YouTube Transcript:
How Generative AI Is Reinventing Global Healthcare

Skip watching entire videos - get the full transcript, search for keywords, and copy with one click.

Video Transcript

Available languages:

View:

Well, good evening, everyone.

It is such a pleasure to be here for so many reasons.

First of all, I'm originally from Brockton, Mass.

Home of champions.

And I remember coming here to the -

here we go, right? Yes.

Marvelous Marvin Hagler, you know. And, I remember coming here,

Tim, back when this was one building and the dinosaur only had a head.

So it is a great honor to be here.

The other thing is that, my relationship with Harvard Business School, the,

Harvard School is where I started my professional career.

So I got a degree at Harvard College, went - got my MBA there,

and then my doctorate - I actually did my doctorate on the economic

implications of adopted expert systems in 1986.

And so it is -

And then I taught in the faculty there for a dozen years.

So it's wonderful to be here.

I have great love and affection

for the institution and it’s great to be able to have that.

And most importantly, the reason I'm happy to be here

is that I really believe, and it's why

Paul Bayer and I started this company that we are at an inflection point where

from a particular point of view, we're returning to a renaissance.

And what I mean by that is when you look at generative AI,

it is categorically different for two fundamental reasons.

First of all, it's the first time in the long

history of automation that the machines are talking back.

Okay. And that is profound.

The second

thing is, if you understand anything about the way the models work,

they have matrix math at the at the middle of them.

And in that you have an - you have a multi-dimensional

semantic matrix that is allowing us to munch down reality,

different data types, and to navigate it in

human time, not infinite time, searching that space.

And that is a renaissance in the sense of being able to do

all kinds of new analysis and to cross disciplines.

And as Leonardo da Vinci did, go all the way from art

to engineering to warfare to science and back again.

And from that point of view, we are very, very lucky to have these two participants

on our panel, Joan and Luba, if you could come up

and I'll do a little introduction here.

Here we have

two unbelievable renaissance professionals.

So Joan Lavorare,

I was in visual and environmental studies at Harvard College.

So was Joan.

But Joan also did the hard part of pre-med, which I didn't do.

And, Joan has got her

masters, in engineering from Saint Andrews.

She's got her MD from Columbia.

She's got her MBA from Sloan and her undergraduate degree,

as I say, both in pre-med and in art.

And you'll see in her work and her experience,

she is a renaissance professional.

Luba Greenwood among her

in addition to being an attorney and actually,

enforcing the writing on things like the enforcement

of the Foreign Corrupt Practices Act early in her career.

She also has been a renaissance professional.

When she was head of business development at Roche,

she was the first person to bring biopharma into the Flat Iron Deal.

For those of you who follow that,

that was a mega-deal that completely changed the landscape of,

biotech and especially pharma, and bringing them into life

sciences at scale and in the cancer area, if I remember right.

And she also took that and went over to Google

and helped Google build out verily, their life sciences activity.

And now she is CEO of Gallup Pharmaceuticals

that is applying new procedures and capabilties.

And also has served on a number of boards and so forth.

So in these two panelists we have the best representatives

of what I think is not only a domain that matters, but a way of thinking

that I invite you all into, because I believe we're at the beginning

of a renaissance of science and commerce like we haven't seen before.

So welcome.

Thank you.

As part of this, we're going to start off with a few,

beginning comments, both from Joan and from Luba,

in terms of setting the table.

And then we'll get into a conversation about some of the key issues.

Being an ex-professor,

I'm going to give you kind of the answer

on what we hope to get to at the beginning of this,

and we can discuss it, but there are three things

that, and preparing for this, we wanted you to at least consider.

What AI is enabling is, first of all,

new discoveries in science that we've never had before.

And you have to think of these

capabilities, not the whole anthropomorphization of generative AI

and AI I think is a tremendous mistake.

It's not about one person replacing - a robot replacing one person.

We're not talking about a collective hivemind.

Think of it more as, you know, cockroaches meet bees, meet

schools of fish that you can call out any kind of intelligence that you want.

It really is a hive mind that we're dealing with, and that's

going to allow us to do science like we've never done before.

There's an announcement by Google today about, you know, a robot that helps

scientists generate whole new ways of doing hypothesis and testing.

So first thing.

Second thing is it's fundamentally changing the liquidity of knowledge.

Now what I mean by that

is that we’re going to see models, and we’ll talk about some of these.

And as consumers of health care,

the ability to move medical knowledge and analysis down to your phone, out

in rural areas, in places

that never had that kind of knowledge before, is fundamental.

And that - that liquidity of knowledge and intelligence is radically changing.

And the third thing is, as we instrument the world

and we instrument our own bodies with Apple, you know, Apple

Watches, Fitbits, the environment that we're in,

we are now able to take that massive amount of data

and make sense of it as we codify and digitize that world,

and to take all kinds of data types unstructured,

structured, traditional, nontraditional, and have new insights.

So those three things: new science, liquidity of knowledge,

and environmental understanding, including our body as an environment,

we we want you to think about as we go through this.

So with that, Joan, you want to start us off?

Absolutely. Thank you so much, John.

I also remember coming here as a child and it is very changed.

It’s wonderful to be here tonight.

I’m going to give a couple of slides

as some background if I can make this work? Okay.

So I am a physician at Boston Children's Hospital in

cardiac-intensive care.

I'm also associate chief medical officer for transformation.

Happy to discuss how AI is impacting

us there at the hospital.

I am also co-founder and president of Virtu Foundation,

which is an NGO with special consultative status to the United Nations.

And that's really how my journey into

AI started fifteen years ago.

And I want to talk a little bit about that tonight and how we've used AI

and now generative AI to map

72 low and middle-income countries and how that's going

to really revolutionize how we can meet the supply/demand needs.

So I have a brief video that I'll just kind of summarize that first.

It's an interesting use case of AI,

so that's why I wanted to bring it.

Virtue Foundation delivers clinical care

and uses data science for good to impact global health

and bring care to millions of people across the globe.

What we're trying to do is continually refresh

our foundational data set, how to create those many knowledge

graphs around each hospital, around each health care NGO.

I think that is one of the most fundamental things that Databricks

is assisting us with today.

For the past 15 years, really under the vision of Doctor

Ebby Elahi, who I co-lead the Actionable Data Initiative

with, we have been focusing on building an AI platform for global health.

By being able to understand the situation on the ground,

where are the hospitals and healthcare NGOs?

What is the work that they're actually doing?

We're able to take that data and match it up to the clinicians

who actually can deliver at that specialist level.

There's currently 143 million patients

every year waiting for surgery in low and middle-income countries.

It's a really large problem,

but if you can create efficiencies in the data, you're talking about millions of lives.

The Databricks gen.

AI capabilities has allowed us to refresh our foundational data set.

I personally do pediatric cardiac-intensive care.

I want to understand where are opportunities to take care of children,

and where could I be able to add skill,

particularly in the care of complex patients.

Postoperatively after cardiac surgery.

Virtue Foundation’s

delighted to be a founding NGO for Databricks For Good.

It's helping us to unlock data to bring impact at scale

to millions of lives on the ground.

