February 20, 2025 • 40 mins
Transforming Pediatric Healthcare with AI: An Interview with Dr. Timothy Chou
In this episode of The Human Code podcast, host Don Finley interviews Dr. Timothy Chou, a pioneer in cloud computing and AI-driven healthcare. They discuss Dr. Chou's journey from his work at Oracle and Stanford to leading the Pediatric Moonshot initiative, which aims to revolutionize pediatric care globally through AI and real-time data. Dr. Chou shares his insights on the progression of AI from automation to real-time decision-making and the impact of moving applications to the data rather than vice versa. They explore the challenges and opportunities in addressing pediatric healthcare shortages, the potential of federated and distributed AI, and the vision of creating real-time, privacy-preserving AI applications for healthcare. The conversation delves into the practical applications and future potential of AI in detecting, diagnosing, and treating rare diseases, emphasizing the importance of early intervention and prevention in healthcare.
00:00 Introduction to The Human Code Podcast
00:49 Meet Dr. Timothy Chou: Pioneer in AI and Healthcare
01:18 The Pediatric Moonshot: Revolutionizing Pediatric Care
02:14 Timothy's Journey: From Stanford to Global Healthcare
03:33 Challenges and Innovations in AI-Driven Healthcare
06:46 The Vision for AI in Medicine
07:53 Addressing Pediatric Healthcare Shortages
13:07 The Future of Healthcare: Early Detection and Investment
16:44 The Role of AI in Mental Health and Drug Trials
32:01 Building a Distributed AI Laboratory
38:37 Final Thoughts and Wisdom from Dr. Timothy Chou
39:44 Closing Remarks and Sponsor Message
Sponsored by FINdustries
Hosted by Don Finley
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Don Finley (00:00):
Welcome to the Human Code, the podcast where
technology meets humanity, andthe future is shaped by the
leaders and innovators of today.
I'm your host, Don Finley,inviting you on a journey
through the fascinating world oftech, leadership, and personal
growth.
Here, we delve into the storiesof visionary minds, Who are not
only driving technologicaladvancement, but also embodying
(00:23):
the personal journeys andinsights that inspire us all.
Each episode, we explore theintersections where human
ingenuity meets the cutting edgeof technology, unpacking the
experiences, challenges, andtriumphs that define our era.
So, whether you are a techenthusiast, an inspiring
entrepreneur, or simply curiousabout the human narratives
(00:44):
behind the digital revolution,you're in the right place.
Welcome to The Human Code.
Today we welcome Dr.
Timothy Chou.
A pioneer in cloud computing, AIdriven healthcare, and business
transformation.
As the former president ofOracle's cloud computing
business and a Stanfordlecturer, Timothy has shaped the
evolution of enterprisetechnology.
(01:06):
Now he's leading the PediatricMoonshot, a bold initiative to
revolutionize global pediatriccare by using AI and real time
data to improve outcomes, lowercosts, and reduce healthcare
inequalities.
In this episode, we dive intothe progression of AI and
organizations from automation toreal time decision making.
(01:26):
must move to the data, not theother way around, and how this
shift is transforminghealthcare.
How AI can address pediatrichealthcare shortages and improve
global access to care.
Timothy's vision is bothdaunting and doable.
And today we're exploring how AIis shaping the future of
medicine and decision making.
Let's get started on the humancode.
(01:48):
I am here with my new favorite friend, Timothy
Chou.
I got to say, we've had such afun time on the pre show talk
that I'm excited to hear whatyou have to say.
You've got a lot of inspirationin you.
and there's a lot of beautifulthings that you've got going on
in this world.
So really appreciate having youhere.
And then again, I just got toask you, what got you interested
(02:10):
in the intersection of humanityand technology?
Timothy Chou (02:14):
I think it happened.
I started a class at Stanford,on cloud computing many years
ago.
And, I always do in my class, Ido the first and last lecture
and in between, I have guestlecturers who are all CEOs of
public companies.
in my last lecture, I end withone phrase and I say, to those
who much is given, much isexpected, not that I quote the
(02:38):
Bible that often, but I wantedto get across to them.
I mean, I'm sitting in front of150 Stanford kids who, yes, much
has been given.
I mean, the fact that they'reeven there is much has been
given and that they have aresponsibility to do something
with that is what I wanted toleave them with.
(02:58):
And, so that's how I ended alecture, like, I don't know, for
10, 15 years.
One day I'm sitting there going,okay, what about you?
and as you heard in our premeeting, I had a very
interesting student show up inclass one year who had an MD, an
MPH, an MBA.
He's chief of pediatriccardiology at the Children's
Hospital of Orange County.
(03:19):
And Dr.
Chang, Anthony, he kind ofadopted me.
And I learned things like, Oh,wow, they're still using CD
ROMs.
Are you kidding me?
And then, and I know you'vespent time in the world of AI.
If we're going to build, youknow, trustworthy AI
applications and healthcare withparticular focus in imaging, the
(03:43):
only way we can do this is wehave to be able to take large
amounts of training data.
So that's why right around thepandemic occurring, I'm sitting
around thinking, Oh, watchNetflix all day or do something
useful, maybe, remembering whatI say to the students.
Don Finley (04:02):
Yep.
Timothy Chou (04:03):
so I grabbed a group of, we were talking about
friends who are multi talentedin engineering, marketing, et
cetera.
And we set ourselves a mission.
And our mission is to reducehealth care inequity.
Lower cost, improve outcomes forchildren locally, rurally, and
(04:25):
globally.
How are we going to do that bycreating real time privacy,
preserving AI applications?
Based on access to data in all 1million healthcare machines in
all 500 children's hospitals inthe world.
And I say to people, I call itmy last great project.
(04:48):
it's really a combination ofpeople I know, technology I
understand, and a mission whichI think is not one you would
take on if you wanted to be anentrepreneur that makes a lot of
money in five years.
But it's one that gives theanswer to the question I ask the
students too much.
I have been given a lot.
(05:08):
This is my answer to thequestion.
Don Finley (05:11):
and I'm calculating the number of miracles that have
to happen in order for yourmoonshot to go.
And it's kind of high, Like,
Timothy Chou (05:18):
Amen.
Don Finley (05:19):
and like typical startup investing, you, you look
for a startup that has maybe onemiracle that needs to occur,
Like if you have too many, youknow, how difficult of a
challenge is, but the visionthat you offer is So endearing
and everybody can see howtangibly the world would be
better if that was the case thatI imagine that you're attracting
some very, significant minds tobe a part of this mission as
(05:42):
well.
And so like, how has thatjourney been for you as far as,
you know, sharing this with theworld and you and Dr.
Chang, building this pediatricmoonshot?
Timothy Chou (05:53):
Yeah, it is.
I mean, you said it's daunting,but I also felt like it was
doable.
I mean, it's daunting, but it'sdoable.
Don Finley (06:02):
Well, yeah,
Timothy Chou (06:03):
Yeah.
Mm hmm.
Don Finley (06:06):
you have training and then you have inference.
And so federated training wassomething that like TensorFlow
has had in it for years.
So like there is research thatgoes on on how to do this.
technologically, like that isachievable.
And then inference, you know,you can distribute that as well.
but then there's the, the legalframework of, getting new
(06:26):
technology into organizationsthat are still using CD ROMs.
Timothy Chou (06:30):
Tell me about it.
Don Finley (06:32):
Yeah.
That's scary.
Timothy Chou (06:34):
Well, you know, as you already alluded to, we took
a step back and went, well, inorder to accomplish the mission,
Do we need to build a newrocket, which happened in the
original moonshot?
And as many of your listenersknow, the current approach to
building AI applications isreally to centralize all the
(06:56):
data, Suck it all into onegiant, computing system, learn
on it.
Right.
In the cloud, and we said thatcan't work in healthcare and
life sciences.
The data size are much larger.
Ultrasounds are a terabyte.
the security requirements, muchstricter privacy law, much
tougher Norwegian Norway sayingwe don't want Norwegian data
(07:19):
leaving Norway.
So we kind of came to theconclusion that in essence,
rather than move the data to theapplication, why don't we move
the application to the data?
