[MUSIC] Hello and welcome to GZERO World.

I'm Ian Bremmer and today we are taking stock of a war that has been 50 years in the making.

And I hope that in the years ahead that we may look back on this day and this action as being the most significant action taken during this administration.

In 1971, President Richard Nixon signed a landmark bill in front of a packed White House audience, the National Cancer Act.

And while unfortunately for President Nixon, it would not go down as the "most significant action taken during his administration," ahem, Watergate cover-up, it did mark the start of America's war on cancer.

Half a century and hundreds of billions of dollars in research later, who is winning that war?

The short answer is cancer.

Today, roughly one in two men and one in three women in the United States will be diagnosed with the disease in their lifetime.

It remains the second leading cause of death in America behind heart disease, and it kills nearly 1,700 people a day.

Nor is the death toll evenly shared.

Black women, for example, are 40 percent more likely to die of breast cancer than white women, despite lower incidence.

And the price of survival can be steep.

Modern cancer drugs often cost over $100,000 a year, with patients shouldering more than $21 billion annually in direct expenses.

That's the short answer.

The longer answer is we are putting up a heck of a fight.

Survival rates, for instance, have increased dramatically.

In the 1960s, just one in three people diagnosed with cancer lived at least five years.

Today, it's more than double that.

Cancer is personal.

Both my parents died from it.

If you haven't had it yourself, you know someone who has.

But my guest today gives me hope that technology may be finally putting us on the winning end of that fight.

As artificial intelligence continues to advance, we could be at the beginning of the most significant medical advancement of our lifetime.

Joining me to talk about how AI is reshaping medicine and the fight against cancer is physician and best-selling author of "The Emperor of All Maladies," Siddhartha Mukherjee.

Don't worry, I've also got your puppet regime.

For more than 2,000 years, you guys have been stealing ideas from America.

But first, a word from the folks helping us keep the lights on.

Funding for GZERO World is provided by our lead sponsor, Prologis.

Every day, all over the world, Prologis helps businesses of all sizes lower their carbon footprint, and scale their supply chains.

With a portfolio of logistics and real estate and an end-to-end solutions platform addressing the critical initiatives of global logistics today.

Learn more at prologis.com.

And by Cox Enterprises is proud to support GZERO.

Cox is working to create an impact in areas like sustainable agriculture, clean tech, health care, and more.

Cox, a family of businesses.

Additional funding provided by Carnegie Corporation of New York, Koo and Patricia Yuen, committed to bridging cultural differences in our communities.

And... Siddhartha Mukherjee, welcome back to GZERO World.

Thank you so much.

I've been doing a lot of shows recently on artificial intelligence and how it seems to be affecting so many fields.

But medicine comes up a lot.

And you exist to a degree at the intersection.

Talk about what is surprising to you about where AI is changing the trajectory of medicine.

I like to think about it in terms of three to four very broad silos.

And I'll talk a little bit about where we are in those silos and where we aren't in those silos.

So the first is the most obvious silo, which is the patient-facing AI silo.

It usually involves applications that patients can use usually on their devices, such as scheduling, simple triage, simple questions like, "Should I, you know, I have X, Y, or Z, do I need to see someone immediately or can I wait?"

Involves billing the movement of electronic medical records, the central records, again to your device and potentially beginning to start scraping some of that information for the patient's use.

They can ask, like they ask a chatbot, you can ask your EMR a question, which is something like, when's the last time I had my cholesterol checked?

That field is very full of actors, of people who are doing this.

It is, I would say, low-lying fruit AI.

It's very helpful to patients, but from the AI standpoint, or from the cures for diseases standpoint, of course, has no real relevance.

Then we come to the second category, or the second big silo, which is doctor or physician-facing AI.

Now we're talking a little bit more complicated things.

We're talking about not just, again, not just the medical record and how you interface with your physician, but potentially projects that have to do with things like diagnostics.

So just to give you a couple of examples, diagnostics that rely on images, visual diagnostics, radiology, so CT scans, MRIs, et cetera, dermatology, so looking at a rash, looking at a lesion, is it melanoma, is it not melanoma?

And finally, pathology.

Is this breast cancer aggressive or non-aggressive?

What do we know about it, what do we understand about it?