So I think that summarizes well

some of the work that we have been doing at Virtue Foundation, in the space.

And it's a really novel use of AI

because you see the burden, the 143 million people

without surgery, almost 3 million disability-adjusted life years.

How do you get to the granular level where you can bring that surgical team

to the right place and deliver the right care?

And I think one of the things that we started thinking about is this is

a lemon market, mobile health philanthropies is a lemon market.

So Doctor Elahi, my colleague, he's an ocular plastic

surgeon at Mount Sinai, he was like, “The folks who - the demands side,

the people who really need us, can’t make their needs known.

And the supply side,” surgeons like Doctor Elahi or specialists like myself,

“There's so many frictions to get engaged.

So how do you - if you start thinking

this is a market problem, how do you solve for it?”

So that was really how we got down this AI pathway.

Oops.

I think it’s gone forward. And seeing this, this data platform

sitting in the center the way you would have financial markets.

So you can have big institutional investors.

That's the Gates, the Gavi's, the governments of countries

working with the W.H.O. or the World Bank.

But you can have the day trader,

the surgical missions.

It can be then scaled, that domain of the small is suddenly scales.

And that brought us down this pathway of finding

where are these health care NGOs, where are these hospital facilities?

We assume there’d be a data set. There wasn't.

So we started using AI.

And this this work was 2020.

We started extracting from the web,

using keyword, creating a machine-learning model to extract it.

And then,

you got 20,000 human man hours.

So that's where the gen. AI becomes important.

We published that and then published, another book coming out later

this year on our approach, and put it onto this web platform.

But there's two things that I think are really important.

And John, you talked about this

is the coupling of that foundational data with these other data sets.

So we're coupling it with nightlight data.

We're coupling it with, road data.

Nurse density, nightlight, daylight,

pollution, sunlight.

And you start to get these incredible, actionable insights.

I'll just hop through that really quickly.

That's just showing our raster maps.

But the other piece

that I think is very important as you think about use cases, the partnering,

I think the enterprise-grade solution was what was so important for us.

And we've partnered with Databricks, Datarobot, Carto to build something

that's enterprise-grade, and putting in those pipelines.

And that's where the gen. AI piece comes in.

I don’t know why it’s not moving forward.

But we've created - and I'd love to talk more about that,

this lens over the web so you can extract multimodal data,

read it, create those knowledge graphs that I mentioned.

So as I understand it, what you've put together is by taking

available data on things like hospitals and capacity and so forth,

then need data on top of that, then collecting, indexing, processing,

collecting additional data on top of that, you essentially get a functional map

of where you can have the biggest impact and where the knowledge and capacity

needs to move around. Is that a fair way to think about it?

Exactly, which it’s starting to show you here. This is -

You can look at this like in an Airbnb-type model.

I can look at this specifcally.

For me and pediatrics, I’m looking in Ghana where the volunteer opportunities

and it’ll showcase for me. Now, I can look more broadly across the globe,

I can look to my needs.

I’m a physician, I wanna go to a place where I can use my Spanish,

and I want to be in an area of conflict.

I want to be in an area that’s safe.

I can target it to me.

And then the other piece that I can look, and I can apply directly and work.

So it's like the AI is now translated into a product

she can use to actively,

have impact on the ground.

And then I think this will go over to a globe mode.

Yeah.

So you can,

oh, I'm sorry, didn’t go forward,

but you can start to look at the actual medical deserts.

So I think I showed that rastor map and it’s not running I think,

which is a shame.

It can start to

show you where are the areas of need by specialty.

So, you can look at this map and see the populations

that, those are the hospitals and the NGO opportunities.

But I think it’ll plug on the right side the layers of data,

that will start to show you the medical deserts.

And we can look at that by specialty.

So access care,

that should show up,

yeah.

So the red is distance

to hospital,

the dark red people are really far in terms of access to any type of care.

Now you put on the actual capacity, that's in the green.

So you can start to target those locations, those red spots.

There's absolutely no care.

And we can slice that by specialty and sub-specialty.

And you can start to, match the resource to need.

That’s just amazing.

I mean, and the AI is helping you both identify

as well as analyze, as well as analyze data

that might have previously been unstructured or, integrable.

Absolutely.

And so the work we’re currently doing

with Databricks is generative AI work

where you can take all the data

that's created around a particular geotagged location

of a hospital, say, in Northern Ghana.

And using their data,

structure, you can have structured and unstructured data

living together.

Yes.

So the structured, we can run some of that analytics and the insights that you see,

but the unstructured can sit there, you can extract it and structure it

later if you want something else, or, what we're also putting on top of it

is a RAG, a conversational interface

where you can start asking it the questions you want answered.

Yes.

So you can torture the data all you want -

Yes.

- which is great.

Awesome. We will get more into that.

Luba.

Do you have a question?

I do have a -

Before you can ask the question, I do have a comment.

Yes.

So first of all I do want to say thank you to

Gotem and Andrew for having me here

and I have to say, my favorite part of the video is, and why

AI is great for augmenting people,

especially healthcare providers and the healthcare system,

and you can definitely tell by that video

and you in it, I love that.

And you actually being right there.

So, yes.

So, what would you like to cover?

What I'd love to cover is as you,

you've got a deep history and bringing really a whole new perspective from data

and intelligence into pharma, biopharma. And in a certain way -

and Joan and I were talking about this before - the long history of medicine

has been basically to squish out a lot of the variability.

If you go back in the history of the AMA,

it was about fighting against the patent medicine knuckleheads

and, you know, just creating, you know, care protocols

that would then get spread around and then controlling, in a certain sense,

controlling and augmenting the science of medicine.

But now we have all this new data, and in a certain sense, it's

kind of swimming upstream against that, or it's getting personalization,

it's new kinds of interpretations.

So and the kind of work that you've done in terms of

bringing data into that whole process,

talk to us a little bit about what use of why you kind of

headed down that path and what you see happening.

Sure, absolutely. And what I see as well.

So we're using - so it's interesting that you're saying data, data is something that -

and when we talk about AI, there really are two

different areas that I want to talk about.

One is the AI that we talked about all the time before, which is

you need to have the data, you need to structure the data,

you need to have longitudinal data.

You have, you know, the best thing is proprietary data

and that's all we

really focused over the last ten, fifteen years and exactly when you started out.

Gen. AI, LLMs, brought in a completely different element

which have, I would say,

stopped us from talking a lot about data, really.

I mean, data is still very important, and it's still key to

understanding and generating insights.

It certainly is very important, for example, for drug discovery.

We have a lot of data that you see in drug discovery

we’re utilizing in order to understand

protein-to-protein interaction.

We’re looking into small molecule development.

Most of the time, actually AI that we can apply right now

is mostly for development, I would say

discovery using small molecules.

So when you hear about that, AI is going to change our opinion

and understanding of biology, just because we know how one

protein interacts with one molecule

and that happens to be a small molecule,

it's a bit of a silly extraction, to now think that we will understand all of biology.

That is where we’re moving,

but we’re still on the way there.