And so we have built adistributed AI cloud
infrastructure for healthcareand life sciences, which
literally puts the cloud serverinside the building at
(07:42):
Children's Hospital of OrangeCounty, or inside the building
of a clinic or inside theambulance, because remember I
said we wanted to build realtime systems.
Don Finley (07:52):
Yeah.
Yeah.
Timothy Chou (07:53):
I mean, I know there's a lot of reasons to do,
you know, after the factanalytics, but that's kind of
like why, I mean, just to giveyou a little statistic in the
United States today, 60 percentof the rural counties have zero
pediatric expertise, zero.
There are only 3000 pediatriccardiologists in the United
States.
(08:14):
By the way, they live inPhiladelphia.
They live in Chicago.
They live in San Francisco.
You go to Salinas in Californiaup to Willits in Northern
California.
There's nobody there.
that's America.
You go to India, there's 300pediatric cardiologists.
You go to Africa in Rwanda,there's one guy.
(08:35):
So if you take a step back andyou go, well, oh, okay, we're
going to go build tons ofmedical schools everywhere and
train everybody.
I mean, the quote, Americansystem.
of medicine.
How can this scale?
There's no way it can scale.
You can't in, you know, if I'm afresh graduate from Stanford
Medicine and PediatricCardiology.
(08:56):
You want me to move to themiddle of Montana?
And earn 1.
98.
I mean, why would I do that?
And you just see that We couldbuild I think some people
because AI we try to put it incompetition with are smartest
people.
We, we, that's how we want tothink about it.
It's going to be better than theworld's best cardiologist or
(09:18):
oncologist or whatever.
And I'm kind of going, I don'tthink that's the point of the
whole thing, What if we couldbuild, I call them, red, green,
yellow applications.
Don Finley (09:27):
Okay.
Timothy Chou (09:28):
Red.
This kid needs to be medevacedout of here.
Or this accident victim, right,has a traumatic brain injury,
needs to go to the ER.
Red.
Green.
Hey, everything's cool.
Yellow.
We need somebody else to take alook at this.
If we could do that, we couldput the brains, in essence, of
(09:50):
the world's best oncologist,orthopedic surgeon, etc.
Right into the rural community,into India, into Africa.
Right.
And so it's not, I always say topeople, if your mom works for
Google and you live in PaloAlto, yeah, what we're working
on is kind of cool andinteresting.
(10:12):
But that's not the point of thewhole thing.
I really think that's why, AI inhealthcare, both is necessary,
but also is.
We're capable of doing that.
Don Finley (10:25):
Absolutely.
And so basically, I think yourfirst kind of iteration of this,
or like what you're talking islike in the simplest level of
the triage of basically sayinglike, Hey, what is going on
here?
What do we need to do?
And how do we pair the AI withthe team on site to basically
know, Hey, here's the situationand here's how quickly we need
to act.
Timothy Chou (10:44):
Well, also, you know, if you look at it, I think
ultimately all diseases aregoing to be classified as rare
diseases.
It's just, we don't know enough
Don Finley (10:55):
That's it.
Yeah.
Timothy Chou (10:56):
do that yet.
Right.
Let me just give you a, anexample.
There's a condition called focalcortical dysplasia.
which is actually a brainlesion.
If this lesion is leftuntreated, a kid has epileptic
seizures.
So there's a kid in Floridawho's had epileptic seizures two
to three times a day.
(11:16):
At night, he wakes up screamingfor 12 years.
I mean, think about that.
Don Finley (11:23):
Ooh.
Okay.
Timothy Chou (11:24):
MRI imaged him early on, didn't see anything,
started taking him down atreatment of drugs, some they
could afford, some they couldn'tafford.
Ultimately, they were startingto talk about, let's go put in
electrodes into his brain, Theyfinally re MRI imaged him.
They now believe that he hasthis condition and he's I
(11:46):
actually have photos or images,MRI images that I could show
you.
I mean, you and I would look atit and go, What's the difference
between A and B?
No, I can't see it.
Don Finley (11:55):
Yeah.
Timothy Chou (11:57):
But, you know, if you can find it, you can
actually surgically remove it.
And if you surgically remove it,the kid's cured for life.
Amazing, right?
Okay, there's a good news andbad news story here, The good
news is there's only 25, 000cases of this a year in the U.
S.
Good news, right?
Don Finley (12:18):
Okay.
Timothy Chou (12:19):
Bad news, no one pediatric neuro oncologist ever
sees enough of this to be ableto recognize it.
Don Finley (12:27):
Exactly.
Yeah.
Timothy Chou (12:28):
now, on the other hand, we all know over in the
world of AI that, damn, if Icould get all the MRI images of
all the kids in the world, Icould build a focal cortical
dysplasia diagnostic all day andall night.
and you can repeat, I think youcould repeat thinking about the
world this way.
Over and over and over again.
(12:49):
It's not, we will end up, like Isaid, I think all diseases will
end up being rare diseases.
It's just we don't know enoughyet.
Don Finley (12:59):
That is probably like, it's a very profound
statement to say that like alldiseases are rare diseases.
And I think that is the paththat we're going down.
which gets me thinking aboutlike, what does that world look
like after that fact, how do yousee your work impacting the
healthcare space?
what is the future state thatyou want to have, as far as the
(13:20):
experience of people goingthrough this?
Timothy Chou (13:23):
Well I think, you know, there's better people to
comment on the healthcaresystem, which I'll comment is
totally broken.
but I think We all have to getto, and this technology is part
of it, is how do we investearlier?
Earlier meaning, you know, weall invest in the last 10 years
(13:46):
of your life, we invest infiniteamounts of money, which are in
essence the end of productivityfor that individual.
We do not invest early, We'remaking a difference in a 10 year
old kid's life.
He has 80 years, 90 years ofproductivity in front of him.
We say kids are our future.
(14:08):
We don't spend that way at all,You want to go do a startup in
pediatric health care?
Are you crazy?
Why would you ever do that?
There's not enough market.
I mean, we say kids are ourfuture.
Don Finley (14:23):
Yes.
Timothy Chou (14:24):
So, we need to invest earlier from the
standpoint of, we'll call it thepopulation.
But also, why are we notinvesting?
Why do we wait until just usingthis kid?
We're waiting for 12 years tofigure out what's really going
on here.
That's how the whole system isbuilt.
We're built for the acute issuethat happens, which either
(14:47):
you're going to go take a drugfor it, or get surgery for it,
That's how the whole thing isbuilt.
Why are we not detecting issuesway earlier in the cycle?
The technology is capable ofdoing that and beginning to
alter what is the other half ofthis, which is lifestyle there.
There's a condition in kids.
This is a kind of a, other typeof story.
(15:08):
kids that have cancer, goodnews.
They are given, you know,chemotherapy, et cetera.
Their cancer is cured
Don Finley (15:15):
Okay.
Timothy Chou (15:16):
news.
There is a tendency that whatends up happening is it affects
their heart.
And so while they survive thecancer, they die of heart
disease.
Now, if you can detect thisearlier, the beginning of this,
there's all sorts of, I mean,back to diet, exercise, all
(15:36):
these sorts of things that willprevent this from happening if
you knew earlier in the cycleuntil, right?
So, I'm just saying, I think theearlier we can know, the earlier
we work on these problems, thelower the, fundamentally, the
lower the cost of the system.
Fundamentally, the better theoutcomes in the system.
(15:59):
And if we can use technology todo this, where, you know, I was
at JPM yesterday.
I was like, you know, thebeauty, a lot of people work on
medical devices and pharma, Isaid, the beauty of software is
in essence, the cost to build itis zero and the cost to deliver
it is zero.
I said, the impact the softwarecan have is huge.
(16:21):
Because I don't have the cost ofdistribution.
I mean, or my cost ofdistribution is very low.
Right.
And I think that's wherehopefully, you know, maybe not
in my lifetime, but hopefully weall get here where we can do
that sort of thing.
Focus earlier in the cycle.
Don Finley (16:43):
There's so many things you touched on.
I know we didn't touch on thisearlier, but like, I'm a big
mental health advocate.
And so one of the things thatI'm supporting is like the MAPS
institution, who is like,They're running the trials for
MDMA to treat PTSD and severedepression.
Timothy Chou (16:58):
huh.