- Is it true that AI in some of these spaces is already more effective than doctors would be in some of these areas?

I've certainly seen studies that have implied that.

So part of the answer is it depends on who the doctor is.

Doctors with years and years and years of experience generally tend to outperform what is currently some of these AI systems.

Whereas obviously a young doctor who's getting experience is not going to outperform these AI systems.

The good news about that is that going forward, it's almost certain that, at least to me, it's almost certain that it's going to be a mix, the physician and the machine, the learning machine, and a mix that is going to be over time refined because, as you know, the first day you start reading a CAT scan is, you know, that's your first CAT scan.

The AI machine, the machine has usually, you know, it starts with 500,000. to diagnostics, especially in these image intensive fields.

Then comes real data mining and data querying.

So your EMR now aggregated across multiple medical records, electronic medical records, creates a huge corpus of knowledge.

That huge corpus of knowledge is untapped for the most part, partly because it's very complex knowledge.

In some cases, like with the Veterans Affairs, in many cases the knowledge is not in fields.

In fact, it's free text knowledge.

It's not on a spreadsheet that's being shared in a broader environment.

Yeah, and it's not-- I would say, more importantly, the structure of the data, the underlying structure of the data, is not divided up appropriately.

So ironically, this unfielded data is actually going to get a move forward.

It moves faster with AI, because many chatbots are able to read it faster than they could read fielded data.

So that's very interesting, because you could ask, you could query aggregated data and ask potentially very interesting questions like, tell me whether or not or what kinds of patients had a response to this drug?

Are there some characteristics that I can draw out of that?

This is now we're moving into territory that is really interesting data mining.

And this is truly complex data.

And I would say that the first applications in this realm in AI are beginning to happen.

So that's the second, I would say, broad silo.

The third broad silo, which is to me the most interesting, is true discovery, generative discovery.

And there, we're essentially trying to find out that if you have a target, whatever it might be, some protein that is not working normally in a cancer cell, can one develop a medicine that will change the activity of that protein?

In other words, we are trying to create a new pharmacopeia of chemicals, potentially of chemicals that have never been created before in human history, to address these very complex diseases.

And allied to that is once we have those proteins, what is the best clinical trial that one can run so that you can get the maximum benefit from this drug.

>> And so how was that developed?

>> So broadly speaking, there are two or three ways that one develops these medicines.

And it's super important to understand them because this is the magic promise.

First of all, you try to understand that every medicine works by binding to something else.

And by binding, it means that it goes and physically attaches itself.

And that attachment has some physical characteristics or biophysical characteristics, which are based on chemistry and physics.

Which will strengthen it, destroy it, change its behavior.

For instance, exactly.

So once we know that, we can ask a couple of questions.

One is, then, if you know the target that you're trying to attack, you can essentially figure out whether-- you can figure out where there's a pocket in that target.

When you identify a pocket, it's like puzzle solving.

Just like in a puzzle, you have a little hole and then you have to put a puzzle piece in.

How do you put a puzzle piece in?

So there are two very broad, it's going to be very broad, but two very broad ways that AI can help in the puzzle fitting algorithm.

So again, think about how a child solves a puzzle.

So one way a child solves a puzzle is by, let's say a puzzle has, it's almost done, there's one piece that's missing, that's that pocket that you want to target, and there are a billion pieces sitting on the side.

So what a child does is, generally speaking, she starts fitting the puzzle, and then she tries several times, and then realizes that the puzzle, the missing piece has a certain set of characteristics.

It has to have a divot on top, and a little hole in the bottom, etc.

So she now goes back, so that's being learned, that's a learned property, and that's why AI comes in, is very important.

So then she says, "Okay, I've learned that property, I've extracted that property as a feature, and now I'm going to go back into the many pieces, I'm going to search for that property, I'm going to eliminate 99% of the pieces that don't have the property.

I'm going to search for that property.

And now I'm going to fit again."

And so now I have a much better fit, but I'm still not there yet.

So then I learned that in fact, the ones that are left have most of the properties, but the one thing that's missing is the little hole on top.

So I refit and so forth.

So that's one way that you can create a new medicine.

- And the interesting thing, of course, is it's not that this is something that humans couldn't do given enough time.

- Absolutely.