The next frontier,

and that's a lot of work I've been doing is investing or running

and starting companies in a drug discovery space where you have

what’s called a platform, everybody likes to talk about a platform,

but, you know, at the end of the day,

you have a certain insight that you have generated from

a lot of data, usually I would say not multimodal data,

but from one area of expertise or one type of data

that you’re generating in order to understand drug discovery. We learned that

just applying AI is not a enough for drug discovery.

We now need to have AI and physics, another component.

So that's all great.

The next stage is we know that just because we can make an amazing molcule

and we know that it will bind with no problems

to a particular protein

or that it will get into a cell with no problems,

we still need to test this on animals before us humans, human animals.

And we know a lot of the discovery that we think, just because it works,

it does not really work in animals to begin with.

And then even if it does, it's actually hard to make it into a drug,

That you can drug it.

So you have to think about those aspects.

So what’s nice right now is

we are using artificial intelligence now for development.

So in the development world, we’re making an actual,

taking a molecule and making it into an actual drug and applying artificial intelligence

towards that to,

for example, predict safety, predict efficacy.

A lot of work, you see a lot of biotechs

that have been started out - biotech is, I would say is a tough

to go into. I’m invested in many of them and ran a number of them.

And a lot of the times your drug does

not necessarily work. It can be safe, but it does not mean that it’s effective.

So right now we're also using artificial intelligence to predict

just because a drug did not work, it does not mean that you need to discard it.

We’ve discarded so many drugs and what we call

“shelved it,” that's what Big Pharma does.

And that's why sometimes you have biotech companies, you have companies like Roivant

that have had that model where they in-license out something.

But this is where generative AI comes in.

That knowledge and understanding of siffting

through enormous amounts of data

and understanding, you yourself applying your own process to understanding

what is going to work in development for a particular indication, is something

that we're now using generative AI; so it’s not just, “Lets dump a lot of data,

let's make a lot of, basically,

assumptions, connections, correlations

that may give us some insight,” but it's really using generative AI

to have almost an agent.

So I’m sure you guys have heard “Agentic AI”

So having an agent to help you make -

and you, meaning drug delivery, drug discovery, a drug development person

understanding what drug is actually going to work.

So that's - on a drug discovery, drug development side

that we’ve been working on.

The second frontier,

of course, is using generative AI for medicine.

And that is, again, where we really struggled before generative AI, before

large language models to even do

simple things such as differential diagnosis.

So when a patient comes in, and by the way,

for your business, I know you guys

use it at Boston Children's Hospital.

and develop your own way of doing that.

But when a patient comes in into the hospital and says,

“I have diarrhea, I'm obese and I have a headache,”

you know, most of the time a physician does kind of what

all of you would do - actually not not even - all of you guys

would probably use Chat GPT. Unfortunately they don’t, they’re like, “I don’t know.

How about we run some tests?

And maybe a stool test, urine, blood tests, and,

I don't know, some, you know, some other test.”

And then after that, you -

but the problem is you can be ordering enormous amounts of tests

and you don’t know what to narrow down on.

So the beauty now of LLMs

is that you can actually train a model

to be the best doctor, the best primary care -

we don’t often talk about primary care,

we always talk about super-specialty and

very fancy doctors. But the ones that really make a difference are primary care.

So you can get the insights from the best primary care, train a model

to say, “You know what? You just need to run these three tests

because it probably is one of these 3 or 5 different diseases.

And then you run some of those tests and then predict what it is most likely.

So.

Yeah, it's - it's amazing.

Luba, I just want you to - just because not everybody's

in medicine, could you just give some examples of how people are taking

multimodal data or, you know, just, as I understand the,

you know, the shelled compounds, some people estimate we've only explored

maybe one percent of the available compounds that might be relevant

for human health at the most.

And the vast majority of those have been shelved.

And so if you could just just make a little more concrete. Yes.

Yeah, sure. Okay, so,

on the drug discovery, drug development side.

So the way that we make drugs is we

we usually tinker around in a lab

and we have an understanding from

a paper that you read. A lot of the times,

this is the researchers do - used to do, I don’t know with NIH cutting,

not sure.

But it used to be, that’s what basic research, basic science is for

is to talk about, let’s say, a certain protein.

And they realize that a certain protein

if it, for example, blocks a certain pathway

then it will stop your immune system,

let’s say a T cell, which is part of your immune system, in recognizing a cancer cell.

So what you’re triyng to do, is you want to, basically block that protein. You want to say,

“That’s it. I want to block that protein from blocking that pathway

because I want that immune cell to go and recognize the cancer cell to go and kill it.”

Now, there are different ways to block it. You can bind to it,

you can basically literally bind to it and block it

so it doesn’t engage with anything else.

You can degrade it. There are these things called degraders

that you can literally just degrade it and kill the protein.

There are so many different ways that now, and they're called modalities,

we call them, which is basically a different way to engage a protein

through various ways,

and also get to the protein, so that’s called delivery.

So we talk about delivery.

that’s why we like mRNA.

Why do we like mRNA?

Because that's the best way to actually go after

and engage the protein, or the target, usually they are proteins, that you want to engage.

So we have the paper that says, “This is the protein responsible

for, basically not -

blocking immune system from recognizing cancer.”

And that’s all it says. Great.

Well, so first of all,

who reads these things? I mean, by the way, they're not always published in nature

or anything like that. I mean, a lot of the times they are published in some random journals.

So you first have to find that.

Number two: let's say that you find that information,

then you say, “Okay, I'm going to come up with a molecule

that now attaches to this protein.”

And you have figured it out,

and then you said, “I am going to go after

prostate cancer because, you know, some

venture capitalist told that that’s a hot one and I’m gonna get a lot of money.”

Okay. So you're going after prostate cancer.

You're making this, it takes you about two three years to do

what’s called the development candidate which is - well actually is more like four years.

So you make this development candidate,

it’s basically the same molecule,

but now you have to attach it to the protein and you see if it works or not.

So you make it,

you’re so proud of yourself, you’re like, “Yes! Okay.”

But now you're going to put it into a mouse that basically

represents a model of prostate cancer.

So you put this into the mouse, it does not work.

Oh, this is really bad.

So, you know, what do you do?

So the nice thing about AI is that

now you can actually go in and you can -

if you had - especially if you had an agent, because if you didn’t have an agent

you have to literally go and search, search words and everything.

You have to go to Chat GPT

and by the way, if you’re not at Harvard and you don’t have the firewall,

Chat GPT gets all of the information you’re putting in

and so, you know, there's a lot of privacy, other issues that we can talk about.

Don't worry, HBS is very good.

They’re very good. So whatever you’re searching in Chat GPT is firewalled

so there’s no problem, you’re not losing any valuable information.

But you have to literally go by hand and you have to find all of these papers.

About 80 percent of the time you’re gonna get actual papers,

20 percent of it is gonna be made up stuff.

Okay, so that's called a hallucination.

So you keep on going, and yes, Chat GPT is great, you’re so excited,

but then you kind of, you get just really tired of it.

So Agentic AI is basically a way to copy you, the reserach scientist,

where all you have to do, you train it, you say, “Okay, I'm going to go after

and look at the list, the pharma - by the way, pharma, they’re not stupid, right?