Don Finley (16:59):
And so one of the challenges that they've come to
in their phase three trials,which they just in the last few
months got rejected by the FDA,two problems with MDMA.
One is you can definitely tellwhen you've had it.
So it's really hard to do adouble blind, study with the
patient.
The second thing is thattreatment is necessary to go
(17:20):
along with the substance.
And so, the FDA doesn't approvetreatments, they approve drugs.
And they can't approve a drugthat requires a treatment that
goes along with it.
The efficacy of it has to standalone.
And so our entire system is,MAPS is showing basically, after
I think like five years, they'reshowing an 85 percent
effectiveness of the treatmentsof like a six month treatment
(17:44):
protocol.
And so like they're showing Lessthan 20 percent are going back
to depression or like havingsigns of PTSD.
And yet at the same time, oursystem that is in place is
really supportive of, is it adrug or is it a medical device?
And if it's a medical device, isit similar to something that is,
(18:05):
you know, You know, old or ifit's completely new, that's an
entirely different process.
But like we, we have patternsand systems that are set to
basically treat symptoms insteadof getting to that early upfront
kind of approach.
And I know that we talked aboutlike triage of somebody who is,
you know, has a trauma headwound, you know, they need to be
medevacked out.
(18:26):
How do you see the pediatricmoonshot playing in that space
of prevention or early kind oflike lifestyle diagnostics?
Timothy Chou (18:36):
if people are interested, we're actually
tracking 140 different, deeplearning imaging applications in
adult and pediatric medicine.
You will not be surprised, thisis at appcommons.
bevelcloud.
io and I know you can edit intoshow notes.
Don Finley (18:55):
Yeah.
Timothy Chou (18:56):
You will not be surprised.
They're just a handful.
of work in pediatrics, And, letme just, you know, if the
listener doesn't really realizethis, pediatric cardiology and
adult cardiology basically don'thave much similarity.
the phrase that's used in, inthe world of pediatrics is,
children are not little adults.
(19:19):
an And this is the rationale isthat the things that cause
mortality and cause issues inkids, which are unresolved, they
die.
So the population of us oldpeople who are smoking and
drinking too much or whatever,the pathologies that we create
are very different than whathappens in children.
Don Finley (19:41):
Ah, true, true.
Timothy Chou (19:42):
Right.
So number one, I think it's justlet's go pay.
In fact, I just spent time withthe head of neonatology at
Stanford.
There's a condition in apremature baby where this is
interesting.
A chemical gets activated in ababy, normal babies gets
activated, which in essencecauses the lung to inflate and
(20:05):
he used a great example.
He says, when you blow up aballoon, The very first puff,
you really have to force it toopen up.
That's exact because the baby'sbeen right underwater.
That exact mechanism has tohappen when a baby's born.
a premature baby, it doesn'thappen in a lot of cases.
(20:27):
But at this point, they actuallyknow this and they have an
artificial synthetic whateverdrug that causes this to happen,
Don Finley (20:36):
Ah, okay.
Timothy Chou (20:37):
So, I mean, just, that's just another example of
that never happens in adults.
Don Finley (20:42):
I don't know, Timmy.
You've never woken up and beenlike, oh, I gotta breathe
Timothy Chou (20:45):
yeah, I gotta breathe now.
Don Finley (20:47):
Gotta breathe now.
Timothy Chou (20:49):
Yeah,
Don Finley (20:49):
Gotta start my day breathing.
Timothy Chou (20:51):
yeah.
And I think the challenge inpediatrics is, I mean, good
news.
Most kids are healthy.
I mean, that's the good news.
So all of us, our perception is,Kids are healthy because, But
the reality is there's a lot of,I'll touch on another one.
I've spent time with the head ofrare diseases at, UPMC.
(21:13):
and in that usage of the word,rare disease is super rare.
And so he's been doing a lot ofwork in genetics.
And so he told me, he said, youknow, a couple of years ago, he
gets called into the adultdepartment.
There are three patients who arein intensive care.
They cannot figure out what'sgoing on.
(21:35):
He goes in and genetic teststhem and comes back with
actually four diagnostics, whichwere pediatric rare diseases.
Which had been undiscovered inthe adult world.
Don Finley (21:50):
Oh.
Timothy Chou (21:52):
So the, you know, we, the greater we, right.
We say, you know, children areour future.
I mean, in a hundred ways,that's true.
And so, I always say to people,if you're interested in what
we're working on, we have tonsof ways to get you involved.
But the more important thing islet's put some attention on
(22:13):
this.
as a society.
And by the way, not onlyhealthcare, but education would
fit into this category as well.
You know, that's the seed corn.
You know, one of the reasons Ilike teaching is you know, you
if you can, I like to say I havethe opportunity to shape young
(22:35):
minds.
it's a responsibility and ablessing at the same time,
right?
Don Finley (22:40):
I, look at it as kind of like, you know, you're,
you're tending a garden, right?
Like you want the flower tobloom in its own way, but at the
same time, you're there tosupport that growth.
Timothy Chou (22:50):
Yeah.
And we're starting to work inover in oncology and adult
oncology.
What we're talking abouttranscends, I mean, all the work
we're doing.
I always say to people, ourcomputers actually have no idea
if they're in a children'shospital or a clinic or an
adult, they don't know really.
Um,
Don Finley (23:06):
yeah.
No!
Timothy Chou (23:10):
cancer, one of the things that happens in, in more
complex cancer cases is thatthey convene what is called a
tumor board.
So because cancer has thismultifaceted side to it, like,
well, I need a pathologist thatlooks at the slides.
I need a radiologist that looksat, you know, the, the lung
(23:31):
images.
I need a radiation oncologistwho knows how to irradiate that.
I need a medical oncologist whoknows all the, And the
incredible number of drugs thatare now being developed, which
are, basically call it genespecific.
Okay.
At the big fancy institutionslike Stanford over here, or you
(23:52):
go to Dana Farber, they have allthese experts and they
literally, I mean, this is kindof fascinating.
They literally glow into a room.
It's like a meeting room withPowerPoint.
And in a one hour period, or oneor two hour period, they will
give every patient five minutes,these complex patients.
Don Finley (24:13):
Okay.
Timothy Chou (24:14):
Prevent the case, et cetera, et cetera.
And then have this conversation.
You look at that.
First of all, that only happensat the fancy places.
You go get cancer in, you know,Western Montana or whatever.
What, what tumor board,
Don Finley (24:31):
Yeah.
Timothy Chou (24:33):
But yeah, and I know you guys have been working
in multi agent, the multi agentworld, if you sit down and think
about it, couldn't you build amulti agent tumor board Right?
You have an expert in the image.
You have an expert in thepathology.
You have an expert in, and atthis stage, we're also spending
(24:56):
time over in the drug trialworld.
You look at it.
I mean, there's so manydifferent drugs coming out.
How does a oncologist, even ageneral oncologist, in the
middle of the country know, it'snot their full time job, the
whole range of what's availableright now?
You
Don Finley (25:15):
And, that's, the thing.
The amount of research that'sgetting released on a daily
basis is more than you can readin a month.
And so, and then to do the jobthat you're doing, there's no
way for anybody in thatprofession to keep up with where
they're at.
Plus, if you have a tumor board,Any collection of humans kind of
has like the dynamics of a humanrelationship.
(25:37):
And so, you've seen the study oryou've heard this, I think it
was a study, about judges as faras if you get, if you have your
sentencing, before lunch,you're, you're done.
Right.
But if you get it, if you'rethere after they eat, you're
going to get a lighter sentence.
Right.
And I think that we'd, we'd seethe same sort of results in any
sort of dynamic when it comes tolike an oncology or a tumor
(26:02):
board.
I love that name, by the way.
So,
Timothy Chou (26:06):
By the way, just to make a comment, tour boards
are volunteer efforts.
The docs do not get compensatedfor this.
I mean, it's back to thisfundamental, we have a broken
system.
I mean, just
Don Finley (26:19):
I mean, that's, I mean, another example of, like,
how broken the system is.
Now, are there other countriesthat you see doing this, like,
better?
That we could, we could gaininspiration from?
Timothy Chou (26:34):
well, just to make a point of it, I think we're all
on the bleeding edge of what AIand healthcare, and when I use
the term, I think a lot ofpeople, the word software and
the world AI areinterchangeable.