- But what AI is doing is creating maximum efficiency given the extraordinary number of permutations you're talking about.

- That's right, yeah.

And also remember that those billion pieces that I talked about, they've also been generated computationally.

So humans could do it, they would first have to make the billion pieces, and then they would have to find the fit.

And absolutely, it's what humans can do.

In fact, that's how we've been making medicines for decades.

And then there's the most interesting one, the one that we do, and the most interesting variant of all of this.

You could have a particularly rascal child who says, "You know what?

"I don't wanna play this stupid game.

What I'm gonna do is I'm gonna take a piece of paper and I'm gonna cut that piece of paper and make it fit into that puzzle piece and I'm gonna put that puzzle piece in."

Now, what does that mean in real life?

It means that you grow the puzzle piece inside the hole.

So you learn the rules and then you begin to take that puzzle piece and say, I understand the rules of that missing piece.

I'm gonna start creating candidates.

I'm gonna forget about the billion pieces.

I'm gonna start creating candidates that will fit into this puzzle piece.

- Which may have never existed before.

- May have never existed before.

This is true generative chemistry.

It is true generative AI and true generative chemistry.

So what I'm trying to emphasize is the scale and power of all of this.

In principle, of course, humans could do it.

But every time we do this in collaboration with a machine, the machine learns it and it learns it forever.

You don't need to train a new chemist, a new generation of chemists, the machine will now learn it for eternity.

So now we've been talking a lot about what AI is doing in the field.

Now if I flip it and say, okay, how does one address cancer specifically?

Sure.

Again, let's divide up the field into three major categories.

As you can see, I'm a big splitter.

Prevention, early detection, and treatment.

But before you can do the elimination, mitigation, et cetera, et cetera, you have to actually find the agent.

And our tests, we had two broad kinds of tests to find cancer-causing agents.

So one broad kind of test was a laboratory test.

So this included putting the chemical onto cells and asking the question, do they start getting mutations?

And of course, mutations are what cause cancer.

So this was called the Ames test.

The second kind of test, which is very labor intensive, to expose animals like mice or rats to a suspect chemical and ask the question, do they get cancer?

The problem is that we're now finding out that the Ames test, even though it's very valuable, misses a whole host of carcinogens or cancer-causing agents, chemicals.

So for instance, I'll give you one example.

Asbestos is a great example of a chemical which clearly causes cancer, lung cancer, mesotheliomas and others, doesn't score on the Ames test.

The other kind of test is the opposite, is a test which is epidemiological.

In other words, you study a population and you ask the question, is there something in that population that gets cancer or some part of the population gets cancer?

And you ask the question, what are they exposed to?

Right.

So asbestos, of course, comes very clearly in that category.

So occupational workers, people who work with, you know, insulation, whereas asbestos clearly have very high rates of lung cancer, mesothelioma, et cetera, et cetera.

So, and we've known that for a long time.

So there was a discrepancy between the epidemiologists and the laboratory scientists.

And it was very hard to sort this discrepancy out because you know you don't know what what is doing what.

Most recently what's happened is that we've figured out that inflammation, inflammation defined as a very particular kind of activation of the immunological response, collaborates with chemicals to cause cancer through a mechanism that we're beginning to identify.

And so going back to again, to asbestos, it's very likely that the asbestos particles cause a particular kind of inflammation.

We don't know this for asbestos fully yet, but cause a very particular kind of inflammation, which ultimately leads to cancer.

Now, why is that important?

It's important because all of a sudden, the universe of things that cause cancer has widened.

And now... - And become specified.

- Exactly.

And now all of a sudden, really I would say years and years of, after years and years of waiting, we have a new theory that it's, cancer is caused by mutations, yes.

It's caused by abnormal cell division, yes.

But it is also caused by an inflammatory milieu.

Now that leads to a whole host of new potential carcinogens that need to be evaluated.

I'll give you a couple of examples.

One great example that people are now very concerned about are these forever chemicals.

Generally speaking, these inflammatory chemicals tend to be like forever chemicals.

So the immune system tries to metabolize them, swallow them up, clean them up.

They can't do it.

And in response to that, you get inflammation and cancer.

And it's a big deal because we are suddenly finding that we can explore the world, the chemical world and the biological world to some extent, and really begin to think about, you know, is it carcinogenic?