They’re not gonna just publish a list for you

of all the things are called “shelved,” so basically compounds that they're not

interested in going after.

So you have to figure out some of the sources and the ways to get it.

Also, not old papers. You can’t really get access to some of the papers,

and some of the models that you can go after have those papers, some models don’t.

So Agentic AI is a way to build an agent to basically do

what you want to do and go and seach everything that big pharma has,

everything that other labs have,

everything that, maybe through publication

if somebody has publicized

and if your publicly traded company in biotech that something

didn’t work, you go through that and you get all that information,

and then you're able to actually discover something

that yes, you know doesn’t work for prostate cancer,

but now you’ve discovered that, oh my god, that actually works,

let’s say, in breast cancer.

And you know, what's most interesting is that most likely you're going to find that

this particular target and the molecule that you have come up with might work in

something like immunology or in urology, something that's completely not related.

And that's the beauty of biology.

Is that helpful?

Very helpful.

Okay.

And one thing that - for those of you who aren't following this every day

like some of us are, in the past two weeks, you might have heard of this thing,

DeepSeek and R1. Literally in the past two weeks, the quality of agents,

I like to call them digital workers, because if you think of them

as humans, that can do certain things, but they're digital,

that the quality of those has gone up

dramatically in the past twenty one days.

So you can go to Perplexity, you can go -

if you buy the $200 a month version of GPT-4 Pro,

you can put in something like, “Give me all the scientific papers on,

you know, harmful chemicals in the groundwater in the Midwest.”

And it will come up with a fifteen page,

deeply researched, with illustrations.

And we've been handing these things just as a thought experiment

to experts in the area

and we said, “Give us a grade, A, B, F, whatever.” It is, routinely coming back

with like a B, B plus rating by people who are expert in the deep field,

and it generates those in anywhere from about 10 to 30 minutes.

That's happened this month.

Okay.

And so this notion of the

all of us are going to have robots that will do stuff for us.

And I believe

HBS is doing some of this.

Jeff's doing some, Bussgang’s doing some and some others.

Part of the education process will be for you to build your own robots,

to share robots with each other and to create the cohort of robots

you will take with you in your career.

And this is - so you start thinking about that in the context of the

the complex problems you're talking about here.

It's a totally different liquidity of knowledge.

By the way, when a scientist gives a B plus to another scientist,

that’s a good thing.

That’s like an A plus in real life. - That's a that's about as high as you get.

So that’s very impressive. - Absolutely, I mean,

Scientists are like plumbers. - They don’t like stuff

they didn’t come up with themselves.

It's like every time I hire a plumber, you know, the first thing they do

when they open the wall is like, “Who did this?” Right?

Same thing with scientists.

Jones.

No, I was just thinking about what you were saying about Children’s,

I mean, we had Boston GPT stood up, John Brownstein stood it up for us

almost a year ago.

So I think it's it's amazing to see how

this is beginning to transform the clinical care.

Absolutely. I mean, what you guys do is amazing.

We talked before, too. I mean I think that Children’s Hospital

is really at the forefront.

Yeah. - Very much. Very much so.

One of the amazing things Joan was telling me

as we're getting going is that Boston Children's Hospital has a lot of,

incredible attributes.

Two of them are they have more clinical data

about children and diseases than any other place on the planet.

And they also have the most PhD MDS who understand

how to bulk through the medicine and the science behind it. Is that correct?

Yeah. I think tha’s one of the important components in all this,

which I hope came through with the global health work.

Having subject matter

experts that also work

with the data science teams and they're speaking a common language,

and I think that takes some time.

But it’s that collaborative work

that really drives the change.

And I think some of the things you and I discussed,

you know, I think of Boston Children’s Hospital where

our population of children with rare

and genetic diseases offers so much potential

when you couple that up with what Lupa is doing.

And we sit in Boston.

So with the level of expertise that we have, coupled with the financial markets,

coupled with the AI innovation that’s coming here,

that’s developed here,

you know, I visited MIT in the AI for Impact Venture Studio class

every Thursday and as we were laughing

I was like, “Oh, Liquid AI, so last year.”

You know, it's - it's changing so fast

and everybody there is in this, like, frontier leading edge.

You can

easily forget, like, the whole world has to catch up.

And I think the unique opportunity that we have here in Boston - Yes.

with the coupling of healthcare, financial markets, and AI

is incredible.

And can I also add too, Joan, you made such a good point about -

you have to have

somebody with a medical background and data science,

and the technical background all working together.

And if I may just share one anecdote. We had

this wonderful, amazing team at Google,

we have access to a lot of data, as you can imagine.

And, one day the team comes in and they said,

“Oh, we figured something out,” I said, “What is this?”

They said, “Did you know that if a 60-year-old male

has a white blood cell count

that is high at 3 AM, they’re 10 times more likely to die

than if it’s high at 3 PM?”

So I was - and I’m not a medical professional but I just have common sense,

and I’m not 22.

If you are at 3 AM, you’re probably at a hospital taking your blood,

you probably are gonna die for other reasons. But at 3 PM it’s just a regular checkup.

But, I mean, this is an extreme case of that. - So accurate. That is so accurate.

So lots of examples like this.

Exactly. - Yes.

Correlation and causation. Yeah. It's a problem.

A number of the questions, that that folks are,

are asking about is around clinic that - okay, research, get it.

Drug discovery, get it. How about clinical application?

And you know there's - there is differential diagnosis clearly being one area. - Yeah.

So differential diagnosis -

But one of the things we also talk edabout before is,

I worked on a project for diabetic retinopathy.

So in the United States and the Western World, for diabetic retinopathy,

if you have diabetes and you get diabetic retinopathy

in the Western World you don’t go blind.

Is actually a fairly easy fix.

You have some drops - - I’m sorry, Luba,

Can I just interject? Just. - Yes.

This is a...right. This is a... - An eye...

Right. A mapping of your retina.

And from that information, with the right resolution - - Yeah, yeah, no, I’m getting that,

Oh. I'm sorry. Okay right. - I’m just saying that, like, if you get,

if you get this diabetic retinopathy, it’s a disease, which is a side effect of

diabetes, - Yes

if you don’t have the drops. - Yeah.

So how do you know you have diabetic retinopathy?

You have to go to an ophthalmologist, an ophthalmologist

has, takes what’s called fundus images.

I’m sure you went to this big, very fancy machine,

and it takes a lot of time, and - and it’s a big fancy machine

which they don’t have everywhere in the world.

So one of the things that we did is,

we basically, again, we, as you were saying, we looked at the data

from fundus images and basically millions of different

points of data,

and trained a very simple algorithm to detect

just from simple photos.

And if you just stick a camera over your eye, just simply from that,

we’ve trained the machine

in order to detect that diabetic retinopathy.

And then what we did is we deployed it

in - and, well, we partnered, we didn’t do it ourselves, we partnered

with a team in India

that went into some of these deserts where

there is no hospital, no nothing, that’s - nothing there,

and then just basically set up these cameras,

and you would have lines of people that would go in, take photos of their eyes,

and then it will tell you if you have diabetic retinopathy,

and if yes, you got the drops and you wouldn’t go blind.

