australia, and how the government is positioning. Albanese has got a much more compliant Backbench, and he's got much more dominance of the Parliament just because of peculiar Australian features of our system. That's Politics Now with me, Patricia Carvellas, funded on ABC Listen. We're being warned that AI called wipeout humanity, but some of the very same experts sounding the alarm are also predicting AI will help cure cancer within the next decade. So how will these competing visions of destruction and salvation play out? And is it cure for cancer really within AI's reach? Today, Doctor and Researcher Amelia Javorski joins us to separate the reality from the height. I'm Alyssa Clark on Gadigal Land in Sydney. This is ABC News Daily. I believe that as this technology progresses, we will see diseases get cured at an unprecedented rate. We look at these tough diseases like cancer, like Alzheimer's. I just think AI is the right thing to cure them. I believe in our lifetime AI will help cure cancer. The strategy is to help scientists and the scientific community cure all diseases. The end of disease? I think that's been within reach, maybe within the next decade or so. I don't see why not. Amelia, what do you think when you hear some of the promises from tech executives about the potential for AI to cure cancer? Does it excite you or does it make you wary when you hear those claims? It makes me reflect on the promises we've heard in the past, because promises from Silicon Valley that we are going to have a health care revolution and cure disease are actually not new. There's a long history here going all the way back to we think of systems like IBM Watson, which was the AI system that was really great at Jeopardy and they tried to bring it to health care and it turned out to be very not good at curing patients. We've heard Tim Cook from Apple say things like Apple's legacy is going to be a health care revolution. That was in 2019 yet to see that materialized. And then really going all the way back to 2013. There was this movement that started that was called the tech bio movement. And this was taking biotech but reversing it to make sure you knew the tech part was really important. And this was the first wave of companies that were using AI for drug discovery. And there was a lot of promise that AI was going to be able to discover new drugs. It was going to transform care. Yet here we are 13 years later and we still don't have in the US at least a single FDA approved drug that has been discovered with AI. So there's a long history here. So I guess that definitely informs how we look at some of the claims that are being made. But Amelia, before we dive into your paper on all the ways AI can and can't cure cancer, can we just take a look inside a cancer tumour, can you just help us understand what is cancer and why we haven't found a cure for it yet? Yeah, when we talk about cancer, we talk about it as if it's one thing, right? A cure for cancer, one cure, one cancer. But cancer is actually thousands of different diseases that are highly personalized. Each of those sort of variants of cancer have their own drivers and their own mutations. So if you looked at a cancer tumour and zoomed in, you would find even within a single tumour, different cells have different types of mutations that are driving that cancer. So it's something that is much, much more complex than a single disease, which is often how we think about it. Cancer is complex, it's adaptive, it responds to treatment, it's really a moving target, not a fixed target. And because of this complexity, the more and more we've studied in oncology and the history of cancer research, has been hoping to find simple answers, but actually finding more complexity. And that's why a universal cure has been so elusive. So tell me some of the ways AI is already helping cancer research. I think that is one of the most exciting aspects of AI in cancer is we don't actually have to wait on future AI. There is so much going on of how AI is being leveraged to actually help patients today. So some examples of that are using AI to predict which patients will respond to which treatments. And actually looking at cancers to help anticipate resistance before it happens, giving you an opportunity to treat that before you get a resistant cancer. Or working with the surgeons in the operating room when they're going to take out tumors to be able to understand where the tumor margin is, so they can get all of it on one go and be confident that they've actually gotten all of the cancer cells out. And I think above all one of the ways AI is being really effective in oncology is helping to catch cancer earlier. So through AI in imaging and AI in blood tests, it's able to detect cancer much earlier than we were previously able to do with our human minds and technologies alone. And that gives you a much better chance of curing it. So there's certainly some of those advances happening here in Australia, which I think is really exciting for people who are either going through cancer themselves or have families who are going through that experience. Australian-made artificial intelligence tool can detect women with a high chance of breast cancer who were missed in normal screening. Researchers say it's the biggest step forward in breast cancer screening in 30 years. University of Queensland researchers have studied biopsies from almost 250 patients with non-small cell lung cancer, scrutinising each sample with the help of AI. There are quite specific tasks, is that what AI can be used for at the moment really well? Those kind of very specific dedicated tasks like where's the edge of a tumour? Or how do I make sure this individual responds to a certain treatment? Yeah, AI is great at those sort of bounded use cases. And I think it's also sort of where, exactly as you say, narrow well-defined tasks, AI does really great with. And also tasks that have a lot of data from AI to learn from. But there are some in the tech sector who are looking more to artificial super intelligence. Some in the tech sector think that this kind of artificial intelligence could do what human brains haven't been able to do and find a way to cure cancer as a singular item. Why do you think that's not the right approach to AI and medical research? Yeah, I think fundamentally curing cancer is not an intelligence limited problem. And we know this from looking at the history of medicine and science, right? As over the past decades, we've learned so much more about medicine and biology and have more scientists on the case and have made breakthroughs in some areas. But in other ways, the cancer mortality has made pretty stubborn and really hasn't