Software could do a lot of coolstuff.
You know, I've been in thebusiness for a lot of years.
but if we're talking about deeplearning, neural networks, et
cetera, we're really still atday one, day two of this whole
(26:57):
thing, right?
Of which ChatGPT exposed us allto the potential of it.
and that's, Really text data,imaging data where, you know, I
don't know we're first minute orsomething.
So I think to tell you there iswork.
we're tracking work and spendingtime with folks in the UK that
(27:18):
are working in this area, folksin Germany that are working in
this area.
So there are other places,particularly when you talk about
distributed learning.
And, swarm learning, federatedlearning.
There are other places andparticularly, you know, Europe
because of their, I'll call itmore strict way of viewing data,
(27:39):
They're much more aggressiveabout working on how do you
build out distributedcapabilities rather than
centralizing it, at motherGoogle or whatever.
So yeah, there are, I mean, butwe're all early, very early.
I mean, there's this uniquecombination of domain expertise
(27:59):
in cardiology, orthopedics, orwhatever, and.
An understanding of being ableto build a team.
Like I said, we're tracking 140of these where I've got some guy
who knows what a convolutionalneural network is.
Right.
and putting those two piecestogether is non trivial.
(28:20):
and, but I think, we're in theearly days.
I always tell people, I was kindof there at the, we'll call it
the birth of cloud computing.
really, I mean, even AWS is,what, 15 years old?
Salesforce is 20 years old,rough 20 25 years old, but even
with cloud computing, whetherapplication or infrastructure,
(28:43):
we're still nowhere.
I mean, plenty of people arerunning on prem, traditional
software models, plenty.
I mean, you know, plenty ofcomputers are being sold that
are put in data centers.
I mean, it's not like it's allquote game set match done.
So when you look at where we arewith AI, I'm kind of like.
we're year one of a 15 yearcycle.
(29:06):
I mean, super early.
Don Finley (29:09):
Oh, I absolutely agree with you.
And I think we end up insituations where we go from
distributed computing tocentralized computing.
And like we find like we go backand forth.
and it's kind of like, even inthe software world, we go
waterfall to agile back andforth as well.
The pendulum swings.
We create new things and it'scool.
And I think we're in that, theirphase of artificial intelligence
(29:31):
being centralized.
But like people like you and,the work you're doing to
decentralize this and showcasethat it is possible to build
private systems that canactually help you and support
you.
And I think that that's a reallyAwesome push for what you're
doing.
And like, we look at even on thesocial media side, right?
(29:52):
Like centralized AI and socialmedia has meant that we are now
the product in which ouremotional engagement is what the
AI is gaining or trying to gainfrom us.
But what I like to see is nowlet's have AI that is looking at
like, what is your purpose?
What is your goals?
and then helping you to achievethose.
And I think that, you know, whatyou're working on as well fits
(30:14):
into that paradigm of givingmore control and access to the
people where they need it.
if, you know, myself and theaudience included had a magic
wand to help you get the, to thenext evolution of your pediatric
moonshot, what could we provideto you?
Timothy Chou (30:30):
I think there's multiple levels.
at one level, it's really, andyou're helping do this, is let's
keep telling the story.
I tell people it took 40, 000people to get to the moon.
We're a few shy of that.
So, spreading the word, numberone.
number two, I think for thosepeople who can engage, do
(30:52):
engage.
Meaning, you know, what does itmean to build a congenital heart
disease?
AI algorithm.
I mean, if you're in the spaceof either the world of
cardiology or the world of AI,go start looking at these
problems.
There's tons of, I said, wealready have examples.
(31:13):
I mean, let me say how specificthis can get.
There is a team at Duke workingon AI analysis of weight
bearing, low power CT imaging.
of the ankle and foot.
So normally, right, when youthink about it, you go take an x
ray of your foot, there's noload on it,
Don Finley (31:33):
Okay.
Okay.
Timothy Chou (31:34):
it's, obviously, there's information when you can
get, weight bearing.
But obviously, if you go dothat, you can't be subjecting it
to a ton of Right, radiation.
Hence, low power CT.
So I'm just using that as anexample of how specialized this
(31:54):
can get.
Low power CT, weight bearingankle and foot,
Don Finley (31:58):
Yeah.
Timothy Chou (31:59):
That.
I think, we're in the middle ofbuilding a Distributed AI
laboratory for healthcare andlife sciences.
What do I mean by that?
We are going to have 32 siteswhere all the imaging data is
available in real time, CT, lowpower, CT, ultrasound, x ray.
(32:21):
MRI, et cetera, all offline dataavailable from the Paxes, which
for the listener doesn't know isin essence, the repository that
the radiologist uses to look at,the images and then all the EMR
data available to authorizedapplications.
And we are want to perfect theart and science.
(32:42):
of taking research work, whetherthat's at University of London
or in Germany or inPhiladelphia, this small
research team, and move it from,we like to say, from the
research bench to the bedside byimplementing it in a laboratory
that sees 2, 000 terabytes ofdata a year.
(33:06):
We'll have 3, 000 servers in it.
Right.
If we can figure this out, wenow have a path to take
brilliance in low power CT forankle and foot, and now deploy
it across all low power CTsystems in the world, So we're
in the middle of that.
I've been funding all of ourefforts to date.
(33:27):
This next step is not going tobe able to be done that way.
So we are actively inconversations with a lot of
people who, you know, I will sayare interested in how
distributed AI will work.
And I'll just make a comment.
If everything moves to, well, Ijust pick on mother Google.
(33:49):
If everything moves to motherGoogle, well, they build their
own chips.
They don't buy servers fromanybody.
So there are a largeconstituency that is interested
in how does a distributed worldwork and we're very eager to
have conversations with anybodyin that space whether that's
(34:09):
from the tech front or the othermajor area we're spending time
in is our new friends inbiopharma.
Where,
Don Finley (34:18):
Bye bye.
Timothy Chou (34:19):
yeah, this is why I said we're spending a lot of
time in, you know, clinicaltrial work where today the
ability to do very preciseconnection of the patients to
the trial is done today.
I'll call it haphazard manual.
(34:39):
Do you know the right people isthe general mechanism.
So what we're right in themiddle of building is a
application.
We call it total recall, whichis connected.
Remember, we have a distributedcomputer sitting inside the
firewall so it can talk to theEMR.
So now I can take it on.
(35:00):
I'll pick on oncology.
I can take an oncology patientwho has a 3, 500 page PDF
document as Her electronicmedical record.
I can rag train a local LLM andnow I can ask questions like,
well, you know, what's thedifference between the last two
CT scans or how many courses ofcisplatin has the patient been
(35:25):
given?
What was the reaction?
What we learned, we shared thiswith 12 oncology nurses around
the country.
I told the team, I said, if wehad a buy now button, I'm pretty
sure they would have hit itbecause we didn't realize how
much manual labor they gothrough in essence with a little
search bar, looking at a giantPDF document and it's like
(35:47):
crazy, right?
so we, we actually, the reasonwe called it total recall is.
We demoed it to these oncologynurses, then we turned over the,
the controls to them to let themask whatever the hell they
wanted to ask.
So, at some point in time, oneof them said, well, what is the
prognosis for this patient?
And you're going, this is whereAI appears like magic, right?
(36:10):
So it does all the other stuff,so it must be able to answer
that.
You're like, no, You humanscan't even answer that question.
Why are we asking the computer?
So that's why we said, no, it'scalled total recall.
It has perfect recall.
of what has happened with thepatient in the past, right?
That's what it does.
Perfect recall, right?
Don Finley (36:30):
recall.
That's it.
Timothy Chou (36:31):
that's, what it does.
Don't
Don Finley (36:32):
it's
Timothy Chou (36:33):
do
Don Finley (36:33):
in the name.
It doesn't, it's notpontificating on what is
actually like going on.
It's not connecting the dots,but it's just total recall.
I love that.
Timothy Chou (36:43):
Now remember, we now, the application now has
complete profile information ofthe patient, Cause it, looking
at the MR record.