So that's one way and one very simple application that we currently have,

and of course, that has changed for global health purposes

that has changed lives of, you know, hundreds of thousands of people.

Other applications just in healthcare, in the hospital,

I think you wanted me to remember to say this, is which is right now,

the first application of AI, unfortunately, is not for patients.

It's not for doctors. It’s for peers.

And that is where AI right now is used,

and it is - that's the number one area is, “Hey, predicting if you’re

going to have -” Bless you. “

“If you’re gonna have, you know, even a worse outcome for some of your comorbidities.

You know, you do you cover you? Do you not cover you?” You know, these kinds of things.

And just very simple claims,

detecting if there are some false claims or double claims.

So doing that

that's not clinical care.

But that is the first application right now of AI.

The next application, again, you have to kind of

think about who's paying for these things.

So if the payer is paying, that’s the number one.

Children’s Hospital is an amazing hospital

so they actually pay for things that are really good for patients.

The majority of hospitals are paying for things that make logistics easier

and by the way it does them is much easier for patients.

I wouldn’t say it increases patient outcomes,

but increases, sometimes, patient happiness.

So, for example, traiging patients

is something that is used right now.

So if you’re coming in, into a hospital, and it can see, okay,

should you be going, should you get a bed immediately?

Because, we all know the trick, right? You want to get VC in an ER

so you’re like, “Oh, I’m getting a heart attack,” right?

Like, everybody does that.

So that - that’s not going to work.

So AI is, you know, pretty good at figuring out,

you know, do you really need that bed?

‘Cause that looked like a cut on your finger

it didin’t look like a, you know, this - this heart attack.

Yeah. So,

so it's one of these that - so that’s the next application, of course,

and that frees up the beds, and that makes things easier.

Another implication that we worked on

is actually detecting who from the ER is going to go to ICU

and another one is who in ICU is going to have sepsis,

‘cause, you know, sepsis is a very scary thing,

would lead to death pretty immediately if not taken care of.

So these are just some of the applications,

but I would love to hear from you

because I know you guys used a lot of it.

Yeah, no, similar.

I think the point you raised, one is that the first uses

and I think that goes with where the funders direct it

because the probability of success is very high

is on the business applications of it.

But I think with generative AI in particular,

it's, it's opened up a whole new vista

the things that you - you have discussed.

My, um...

colleague and direct boss, Dr. Peter Laussen,

he’s our EVP of health affairs,

is really driving

this aligned intelligence platform

across Boston Children's as we start to think about

AI integration and, and really

bringing this more into the care spectrum.

I think standing up Boston GPT

was a very important part of that step.

But there's still a long way to go, I think to your renaissance point.

We are only in the beginnings of this.

Really only in the beginnings of this.

And what, from my perspective,

when I think about it with the work with Virtue Foundation

is that it's going to democratize care more over time

because it accelerates access to knowledge and technology where,

say, I can now identify these medical deserts.

You gave the example of diabetic retinopathy.

And I get a lot of companies now, inbound,

reaching out, one of which is an AI pathology company

built out of Mass General.

Right, they’ve got this incredible tech, can read the slides

the better than a rad - better than a pathologist.

They have agents, they have Agentic AI running things in the background

so it can read the very basics, the edge cases get put forward

for a pathologist to inspect further.

This is great.

It's built for the US.

They've created a camera app

that can then attach onto a very basicmicroscope

in low and middle-income countries so that the same level of pathology skill

that one would receive at MGB could now be in

you know, northern Ghana, or eastern Ivory Coast,

you know. So being able to plug that into an AI-based

data platform like we're building to see where - where are those labs,

then they know where to target, to direct that level of technology.

So I think it's going to help

accelerate the speed at which care clinically improves

in markets that have really struggled.

Yeah this - this whole notion

of the changing and liquidity of knowledge and its applications, it's important

to remember that there are, what, about 6.5 billion of the smartphones

out there and about another billion of non-smartphones.

The refresh cycle’s about 10 to 15 percent per year.

And even the low end, Apple, it just got announced this week -

or last week - has a neural chip on it.

So and, you know, by the end of this year, we'll have a billion people

who have AI chips, or more in their phones, and those are running

all the way up to like the medium sized models, about 70 billion parameters.

So this device is changing the nature of the distribution of intelligence.

And five, you know, five, six years, we're going to have 5 billion people with it.

And so it's really important to to think about that.

And you guys may be aware of GV, which is an application

being launched in India was launched, soft launch in August, is out of Andrew Ng’s

lab in Stanford does differential diagnosis on the phone.

One of the backers of GV is the largest ambulance company in India,

and they are within 20 minutes of 920 million people.

So the idea of having this intelligence either in the ambulance

are available with that.

And then you start thinking about, okay, well,

now I can get longitudinal real world data that maybe I can't get someplace else.

And, you know, hundreds of millions of people gets very interesting very fast.

Can I just tell one quick business story, on this issue,

that I think has a kind of nice outcome, or at least I think was a nice outcome.

Turns out that most doctors spend,

four minutes with a patient and ten minutes with the computer

basically filling out the, you know, the paperwork, right?

One of the large payers put

in an ambient collection system so that the doctors, it listens

to what the doctor patient conversation has, it records that it codifies it.

And so now the instead of 14 minutes with ten minutes with the paperwork,

the office visit is seven minutes, but it's six minutes for the patient

instead of four minutes for the patient and one minute on the paperwork.

So I don't know if that's tremendously inspiring,

but it is better.

So I was just - actually I’m glad that you brought this up

because I was going to say, what are some - kind of

the hurdles to implementation of all this amazing data.

And, number one is certainly, Epic, I would imagine,

so that - that’s uh, that’s kind of number one.

So I hope somebody does some sort of regulation around that.

And because - - That’s the - that’s the data platform

that we use for patient notes. You’ve seen your physician turn around

with their back to you and start typing, it’s generally in Epic.

Yes. Exactly.

And it’s so hard because it’s made as a billing system

it’s not made as a system for you to input all of this amazing

information, data about a patient. - Right.

And a lot of the times, information in there, such as,

let’s say you come up, you say, “Okay, I’m an AI,

I’m so smart. I will make this new app.

And this new app is gonna do something very simple,”

which, by the way, no app can do right now because of Epic, which is

if you come in and you’re looking at a patient,

and, okay, the patient’s talking, you might even have

a summary of your notes because you have some of that information.

But let's say that they are taking medication,

the patient comes in and says, you know, “I have very high potassium levels.”

And the doctor’s like, “Oh my god, you’re going to die,

this is really, really bad.”

You know, the doctor’s not going into your whole, giant list

of all the drugs and all the counter indications of what’s happening.

And, you know, one of those drugs could actually be - is the reason

why you either have low or super high potassium levels.

So wouldn’t that be great?

You say, “I come up with this really cool app that I'm going to upload

all of your information about your drugs

that you’re also taking.

First of all, check for counter indications,

make sure that you can take them all.

And then see, maybe you’re coming into the hospital

because of a side effect of a drug. That seems like

a very simple thing, right? Well, good luck

incorporating that into Epic,

I mean, and into the workflow of the physician.