Is it not?

And how to mitigate and remove them, etc.

We don't know the answer to, for instance, these forever chemicals.

But just to give you, you know, how it changes practice.

I, for instance, I threw away all my Teflon wear, because Teflon is a consequence of one of these forever chemicals.

So I now cook only on non-Teflon pans.

Now, is that proven in your view?

No, no, no, no, no.

So we don't need to tell everybody yet to get rid of Teflon.

Please don't do that.

Because, I mean, you know, but you're being overly, abundantly cautious.

I'm being abundantly cautious.

Okay, fine.

All right, so then we come to early detection.

And so what's going on there?

We were using five marginally effective mechanisms to detect cancer early.

Colonoscopy, pretty effective.

Pap smears for cervical cancer.

Mammography for breast cancer.

So probably misses a lot of cancers.

Low dose CT for high risk people such as smokers for lung cancer.

And then the PSA test for prostate cancer, which as you know is very complicated.

It has a lot of false positives.

And it's not really a great test for prostate cancer.

So those were the five.

What's happened in the interim is that we've invented, scientists have invented blood tests that can detect cancer at its very early stages based on DNA that's shed by the cancer cells into the blood.

So there are many companies that have grown around this.

One prominent one is called GRAIL Therapeutics.

There are several others.

The problem with these is that they're still undergoing testing.

We don't know whether these will ultimately lead to lives saved or lives not saved.

- And treatment?

- So treatment is a completely different universe because this is where AI has the maximum impact.

And so in treatment, we're using new immunotherapy.

We are using new modes of genetically dissecting patients and understanding what drug they're most likely to respond to.

And we're using AI in every dimension of guiding treatment so that we can get better and better at treating patients.

And that ranges from, you know, more directed, guided treatment for the delivery of therapy, all the way to, you know, the invention of new drugs, as we've just mentioned.

So that's what's going on with treatment.

- Sid Mukherjee, thanks for the work you're doing.

- Pleasure, thank you for having me.

(electronic music) - Now we go from the cutting edge of scientific discovery to the equally thrilling world of legless puppets with inferiority complexes.

I've got your Puppet Regine.

- Welcome back to This Authoritarian Life, me and Xi's podcast, where we seek to establish total control over emotional health and well-being.

(laughs) This week, we're welcoming back old friend, Donald Trump.

- Glad to be here as always.

Hey, you wanna end the war in Ukraine?

Still no.

- Okay, worth a shot.

Forget Ukraine, guys.

I want to talk about big news recently from America.

You mean the Travis and Taylor engagement?

I hate Taylor.

I love Taylor.

I love Taylor.

She's a terrific person.

But I meant more some big moves by our boy Donald.

I like that he banned burning flags.

No one should ever insult the government like that.

Well, that's right.

But also, you know, I couldn't just let people destroy something that is so important to... your economy, Xi.

I mean, frankly, most of those flags were made by you.

Strong facts.

I really enjoyed you bullying Central Bank, forcing government to take over technology company, and sending police to home of disloyal former employee.

I couldn't have done it better myself.

Should sue this guy for IP theft.

He really is ripping off all kinds of material from us.

You know, Vladimir is right.

For more than 2000 years, you guys have been stealing ideas from America.

Well, the table's a turn, suckers, because under my leadership, we will now be very strongly stealing ideas from you.

Huh.

Hmm.

So where do you guys think Swelce wedding will be?

♪ Puppet Regime ♪ That's our show this week.

Come back next week.

And if you like what you've seen, or even if you don't, but you've discovered a cure for something really bad, check us out at gzeromedia.com.

[music] Funding for GZERO World is provided by our lead sponsor, Prologis.

Every day, all over the world, Prologis helps businesses of all sizes lower their carbon footprint, and scale their supply chains.

With a portfolio of logistics and real estate and an end-to-end solutions platform addressing the critical initiatives of global logistics today.

Learn more at prologis.com And by Cox Enterprises is proud to support GZERO.

Cox is working to create an impact in areas like sustainable agriculture, clean tech, health care and more.

Cox, a family of businesses.

Additional funding provided by Carnegie Corporation of New York, Koo and Patricia Yuen, committed to bridging cultural differences in our communities, and... [music]