So what we have come to learn isthat in phase 2, phase 3 drug
trials, the real data about whatthose protocols are, are sitting
inside protocol documents thatcan be 300 pages of PDF and they
(37:10):
are sitting behind the firewallbecause the pharma company
doesn't want to let everybodyknow what they're working on.
Okay, if you have a distributedinfrastructure, I park a server
inside Roche, Bristol Myers,Squibb, whatever, LLM, RAG, the
protocol document, and now multiagent, I can have the total
recall agent talk to theprotocol agent, and now the hit
(37:33):
rate has got to skyrocketbecause you have now complete
knowledge of the patient andcomplete knowledge of the
protocol.
And just to say it, This.
is not that hard to do.
Don Finley (37:47):
No, that's entirely you're, you're talking about
things that we do in otherareas, right?
Like there are
Timothy Chou (37:53):
bad.
Don Finley (37:54):
of this.
And in fact, we have ragsolutions running in our own
systems.
We distributed for our clients,right?
Like all that question andanswer stuff is.
is relatively, I mean, it's nota, a solve solve problem, but
they are solvable.
They require a little, it's likethe early days of Oracle, right?
Like, you know, it was adatabase company, but they were
(38:15):
still like building and like ragsolutions are context based.
For the most part, you got tofigure out how to get the
information in, in a way thatthe machine can process it.
But at the same time, likeyou're talking about stuff that
is, you know, day oneimplementable, not.
Two years, five years down theline.
I love it.
So Timothy, I gotta wrap thisup.
(38:38):
Is there any other like lastbits of wisdom that you would
like to share with the audienceof like any life lessons as far
as how to, how to live in this,you know, future AI world that
we're talking about?
Timothy Chou (38:49):
Well, it's something I, I said to you
during the pre meeting.
When I start out my Stanfordclass, I always say to the
students, I go, you know, thequality of a student is not
measured by what you know, butby the quality of the questions
you ask.
And maybe in a world of LLMs, itis even more true.
Don Finley (39:11):
Yes.
Timothy Chou (39:11):
It's the quality of the questions you ask.
And I think the, the more we'reable to be curious, the more
we're able to ask the rightquestions, the faster this is
all going to happen, right?
And
Don Finley (39:28):
I absolutely love it.
Curiosity is one of the thingsthat drives this world and
Timothy, thank you so much forbeing on the show today.
Like I really appreciate thetime we've had together.
Timothy Chou (39:40):
I have as well, Don, it's been a lot of fun.
Don Finley (39:43):
Absolutely.
Thank you for tuning into TheHuman Code, sponsored by
Findustries, where we harness AIto elevate your business.
By improving operationalefficiency and accelerating
growth, we turn opportunitiesinto reality.
Let Findustries be your guide toAI mastery, making success
inevitable.
(40:04):
Explore how at Findustries.
co.
Welcome to the Human Code, the podcast where
technology meets humanity, andthe future is shaped by the
leaders and innovators of today.
I'm your host, Don Finley,inviting you on a journey
through the fascinating world oftech, leadership, and personal
growth.
Here, we delve into the storiesof visionary minds, Who are not
only driving technologicaladvancement, but also embodying
(00:23):
the personal journeys andinsights that inspire us all.
Each episode, we explore theintersections where human
ingenuity meets the cutting edgeof technology, unpacking the
experiences, challenges, andtriumphs that define our era.
So, whether you are a techenthusiast, an inspiring
entrepreneur, or simply curiousabout the human narratives
(00:44):
behind the digital revolution,you're in the right place.
Welcome to The Human Code.
Today we welcome Dr.
Timothy Chou.
A pioneer in cloud computing, AIdriven healthcare, and business
transformation.
As the former president ofOracle's cloud computing
business and a Stanfordlecturer, Timothy has shaped the
evolution of enterprisetechnology.
(01:06):
Now he's leading the PediatricMoonshot, a bold initiative to
revolutionize global pediatriccare by using AI and real time
data to improve outcomes, lowercosts, and reduce healthcare
inequalities.
In this episode, we dive intothe progression of AI and
organizations from automation toreal time decision making.
(01:26):
must move to the data, not theother way around, and how this
shift is transforminghealthcare.
How AI can address pediatrichealthcare shortages and improve
global access to care.
Timothy's vision is bothdaunting and doable.
And today we're exploring how AIis shaping the future of
medicine and decision making.
Let's get started on the humancode.
(01:48):
I am here with my new favorite friend, Timothy
Chou.
I got to say, we've had such afun time on the pre show talk
that I'm excited to hear whatyou have to say.
You've got a lot of inspirationin you.
and there's a lot of beautifulthings that you've got going on
in this world.
So really appreciate having youhere.
And then again, I just got toask you, what got you interested
(02:10):
in the intersection of humanityand technology?
Timothy Chou (02:14):
I think it happened.
I started a class at Stanford,on cloud computing many years
ago.
And, I always do in my class, Ido the first and last lecture
and in between, I have guestlecturers who are all CEOs of
public companies.
in my last lecture, I end withone phrase and I say, to those
who much is given, much isexpected, not that I quote the
(02:38):
Bible that often, but I wantedto get across to them.
I mean, I'm sitting in front of150 Stanford kids who, yes, much
has been given.
I mean, the fact that they'reeven there is much has been
given and that they have aresponsibility to do something
with that is what I wanted toleave them with.
(02:58):
And, so that's how I ended alecture, like, I don't know, for
10, 15 years.
One day I'm sitting there going,okay, what about you?
and as you heard in our premeeting, I had a very
interesting student show up inclass one year who had an MD, an
MPH, an MBA.
He's chief of pediatriccardiology at the Children's
Hospital of Orange County.
(03:19):
And Dr.
Chang, Anthony, he kind ofadopted me.
And I learned things like, Oh,wow, they're still using CD
ROMs.
Are you kidding me?
And then, and I know you'vespent time in the world of AI.
If we're going to build, youknow, trustworthy AI
applications and healthcare withparticular focus in imaging, the
(03:43):
only way we can do this is wehave to be able to take large
amounts of training data.
So that's why right around thepandemic occurring, I'm sitting
around thinking, Oh, watchNetflix all day or do something
useful, maybe, remembering whatI say to the students.
Don Finley (04:02):
Yep.
Timothy Chou (04:03):
so I grabbed a group of, we were talking about
friends who are multi talentedin engineering, marketing, et
cetera.
And we set ourselves a mission.
And our mission is to reducehealth care inequity.
Lower cost, improve outcomes forchildren locally, rurally, and
(04:25):
globally.
How are we going to do that bycreating real time privacy,
preserving AI applications?
Based on access to data in all 1million healthcare machines in
all 500 children's hospitals inthe world.
And I say to people, I call itmy last great project.
(04:48):
it's really a combination ofpeople I know, technology I
understand, and a mission whichI think is not one you would
take on if you wanted to be anentrepreneur that makes a lot of
money in five years.
But it's one that gives theanswer to the question I ask the
students too much.
I have been given a lot.
(05:08):
This is my answer to thequestion.
Don Finley (05:11):
and I'm calculating the number of miracles that have
to happen in order for yourmoonshot to go.
And it's kind of high, Like,
Timothy Chou (05:18):
Amen.
Don Finley (05:19):
and like typical startup investing, you, you look
for a startup that has maybe onemiracle that needs to occur,
Like if you have too many, youknow, how difficult of a
challenge is, but the visionthat you offer is So endearing
and everybody can see howtangibly the world would be
better if that was the case thatI imagine that you're attracting
some very, significant minds tobe a part of this mission as
(05:42):
well.
And so like, how has thatjourney been for you as far as,
you know, sharing this with theworld and you and Dr.
Chang, building this pediatricmoonshot?
Timothy Chou (05:53):
Yeah, it is.
I mean, you said it's daunting,but I also felt like it was
doable.
I mean, it's daunting, but it'sdoable.
Don Finley (06:02):
Well, yeah,
Timothy Chou (06:03):
Yeah.
Mm hmm.
Don Finley (06:06):
you have training and then you have inference.
And so federated training wassomething that like TensorFlow
has had in it for years.
So like there is research thatgoes on on how to do this.
technologically, like that isachievable.
And then inference, you know,you can distribute that as well.
but then there's the, the legalframework of, getting new
(06:26):
technology into organizationsthat are still using CD ROMs.