So that is kind of the reality and the first kind of block to this

Especially in the hospital systems.

Number two are actual computers.

So we’re talking about, we design a system that can diagnose

metastasises of breast cancer better than any doctor.

Okay, fantastic, this is great,

we have a special software for it.

All we have to do, you know, patient comes in, you have a scan,

you upload the software, you click on it

it’s gonna do the analysis for you and...

Wonderful, poof, you can see if breat cancer

metastasized or not.

Only one problem: you have about 15 minutes

to review this slide and it takes 20 to 25 minutes to upload the program or to the.

to upload the program onto your old computer.

So that is another problem.

I mean, we don’t - our computers, you know, these systems, the software is designed

in, by people that have really, really fast computers

that are also on really, really fast networks

and not tied to, you know, multi-factor authentification

all these things that, they’ll crash and everything.

So there is even simple things like that that is a challenge still.

But we’ll get through it. But, you know, that’s just the reality.

Completely agree.

I would like to ask you a question, though, because when I,

when I look at this from all the clinical years,

part of what got me interested in some of this is

the cardiac patients, seeing that longitudinal decision-making

is so important to your end outcome.

And I see, you know, so much of the tech coming out of MIT, and,

and I see a lot of startup companies, too,

who are kind of,

looking at taking this out of the medical centers.

So that when you start thinking about the primary care piece,

and Epic, and just the... - Grind.

the grind of it, you kind of see us heading towards

a world of this continual longitudinal

real-time collection of new data,

kind of the way we do it in ICU, right?

I’ve gotten my patients wired up for sound

and beat to beat, I’m collecting everything.

So - and - and that's another area in

which we're looking at AI obviously at Boston Children’s Hospital

if that’s just a treasure trove of data.

But I think ultimately we're going to get to a point where all of us

are continually censored. Like, we're - our genome is done at birth.

You can predict the directionality of your, of your, of your life

and ways in which you might choose to live your life.

And then you're collecting things in real time, to your point, John,

of your own biology,

external factors to your biology, your microbiome, all this interplay,

because AI is going to enable it.

You’re gonna talk to your doctor and as they look at your face,

you can read your blood pressure and your heart rate just directly off the image.

There’s all of this tech coming that I think we’re not

going to recognize ten years from now

where we are in terms of that clinical care.

But I’d be curious what your perspective is on that.

Yeah, no, I completely agree with you.

I mean, there’s voice, too, and you can recognize by voice

if it’s depression, and other - - Yes, voice.

there are so many markers of health. - Voice actigraphs in terms of

diagnosing your Parkinson's, your voice in terms of neuro - - Exactly.

you can predict the neurodegenerations before it's even happened.

And - and those are all fascinating, and I completely agree with you

and it’s gonna be pretty hard to integrate that.

So even though in five years, all of that -

actually, all of that technically is available now,

and it's even difficult to integrate into concierge medicine.

So there’s quite a bit of concierge medicine

here in Boston, in New York, and other areas,

and it's still difficult because you are still

relying on many of the same, very old systems.

So as soon as we can get rid of that,

that will be utilized.

But there is another issue is human behavior.

It’s actually pretty hard to change human behavior.

Just because we know and can predict

that, for example, for you in particular, you have, you know,

the kind of cardiovascular health that you can be skinny, you think you’re skinny

so this was good, but actually you have really high cholesterol

and are much more likely to have a heart attack than the next person.

But convincing you to actually do something without -

and this is what's going to happen in five years -

you're not going to want to do it.

So it’s gonna be tied to your insurance,

it’s gonna be tied to your employer.

So your employer is basically going to either

penalize you or, probably start with incentivizing you

but eventually penalizing you

for not following - because they have to pay for it.

So it's going to be there, but it’s not going

to be so rosy and, and wonderful, unfortunately.

You know, a number of the questions that are coming up on Slido here

have to do with, kind of truth and trust.

So we've been talking a lot about the potential and the longitudinal stuff and,

and the stuff that's available and voice alone.

I think if people understood would blow their minds, I would, there

there's a woman,

Rita Singh, out of Carnegie Mellon that took a sixty second sample of my voice,

and with that sixty second sample was able to generate a profile of my face,

my height within an inch, my age within a year,

my blood pressure within five points, my weight within within five pounds. And,

that's from sixty seconds of voice, personality profile, like the whole routine.

And that was six years ago.

So, so voice is amazing.

But how do we think about - is the - when - when we're getting information

from patients, how do we know they're telling us the truth?

And then when we feed it to the model, how do we know the model is

telling us the right answer?

That’s why I went into pediatrics.

Kids are lousy.

You're liars.

I love it.

No, because all the adults I took care of in medical school

were lying to me and lying to themselves.

They’re like, “I don’t drink, I don’t eat too much, I don’t - I don’t -

I exercise!” and I was like, “I don’t think so.”

In the kids, there was so much hope, you know?

They’re much more honest.

Yeah, I mean, I think it’s good.

I think that, you know, you’re only as good as your data.

And I - I do think that the - ultimately, the data that really matters,

that’s gonna - is gonna say whether or not

you’re gonna have diseases, all the stuff that you had just mentioned.

You know, all of your family history,

your genomics, you know, epigenetics, whether or not

you’re actually gonna get something

or not going to get something.

It also depends as something that you can be monitored,

right? I mean, I’m wearing all sorts of monitors here.

And - that you can then predict.

Okay, this is how much you’re walking,

this is probably how much you’re drinking,

so it's going to be much harder

to lie in the future - Yeah.

because you’ll be tracked at all times.

I encourage everybody to put a glucometer on

and just - they last ten days, two weeks,

depends upon whether you get Dexcoms or you get the Abbott version.

You put it on and you watch what happens as you just eat your diet across,

you know, two weeks. I promise you, you will think differently,

because you will recognize what each food item is doing to you.

And it's actually quite personal.

Not everybody's going to react the same way.

Like, I can eat pasta, sugar is not really going up.

You know, I eat a bowl of rice, sugar’s going up.

And then you can start thinking about, well, do I really want these insulin spikes?

I'm not diabetic, you know?

But it's - I want information around

how my body is responding to what I'm doing to it

every day.

Because ultimately, it's that daily decision that's determining my end

outcome ten years from now, twenty years from now, thirty years from now.

So I'd rather make better decisions and have data

that's going to help drive my decision making.

I'd be happy to sign up for it all right now

for me personally. Other people may not feel that way.

But I like being able to action on it. I'd rather know

and from all my experience of taking care of patients,

if you come to me early or you come to me late, my ability

to get you to a good location for your life is much better

if I see you early.

It's much harder to pull you back from the brink.

So having that data that we can action on,

and I think these are the things that AI is enabling,

it’s - it is transformative.

There are going to be a whole host of issues.

Yeah.

Trust, payers, insurance, all the things that you just wish

weren’t so in terms of problems.

But I think we fundamentally have to work through them

and we will end up moving in that direction no matter what.

I just think it's - it's moving, the train’s moving.

Yeah, no, I completely agree with you, and we were actually - did the DexCom

together with Verily, so we did that, which is the - continuous glucose monitors

so absolutely, highly recommend it.