moved. And I think this comes down to you. Do you have the data that's necessary to actually learn from? So AI is amazing in domains like physics and like math. And we've heard a lot about how AI is making all kinds of breakthroughs in those fields. But what those fields have is basic rules of the road. And so when you have basic rules of the road, it's quite easy to, you know, have more intelligence, mean more breakthroughs. Whereas with biology, that doesn't really exist. Biology doesn't have first principles. Our bodies are complex, adaptive, evolved systems were not engineered. And I think people often say, well, we'll just work up from the physics, right? We know there were rules of physics, so we could then know the rules of humans. Now, it's not only like computationally infeasible that it would take so much compute to simulate a human being. But it also just doesn't work that way, because it's not linear from physics to. our biology, you know, as you move from an atom to a cell to a tissue to a full human, there's different properties that arise that you actually can't predict from lower layers. Are there big sky thinkers who might say, "Well, there might be rules of the road or patterns that we just don't understand, yes, and maybe AI can help us understand them?" I suppose that might be their rebuttal to your view. There's a great scientist by the name of Stephen Wolfram, who came up with this idea of computational irreducibility, which really means, and biology is a use case of this, that there are some things that you just cannot reduce to a certain amount of parts. They need to be measured, not simulated, and biology is really in that domain. So I think our big bottleneck in biology is how do we measure it better? How do we measure what health looks like, what disease looks like at scale, and generate that data to then apply AI to learn from and denote insights from? Amelia, there's something about the promise of curing cancer that I think hits people really hard, probably because most of us know someone, a family member or a friend who's been affected by cancer, and I understand that's the case for you as well, losing your father to cancer, there's a bit of a personal response that I think many people have that go, "Well, even if there's just the tiniest chance that it helps, or we can put a lot of energy and resources and finances into developing artificial superintelligence, and maybe it won't deliver everything at promises, but even if it can help with something, isn't that worth it? How do you think about that idea? So, cancer is, thank you for that, and cancer is very deeply personal to me from losing my father to cancer, and also by profession being a physician by training. And I completely agree that we should be racing full speed to a cure for cancer. I just don't think that superintelligence is the fastest road to get us there. And all of this focus on superintelligence clouds out the sort of other things we need to do in order to get to a cure. And there is so much promising science that is going on among scientists and reformers and small startups that is really underfunded and overshadowed by this overarching narrative that we're going to race to this superintelligent genie that one day will give us a cure for cancer. If it does what we say and it listens to us and somehow magically comes up with the solution. But I think that something that is an onus on the companies to really start to get into, which is how do we anticipate that a system like that would do that? And up until now, the conversation around it has been very light on specifics. It's sort of, we will build the superintelligent genie, there's some hand waving and somehow it will cure cancer. And I think that is, you know, a great promise, but it's a promise without a plan. I think that obscures the amazing scientists that are actually hard at work with plans, with promising data, with great, you know, high potential solutions to actually cure cancer in the here and now it takes attention and capital away from their efforts. And so when you hear those promises, do you think the leaders of these tech companies misunderstand where the most promises or do you think they're misappropriating people's desire to see a cure for cancer or to sell investment in their products? I think it is a mixed bag. I certainly think that there are well-meaning individuals who are a bit naive about, you know, the challenge that is ahead. And that same naivete that we have seen in the past with Silicon Valley coming to biology and healthcare, assuming something will work only to be confronted with. This is a lot more complicated and a lot harder than I thought it was. I do think there are also folks that are leveraging this narrative, not in sort of a benevolent naive point of view, but for, you know, a more sales-oriented point of view, but also to distract from the risks of the technology that they are creating. So there's been a lot of conversation in the media about the risks of developing superintelligence. And I think that this conversation in particular is one that can be used to say, "Oh, but wait, but we make your cancer," right? Don't look at the risks, don't look at the things we make your cancer and steering the conversation in that direction. So where you sit right now, you have seen some of the advances that AI has been able to help with in specific circumstances. You've cast a skeptical eye on some of the grander claims coming from the sector. But how optimistic are you about the potential for really helpful breakthroughs over the coming years and decades? I have so much hope for the future of where we're sitting in cancer research at the moment. And that actually informs a bit of my critical view is seeing so much attention and funds be diverted away from the amazing hopeful research that is actually happening in academic institutions and small companies and even in large pharmaceutical companies. Like, I think there's a lot happening that makes us pretty close to maybe not curing cancer, certainly curing some cancers, but being able to transform cancer into a manageable chronic condition. Much like HIV went from being something that was a death sentence to being something that if you had the right medical management, you can live a normal quality of life. Amelia Javorsky is the director of the futures program at the Future of Life Institute, a non-profit aimed at reducing AI risks. She's also a scientist and mentor at the Vis Institute at Harvard University. This episode was produced by Ilaria Brophy and Louise Mylon, audio production by Sam Dunn. A supervising producer is Sydney Pied. I'm Melissa Clark. Sam Holi will be back tomorrow, thanks for listening.