Timothy Chou (06:30):
Tell me about it.
Don Finley (06:32):
Yeah.
That's scary.
Timothy Chou (06:34):
Well, you know, as you already alluded to, we took
a step back and went, well, inorder to accomplish the mission,
Do we need to build a newrocket, which happened in the
original moonshot?
And as many of your listenersknow, the current approach to
building AI applications isreally to centralize all the
(06:56):
data, Suck it all into onegiant, computing system, learn
on it.
Right.
In the cloud, and we said thatcan't work in healthcare and
life sciences.
The data size are much larger.
Ultrasounds are a terabyte.
the security requirements, muchstricter privacy law, much
tougher Norwegian Norway sayingwe don't want Norwegian data
(07:19):
leaving Norway.
So we kind of came to theconclusion that in essence,
rather than move the data to theapplication, why don't we move
the application to the data?
And so we have built adistributed AI cloud
infrastructure for healthcareand life sciences, which
literally puts the cloud serverinside the building at
(07:42):
Children's Hospital of OrangeCounty, or inside the building
of a clinic or inside theambulance, because remember I
said we wanted to build realtime systems.
Don Finley (07:52):
Yeah.
Yeah.
Timothy Chou (07:53):
I mean, I know there's a lot of reasons to do,
you know, after the factanalytics, but that's kind of
like why, I mean, just to giveyou a little statistic in the
United States today, 60 percentof the rural counties have zero
pediatric expertise, zero.
There are only 3000 pediatriccardiologists in the United
States.
(08:14):
By the way, they live inPhiladelphia.
They live in Chicago.
They live in San Francisco.
You go to Salinas in Californiaup to Willits in Northern
California.
There's nobody there.
that's America.
You go to India, there's 300pediatric cardiologists.
You go to Africa in Rwanda,there's one guy.
(08:35):
So if you take a step back andyou go, well, oh, okay, we're
going to go build tons ofmedical schools everywhere and
train everybody.
I mean, the quote, Americansystem.
of medicine.
How can this scale?
There's no way it can scale.
You can't in, you know, if I'm afresh graduate from Stanford
Medicine and PediatricCardiology.
(08:56):
You want me to move to themiddle of Montana?
And earn 1.
98.
I mean, why would I do that?
And you just see that We couldbuild I think some people
because AI we try to put it incompetition with are smartest
people.
We, we, that's how we want tothink about it.
It's going to be better than theworld's best cardiologist or
(09:18):
oncologist or whatever.
And I'm kind of going, I don'tthink that's the point of the
whole thing, What if we couldbuild, I call them, red, green,
yellow applications.
Don Finley (09:27):
Okay.
Timothy Chou (09:28):
Red.
This kid needs to be medevacedout of here.
Or this accident victim, right,has a traumatic brain injury,
needs to go to the ER.
Red.
Green.
Hey, everything's cool.
Yellow.
We need somebody else to take alook at this.
If we could do that, we couldput the brains, in essence, of
(09:50):
the world's best oncologist,orthopedic surgeon, etc.
Right into the rural community,into India, into Africa.
Right.
And so it's not, I always say topeople, if your mom works for
Google and you live in PaloAlto, yeah, what we're working
on is kind of cool andinteresting.
(10:12):
But that's not the point of thewhole thing.
I really think that's why, AI inhealthcare, both is necessary,
but also is.
We're capable of doing that.
Don Finley (10:25):
Absolutely.
And so basically, I think yourfirst kind of iteration of this,
or like what you're talking islike in the simplest level of
the triage of basically sayinglike, Hey, what is going on
here?
What do we need to do?
And how do we pair the AI withthe team on site to basically
know, Hey, here's the situationand here's how quickly we need
to act.
Timothy Chou (10:44):
Well, also, you know, if you look at it, I think
ultimately all diseases aregoing to be classified as rare
diseases.
It's just, we don't know enough
Don Finley (10:55):
That's it.
Yeah.
Timothy Chou (10:56):
do that yet.
Right.
Let me just give you a, anexample.
There's a condition called focalcortical dysplasia.
which is actually a brainlesion.
If this lesion is leftuntreated, a kid has epileptic
seizures.
So there's a kid in Floridawho's had epileptic seizures two
to three times a day.
(11:16):
At night, he wakes up screamingfor 12 years.
I mean, think about that.
Don Finley (11:23):
Ooh.
Okay.
Timothy Chou (11:24):
MRI imaged him early on, didn't see anything,
started taking him down atreatment of drugs, some they
could afford, some they couldn'tafford.
Ultimately, they were startingto talk about, let's go put in
electrodes into his brain, Theyfinally re MRI imaged him.
They now believe that he hasthis condition and he's I
(11:46):
actually have photos or images,MRI images that I could show
you.
I mean, you and I would look atit and go, What's the difference
between A and B?
No, I can't see it.
Don Finley (11:55):
Yeah.
Timothy Chou (11:57):
But, you know, if you can find it, you can
actually surgically remove it.
And if you surgically remove it,the kid's cured for life.
Amazing, right?
Okay, there's a good news andbad news story here, The good
news is there's only 25, 000cases of this a year in the U.
S.
Good news, right?
Don Finley (12:18):
Okay.
Timothy Chou (12:19):
Bad news, no one pediatric neuro oncologist ever
sees enough of this to be ableto recognize it.
Don Finley (12:27):
Exactly.
Yeah.
Timothy Chou (12:28):
now, on the other hand, we all know over in the
world of AI that, damn, if Icould get all the MRI images of
all the kids in the world, Icould build a focal cortical
dysplasia diagnostic all day andall night.
and you can repeat, I think youcould repeat thinking about the
world this way.
Over and over and over again.
(12:49):
It's not, we will end up, like Isaid, I think all diseases will
end up being rare diseases.
It's just we don't know enoughyet.
Don Finley (12:59):
That is probably like, it's a very profound
statement to say that like alldiseases are rare diseases.
And I think that is the paththat we're going down.
which gets me thinking aboutlike, what does that world look
like after that fact, how do yousee your work impacting the
healthcare space?
what is the future state thatyou want to have, as far as the
(13:20):
experience of people goingthrough this?
Timothy Chou (13:23):
Well I think, you know, there's better people to
comment on the healthcaresystem, which I'll comment is
totally broken.
but I think We all have to getto, and this technology is part
of it, is how do we investearlier?
Earlier meaning, you know, weall invest in the last 10 years
(13:46):
of your life, we invest infiniteamounts of money, which are in
essence the end of productivityfor that individual.
We do not invest early, We'remaking a difference in a 10 year
old kid's life.
He has 80 years, 90 years ofproductivity in front of him.
We say kids are our future.
(14:08):
We don't spend that way at all,You want to go do a startup in
pediatric health care?
Are you crazy?
Why would you ever do that?
There's not enough market.
I mean, we say kids are ourfuture.
Don Finley (14:23):
Yes.
Timothy Chou (14:24):
So, we need to invest earlier from the
standpoint of, we'll call it thepopulation.
But also, why are we notinvesting?
Why do we wait until just usingthis kid?
We're waiting for 12 years tofigure out what's really going
on here.
That's how the whole system isbuilt.
We're built for the acute issuethat happens, which either
(14:47):
you're going to go take a drugfor it, or get surgery for it,
That's how the whole thing isbuilt.
Why are we not detecting issuesway earlier in the cycle?
The technology is capable ofdoing that and beginning to
alter what is the other half ofthis, which is lifestyle there.
There's a condition in kids.
This is a kind of a, other typeof story.
(15:08):
kids that have cancer, goodnews.
They are given, you know,chemotherapy, et cetera.
Their cancer is cured
Don Finley (15:15):
Okay.
Timothy Chou (15:16):
news.
There is a tendency that whatends up happening is it affects
their heart.
And so while they survive thecancer, they die of heart
disease.
Now, if you can detect thisearlier, the beginning of this,
there's all sorts of, I mean,back to diet, exercise, all
(15:36):
these sorts of things that willprevent this from happening if
you knew earlier in the cycleuntil, right?
So, I'm just saying, I think theearlier we can know, the earlier
we work on these problems, thelower the, fundamentally, the
lower the cost of the system.