What is very interesting is, I know and I'm sure you know

a lot of people that are trying to “hack” biology

and we all are trying to have, you know, a biological age that is much younger

than we all are.

And, you know, they're - they're using

the continuous glucose monitors and they’re using everything

you know, taking Metformin and all sorts of,

you know, CoQ10 and all sorts of

exciting things, and monitoring everything.

They all, I mean, again, it’s like

N100 but they all look pretty unhealthy to me.

You know? I don’t know if you’ve noticed the same,

they - they look kind of sickly, I don’t - I don’t know what it is.

They’re - they’re so worried about their health, and...

They’ve biohacked to a point where they don’t necessarily look normal.

Yes. So be careful. - Yes. Yes. Yes.

But I’m not saying that. I’m just saying that - - No, I know, I know.

I love the continuous glucose monitors. - No.

Is Jack Dorsey looking good? No, I don’t know.

Yeah. But I think that it’s that collection of

large amounts of longitudinal data.

And then you can put analytics on top of that.

We're going to derive insights that we couldn’t

see before. We can diagnose things earlier than we could before.

It's opening up opportunities.

Absolutely. And I mean, the one thing that I would like

just for - especially if you’re not in healthcare, um.

But all of us go to the doctor, I mean

some of the things that we use today

I’m sure probably all of you are already using,

but you have to know how to advocate because it’s gonna

take us a while to fix the healthcare system, unfortunately.

And unless - again, I just love Boston Children’s, they’re amazing -

but unless you’re going there, it’s - you have to advocate

even when you go to some of the top places,

you just have to advocate for yourself.

And the only way you can advocate is, first of all, is

you know, if you have any aches and pains

or anything, and you’re going to the physician

just do what Agentic AI would do on their end

which it’s not doing, you can actually do it.

So what you would do is,

let’s say you have some tests, and recent tests, or right now

if you go into - if you’re part of Patient Gateway or something like that,

download all those tests and then put them through Chat GPT

and say, “Okay, now I’m having these sorts of pains and aches

and pains, and this is what’s happening. And by the way, I just

started on these two or three other medications.”

Upload all your medication data.

It does wonders. I mean, you can do differential diagnosis

yourself today. - Yes.

So you don’t actually even need somebody to do that for you.

If you don't - you know, I have parents that don’t speak English.

You know, there’s so many apps and ways to also

do the same thing but with translations so that it’s

real-time translation, it can help you

So it's amazing what we can actually do today.

A lot of the times, also, you know, if you're in a hospital,

you know, and again, a hospital, sadly, you know,

with elderly parents, or if you’re in a hospital with children

you’re stressed, you're waiting to get your test results.

You will get the test results before the physician would get it.

Yes.

So you just go in, you upload it,

again, Chat GPT it, so when the physician

finally, an hour and a half later,

comes and says what you already know, you can actually -

you can even ask Chat GPT, you either come up with the

questions yourself or you say, “Okay, you know,

the ER doctor is coming, the pediatrician is coming

you know, the caridiologist is coming, what do I ask based on this?”

So I would say over the next five years is probably,

or at least three years, that’s the best that we can do,

is just advocate for ourselves and our own health.

I'm already seeing that with my patients at Children’s.

Is that annoying?

No, no, no, it’s not.

But it's - it's amazing to see how it's empowering patients.

And I think particularly I see it a lot, you know, moms of,

patients that are, like, chronic patients, right?

They’ve been - I feel for them.

They’ve been going through health systems, actually,

sometimes all over the world.

So it's a way in which they can really advocate

for their child and, and be on top of things.

And then we can dialogue it, you know,

because as you said, not everything's going to be 100% accurate

out the models yet, but they’re moving in that direction.

The other thing I was going to ask you about is

Med-Gemini, I don’t know if you were involved with that,

with Google, I mean, I think - I presume most people

in the audience, maybe you’re familiar with it, but seeing

demos of Med-Gemini

and where that has gotten to in terms of being able to

enable diagnosis.

You know, I think

you talked about Chat GPT, but the use of Med-Gemini

by the individual in there advocating for themselves.

I completely agree. I mean, that’s, like, another level.

That’s next level.

Because I want to utilize that with our, with a platform

in low and middle-income countries, because it’s like,

in your pocket. - It’s amazing, yeah.

It’s amazing, diagnostician, it’s like a position in your pocket.

One of the questions in here is will AI increase or decrease health inequality.

And I would predict it's gonna increase health equality

because of the availability.

Just for those of you who aren't,

who aren't using these models on a daily basis, just a show of hands,

how many people are using, you know,

large language models of any kind, three - once or whatever, every day.

Almost everyone.

Okay, so maybe about what, fifteen, ten, 15% of folks, for those of you

the - a really important thing you have to remember,

if you're intimidated by these things, or having to use them is

you can ask the machine what it needs, you know, describe yourself,

“I've never used one of these things before. Here's what I want to do.

Tell me what I need to tell you

to get what I want,” and it will tell you what it needs.

Okay?

So don't -

as I say, the machines talk back.

Tell them your intent.

They will help you.

Talk to them in a way that will get you something.

And we've never had a technology like that, so don't don't be intimidated.

Well, you might be totally intimidated

by the whole idea, but don't be intimidated by using them.

I know we're coming toward the end time here.

I wanted to - there are two, big things I wanted to ask. First of all,

oh, and, there was a question about the monitoring

your glucose, those are glucometers, right?

You can just buy those from the pharmacy,

right?

Yes, you can buy those from a pharmacy.

I don't - I don't think you need a prescription anymore,

but I'm not 100% sure, you used to need a prescription.

There's lots of companies that also sell them online.

And lots of companies that utilize them,

you know. - Right.

Like Levels Health, or, yeah. - And now the Silk Road guy’s out of jail

you could probably get anything you want. No, the, uh,

that's another whole conversation.

What are things that people should do differently

given our conversation here, whether it's about their own health, their family's

health, their community, how should they - what should they do differently?

Well, one thing I forgot to mention, which

just be mindful, and hopefully we’ll get this type of technology,

I’m actually on a board of a comapny that does something like that

but there will be more and more companies,

is the dosing. So when we find a dose for a patient for a drug,

we often exclude the majority of population,

we exclude many women, minorities, and, you know.

So think when, and, if you’re not 250 lbs male, white male,

just think about it when you’re given,

you know, multiple horse pills of something to take.

Just think, is that the right thing for you?

So hopefully - just, again, awareness

and this is somewhere where AI is going towards

which is optimal dosing for you as an individual

and the more data that we’re collecting, the better it’s going to become.

So even if something is already - has been labeled and

has that dose, there is an opportunity to actually

do something for you as an individual.

So I completely forgot that, but it’s an important thing.

I’m not advocating, “Do not take the dose that your doctor has prescribed,”

but just think about it.

You’re the docotor. I would be curious what you think.

No, I - it’s a great point because in pediatrics, we dose by weight.