Fundamentally, the better theoutcomes in the system.
(15:59):
And if we can use technology todo this, where, you know, I was
at JPM yesterday.
I was like, you know, thebeauty, a lot of people work on
medical devices and pharma, Isaid, the beauty of software is
in essence, the cost to build itis zero and the cost to deliver
it is zero.
I said, the impact the softwarecan have is huge.
(16:21):
Because I don't have the cost ofdistribution.
I mean, or my cost ofdistribution is very low.
Right.
And I think that's wherehopefully, you know, maybe not
in my lifetime, but hopefully weall get here where we can do
that sort of thing.
Focus earlier in the cycle.
Don Finley (16:43):
There's so many things you touched on.
I know we didn't touch on thisearlier, but like, I'm a big
mental health advocate.
And so one of the things thatI'm supporting is like the MAPS
institution, who is like,They're running the trials for
MDMA to treat PTSD and severedepression.
Timothy Chou (16:58):
huh.
Don Finley (16:59):
And so one of the challenges that they've come to
in their phase three trials,which they just in the last few
months got rejected by the FDA,two problems with MDMA.
One is you can definitely tellwhen you've had it.
So it's really hard to do adouble blind, study with the
patient.
The second thing is thattreatment is necessary to go
(17:20):
along with the substance.
And so, the FDA doesn't approvetreatments, they approve drugs.
And they can't approve a drugthat requires a treatment that
goes along with it.
The efficacy of it has to standalone.
And so our entire system is,MAPS is showing basically, after
I think like five years, they'reshowing an 85 percent
effectiveness of the treatmentsof like a six month treatment
(17:44):
protocol.
And so like they're showing Lessthan 20 percent are going back
to depression or like havingsigns of PTSD.
And yet at the same time, oursystem that is in place is
really supportive of, is it adrug or is it a medical device?
And if it's a medical device, isit similar to something that is,
(18:05):
you know, You know, old or ifit's completely new, that's an
entirely different process.
But like we, we have patternsand systems that are set to
basically treat symptoms insteadof getting to that early upfront
kind of approach.
And I know that we talked aboutlike triage of somebody who is,
you know, has a trauma headwound, you know, they need to be
medevacked out.
(18:26):
How do you see the pediatricmoonshot playing in that space
of prevention or early kind oflike lifestyle diagnostics?
Timothy Chou (18:36):
if people are interested, we're actually
tracking 140 different, deeplearning imaging applications in
adult and pediatric medicine.
You will not be surprised, thisis at appcommons.
bevelcloud.
io and I know you can edit intoshow notes.
Don Finley (18:55):
Yeah.
Timothy Chou (18:56):
You will not be surprised.
They're just a handful.
of work in pediatrics, And, letme just, you know, if the
listener doesn't really realizethis, pediatric cardiology and
adult cardiology basically don'thave much similarity.
the phrase that's used in, inthe world of pediatrics is,
children are not little adults.
(19:19):
an And this is the rationale isthat the things that cause
mortality and cause issues inkids, which are unresolved, they
die.
So the population of us oldpeople who are smoking and
drinking too much or whatever,the pathologies that we create
are very different than whathappens in children.
Don Finley (19:41):
Ah, true, true.
Timothy Chou (19:42):
Right.
So number one, I think it's justlet's go pay.
In fact, I just spent time withthe head of neonatology at
Stanford.
There's a condition in apremature baby where this is
interesting.
A chemical gets activated in ababy, normal babies gets
activated, which in essencecauses the lung to inflate and
(20:05):
he used a great example.
He says, when you blow up aballoon, The very first puff,
you really have to force it toopen up.
That's exact because the baby'sbeen right underwater.
That exact mechanism has tohappen when a baby's born.
a premature baby, it doesn'thappen in a lot of cases.
(20:27):
But at this point, they actuallyknow this and they have an
artificial synthetic whateverdrug that causes this to happen,
Don Finley (20:36):
Ah, okay.
Timothy Chou (20:37):
So, I mean, just, that's just another example of
that never happens in adults.
Don Finley (20:42):
I don't know, Timmy.
You've never woken up and beenlike, oh, I gotta breathe
Timothy Chou (20:45):
yeah, I gotta breathe now.
Don Finley (20:47):
Gotta breathe now.
Timothy Chou (20:49):
Yeah,
Don Finley (20:49):
Gotta start my day breathing.
Timothy Chou (20:51):
yeah.
And I think the challenge inpediatrics is, I mean, good
news.
Most kids are healthy.
I mean, that's the good news.
So all of us, our perception is,Kids are healthy because, But
the reality is there's a lot of,I'll touch on another one.
I've spent time with the head ofrare diseases at, UPMC.
(21:13):
and in that usage of the word,rare disease is super rare.
And so he's been doing a lot ofwork in genetics.
And so he told me, he said, youknow, a couple of years ago, he
gets called into the adultdepartment.
There are three patients who arein intensive care.
They cannot figure out what'sgoing on.
(21:35):
He goes in and genetic teststhem and comes back with
actually four diagnostics, whichwere pediatric rare diseases.
Which had been undiscovered inthe adult world.
Don Finley (21:50):
Oh.
Timothy Chou (21:52):
So the, you know, we, the greater we, right.
We say, you know, children areour future.
I mean, in a hundred ways,that's true.
And so, I always say to people,if you're interested in what
we're working on, we have tonsof ways to get you involved.
But the more important thing islet's put some attention on
(22:13):
this.
as a society.
And by the way, not onlyhealthcare, but education would
fit into this category as well.
You know, that's the seed corn.
You know, one of the reasons Ilike teaching is you know, you
if you can, I like to say I havethe opportunity to shape young
(22:35):
minds.
it's a responsibility and ablessing at the same time,
right?
Don Finley (22:40):
I, look at it as kind of like, you know, you're,
you're tending a garden, right?
Like you want the flower tobloom in its own way, but at the
same time, you're there tosupport that growth.
Timothy Chou (22:50):
Yeah.
And we're starting to work inover in oncology and adult
oncology.
What we're talking abouttranscends, I mean, all the work
we're doing.
I always say to people, ourcomputers actually have no idea
if they're in a children'shospital or a clinic or an
adult, they don't know really.
Um,
Don Finley (23:06):
yeah.
No!
Timothy Chou (23:10):
cancer, one of the things that happens in, in more
complex cancer cases is thatthey convene what is called a
tumor board.
So because cancer has thismultifaceted side to it, like,
well, I need a pathologist thatlooks at the slides.
I need a radiologist that looksat, you know, the, the lung
(23:31):
images.
I need a radiation oncologistwho knows how to irradiate that.
I need a medical oncologist whoknows all the, And the
incredible number of drugs thatare now being developed, which
are, basically call it genespecific.
Okay.
At the big fancy institutionslike Stanford over here, or you
(23:52):
go to Dana Farber, they have allthese experts and they
literally, I mean, this is kindof fascinating.
They literally glow into a room.
It's like a meeting room withPowerPoint.
And in a one hour period, or oneor two hour period, they will
give every patient five minutes,these complex patients.
Don Finley (24:13):
Okay.
Timothy Chou (24:14):
Prevent the case, et cetera, et cetera.
And then have this conversation.
You look at that.
First of all, that only happensat the fancy places.
You go get cancer in, you know,Western Montana or whatever.
What, what tumor board,
Don Finley (24:31):
Yeah.
Timothy Chou (24:33):
But yeah, and I know you guys have been working
in multi agent, the multi agentworld, if you sit down and think
about it, couldn't you build amulti agent tumor board Right?
You have an expert in the image.
You have an expert in thepathology.
You have an expert in, and atthis stage, we're also spending
(24:56):
time over in the drug trialworld.
You look at it.
I mean, there's so manydifferent drugs coming out.
How does a oncologist, even ageneral oncologist, in the
middle of the country know, it'snot their full time job, the
whole range of what's availableright now?
You
Don Finley (25:15):
And, that's, the thing.
The amount of research that'sgetting released on a daily
basis is more than you can readin a month.
And so, and then to do the jobthat you're doing, there's no
way for anybody in thatprofession to keep up with where
they're at.
Plus, if you have a tumor board,Any collection of humans kind of
has like the dynamics of a humanrelationship.