And I used to - my mother would always ask me

to come with her, because she had some complex needs,

to explain why she couldn't take the adult dose,

and it always was dumbfounding to me that we didn’t

take a similar approach, because a 50 kilo woman

and a 120 kilo man are not the same in terms of

dosing medications. But I guess

the point you made and the point I would say

to everybody is take agency of your health.

I think if Covid taught us anything, it's that.

You know, don't hand it over.

You know, you're - you're like copilots in your healthcare

in terms of thinking about working with your physician.

And now you have another copilot in your pocket with AI

and access to large language models.

So I think you you can start taking control of your health

in a way that wasn't possible before.

And to work with your physicians in ways that you’re really

thinking about it and taking agency for yourself.

Yeah, I would say to folks to, to pay attention to the edge cases.

I think we're going to see innovations in care from the developing world,

from rural health, for things that are,

I think of a hospital like an old mainframe computer.

Mainframe computers are still around, but they're not the center

of our interaction with information anymore.

And I think hospitals are going to be like that, right?

That they're going to be the mainframes and the whole ecosystem of many computers

and personal computers and whatever, smart televisions, right, that we're going

to have versions of that and health care in terms of care delivery.

And my guess would be that's going to happen a lot faster, in the -

I think of the edge cases -

so developing world, rural health, and so forth.

And to keep an eye out for those

because I think there'll be some interesting new models.

That won't be as constrained by a combination of Epic and the payer system,

Completely agree with you, we’re already seeing it on the ground

being in some of these countries where's no legacy system

to, like, hold them back.

So, and I think what sometimes happens,

I don't know if people know the story of the ultrasound.

GE developed the portable ultrasound that almost everybody’s had

a study done on them for something.

And they couldn't develop it here because the GE engineers were like,

couldn’t get their head around it. They made big machines that

cost, you know, $150,000.

So they had to put Innovation Hubs, they put one in India and one in China.

And they designed it

as a portable ultrasound for that market because they couldn't have

an echocardiogram, they couldn’t have the types of machines that we had.

What was interesting is, then they developed that,

and it was wonderful. It gave access to ultrasound care

to low and middle-income countries.

But it opened up a whole new market and a whole transformation

in radiology in the United States.

Right.

So I do think you're correct.

I think some of the innovations that are coming in, I -

some of the things I see when we’re on the ground

will come back and benefit our health system.

Yeah.

And the last question I have for you before we go here

is a question that, as a

grandfather, I worry about a lot, and

I get a lot of questions from executives and so forth.

What should my children learn and why?

Yeah, it’s a - goodness, it’s a great question

I mean, I think of the same for my kids.

I mean, I certainly do not want them to go into computer science.

That is not - no, and my daughter does love computational biology

and I said, you know, “Stick to biology. The computational,

not so sure.”

You know, I would say that the key fundamentals still,

I would say biology and engineering is something that we’ll still need to know.

and still need to innovate around. And having, also, high IQ.

We, you know,

yeah, some people are born with high IQ, but it’s

something that certainly can be developed and,

ability to have leadership skills.

And when I say “leadership skills” not to boss people around,

but to collaborate and get people from -

like you were saying, Joan, at the beginning -

from interdisciplinary groups to come and work together

and also understand enough, of computer science.

But you’re not gonna - the way that we build models right now

you really don’t even need to do that.

I myself don’t understand why as a - I don’t want to

offend the computer scientists here - but why would you train a machine

to literally take you out of a job?

That is, uh,

because physicians, they’re not gonna be out of jobs

you still need physicians, you still need someone to

actually physically hold your hand

and machines are still gonna have errors

and we talked about some of the trust.

I remember I worked, actually, on detecting whether or not

you will have sepsis. And when you work on something for so long

you yourself start thinking you have sepsis.

So I was very sick, like, you know, not that sick,

I’m sure, like, all of you have had, probably, a recent flu

or cough that was very very bad.

And I thought, “Oh my god, I’m dying of sepsis.”

And a friend of mine who’s an ER doctor

she literally just went and looked at me,

you know, through WhatsApp and she said,

in two seconds she said, “You don’t have sepsis.”

And so, you know, but - - It’s true.

but the app - apps that I had - - I can look at you and say that.

maybe. Maybe, I said, “It said 70 percent,”

but yeah, she said, “No.” And she said,

“The fact that you actually looked at the app,

you can say 70 percent, and you called me

and you’re holding up and you’re smiling,

you definitely don’t have it.” - Yeah. You don’t have it.

So there’s common sense, so yes, what would I learn

is continue having common sense because sadly,

that is disappearing very quickly.

Yeah. Yeah, definitely.

And, um, a colleague who is in AI research in exactly this space

came up with what do we need for our children.

And it was creative, adaptive problem-solvers.

Those were the three key elements

that this next generation was going to need.

Because so much is changing so quickly,

so you do have to be able to think creatively

because, as you said, you may be out of a job but now you have to think about a new one.

So you have to be adaptive.

And it's problem-solving skills, which does involve STEM skills.

And I think those are the things I think about with my own children, preparing them

for that world, that they’re entering that is so different and.

will be so different.

Yeah, my - - Sorry, there is something else

other than STEM? There’s, like, another...?

Right. What else - what else is there?

Yeah.

My, I, I think about it a lot and the, the,

my current answer is, two things.

First of all,

I think that, I think literacy of how to interact with these robots, build them,

use them, share them, learn from them is going to be as important

as literacy, literacy.

So I think that we have to,

I want a new Second Amendment right, the right to bear AI.

um you know, and that I think

we need to teach our children how to use these things.

That's one huge thing.

It's essentially Socrates at scale, you know.

Right.

And having Socrates as a tutor must have been pretty good.

I mean, Aristotle did a good job with Alexander the Great.

I guess the, there's that.

I think the other thing in this, I think, has something

to do with the common sense is to,

to take digital fasts, to desperately seek analog.

I think that we have to get I mean, the, the neuroscience

and so forth is very clear. 72 hours in nature.

Your body changes radically, your cognition changes radically.

And if we don't have those, you know, digital fasts,

I think we have real problems.

So anyway, that's, anyway, my current answer.

Love the robots and get away from them. So.

Well, I hope you join me.

Thank you so much for taking the time.

This wonderful - thank you.

Click on any text or timestamp to jump to that moment in the video

Most transcripts ready in under 5 seconds

One-Click Copy125+ LanguagesSearch ContentJump to Timestamps

Paste YouTube URL

Enter any YouTube video link to get the full transcript

Most transcripts ready in under 5 seconds

Get Our Chrome Extension

Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.

Add to Chrome — Free

Works with YouTube, Coursera, Udemy and more educational platforms

Get Instant Transcripts: Just Edit the Domain in Your Address Bar!

YouTube

←

→

↻

https://www.youtube.com/watch?v=UF8uR6Z6KLc

YoutubeToText

←

→

↻

https://youtubetotext.net/watch?v=UF8uR6Z6KLc

YouTube Transcript:How Generative AI Is Reinventing Global Healthcare

Video Transcript

Paste YouTube URL

Transcript Extraction Form

Get Our Chrome Extension

Get Instant Transcripts: Just Edit the Domain in Your Address Bar!

YouTube Transcript:
How Generative AI Is Reinventing Global Healthcare