(25:37):
And so, you've seen the study oryou've heard this, I think it
was a study, about judges as faras if you get, if you have your
sentencing, before lunch,you're, you're done.
Right.
But if you get it, if you'rethere after they eat, you're
going to get a lighter sentence.
Right.
And I think that we'd, we'd seethe same sort of results in any
sort of dynamic when it comes tolike an oncology or a tumor
(26:02):
board.
I love that name, by the way.
So,
Timothy Chou (26:06):
By the way, just to make a comment, tour boards
are volunteer efforts.
The docs do not get compensatedfor this.
I mean, it's back to thisfundamental, we have a broken
system.
I mean, just
Don Finley (26:19):
I mean, that's, I mean, another example of, like,
how broken the system is.
Now, are there other countriesthat you see doing this, like,
better?
That we could, we could gaininspiration from?
Timothy Chou (26:34):
well, just to make a point of it, I think we're all
on the bleeding edge of what AIand healthcare, and when I use
the term, I think a lot ofpeople, the word software and
the world AI areinterchangeable.
Software could do a lot of coolstuff.
You know, I've been in thebusiness for a lot of years.
but if we're talking about deeplearning, neural networks, et
cetera, we're really still atday one, day two of this whole
(26:57):
thing, right?
Of which ChatGPT exposed us allto the potential of it.
and that's, Really text data,imaging data where, you know, I
don't know we're first minute orsomething.
So I think to tell you there iswork.
we're tracking work and spendingtime with folks in the UK that
(27:18):
are working in this area, folksin Germany that are working in
this area.
So there are other places,particularly when you talk about
distributed learning.
And, swarm learning, federatedlearning.
There are other places andparticularly, you know, Europe
because of their, I'll call itmore strict way of viewing data,
(27:39):
They're much more aggressiveabout working on how do you
build out distributedcapabilities rather than
centralizing it, at motherGoogle or whatever.
So yeah, there are, I mean, butwe're all early, very early.
I mean, there's this uniquecombination of domain expertise
(27:59):
in cardiology, orthopedics, orwhatever, and.
An understanding of being ableto build a team.
Like I said, we're tracking 140of these where I've got some guy
who knows what a convolutionalneural network is.
Right.
and putting those two piecestogether is non trivial.
(28:20):
and, but I think, we're in theearly days.
I always tell people, I was kindof there at the, we'll call it
the birth of cloud computing.
really, I mean, even AWS is,what, 15 years old?
Salesforce is 20 years old,rough 20 25 years old, but even
with cloud computing, whetherapplication or infrastructure,
(28:43):
we're still nowhere.
I mean, plenty of people arerunning on prem, traditional
software models, plenty.
I mean, you know, plenty ofcomputers are being sold that
are put in data centers.
I mean, it's not like it's allquote game set match done.
So when you look at where we arewith AI, I'm kind of like.
we're year one of a 15 yearcycle.
(29:06):
I mean, super early.
Don Finley (29:09):
Oh, I absolutely agree with you.
And I think we end up insituations where we go from
distributed computing tocentralized computing.
And like we find like we go backand forth.
and it's kind of like, even inthe software world, we go
waterfall to agile back andforth as well.
The pendulum swings.
We create new things and it'scool.
And I think we're in that, theirphase of artificial intelligence
(29:31):
being centralized.
But like people like you and,the work you're doing to
decentralize this and showcasethat it is possible to build
private systems that canactually help you and support
you.
And I think that that's a reallyAwesome push for what you're
doing.
And like, we look at even on thesocial media side, right?
(29:52):
Like centralized AI and socialmedia has meant that we are now
the product in which ouremotional engagement is what the
AI is gaining or trying to gainfrom us.
But what I like to see is nowlet's have AI that is looking at
like, what is your purpose?
What is your goals?
and then helping you to achievethose.
And I think that, you know, whatyou're working on as well fits
(30:14):
into that paradigm of givingmore control and access to the
people where they need it.
if, you know, myself and theaudience included had a magic
wand to help you get the, to thenext evolution of your pediatric
moonshot, what could we provideto you?
Timothy Chou (30:30):
I think there's multiple levels.
at one level, it's really, andyou're helping do this, is let's
keep telling the story.
I tell people it took 40, 000people to get to the moon.
We're a few shy of that.
So, spreading the word, numberone.
number two, I think for thosepeople who can engage, do
(30:52):
engage.
Meaning, you know, what does itmean to build a congenital heart
disease?
AI algorithm.
I mean, if you're in the spaceof either the world of
cardiology or the world of AI,go start looking at these
problems.
There's tons of, I said, wealready have examples.
(31:13):
I mean, let me say how specificthis can get.
There is a team at Duke workingon AI analysis of weight
bearing, low power CT imaging.
of the ankle and foot.
So normally, right, when youthink about it, you go take an x
ray of your foot, there's noload on it,
Don Finley (31:33):
Okay.
Okay.
Timothy Chou (31:34):
it's, obviously, there's information when you can
get, weight bearing.
But obviously, if you go dothat, you can't be subjecting it
to a ton of Right, radiation.
Hence, low power CT.
So I'm just using that as anexample of how specialized this
(31:54):
can get.
Low power CT, weight bearingankle and foot,
Don Finley (31:58):
Yeah.
Timothy Chou (31:59):
That.
I think, we're in the middle ofbuilding a Distributed AI
laboratory for healthcare andlife sciences.
What do I mean by that?
We are going to have 32 siteswhere all the imaging data is
available in real time, CT, lowpower, CT, ultrasound, x ray.
(32:21):
MRI, et cetera, all offline dataavailable from the Paxes, which
for the listener doesn't know isin essence, the repository that
the radiologist uses to look at,the images and then all the EMR
data available to authorizedapplications.
And we are want to perfect theart and science.
(32:42):
of taking research work, whetherthat's at University of London
or in Germany or inPhiladelphia, this small
research team, and move it from,we like to say, from the
research bench to the bedside byimplementing it in a laboratory
that sees 2, 000 terabytes ofdata a year.
(33:06):
We'll have 3, 000 servers in it.
Right.
If we can figure this out, wenow have a path to take
brilliance in low power CT forankle and foot, and now deploy
it across all low power CTsystems in the world, So we're
in the middle of that.
I've been funding all of ourefforts to date.
(33:27):
This next step is not going tobe able to be done that way.
So we are actively inconversations with a lot of
people who, you know, I will sayare interested in how
distributed AI will work.
And I'll just make a comment.
If everything moves to, well, Ijust pick on mother Google.
(33:49):
If everything moves to motherGoogle, well, they build their
own chips.
They don't buy servers fromanybody.
So there are a largeconstituency that is interested
in how does a distributed worldwork and we're very eager to
have conversations with anybodyin that space whether that's
(34:09):
from the tech front or the othermajor area we're spending time
in is our new friends inbiopharma.
Where,
Don Finley (34:18):
Bye bye.
Timothy Chou (34:19):
yeah, this is why I said we're spending a lot of
time in, you know, clinicaltrial work where today the
ability to do very preciseconnection of the patients to
the trial is done today.
I'll call it haphazard manual.
(34:39):
Do you know the right people isthe general mechanism.
So what we're right in themiddle of building is a
application.
We call it total recall, whichis connected.
Remember, we have a distributedcomputer sitting inside the
firewall so it can talk to theEMR.
So now I can take it on.
(35:00):
I'll pick on oncology.
I can take an oncology patientwho has a 3, 500 page PDF
document as Her electronicmedical record.
I can rag train a local LLM andnow I can ask questions like,
well, you know, what's thedifference between the last two
CT scans or how many courses ofcisplatin has the patient been
(35:25):
given?
What was the reaction?
What we learned, we shared thiswith 12 oncology nurses around
the country.
I told the team, I said, if wehad a buy now button, I'm pretty
sure they would have hit itbecause we didn't realize how
much manual labor they gothrough in essence with a little
search bar, looking at a giantPDF document and it's like
(35:47):
crazy, right?
so we, we actually, the reasonwe called it total recall is.
We demoed it to these oncologynurses, then we turned over the,
the controls to them to let themask whatever the hell they
wanted to ask.
So, at some point in time, oneof them said, well, what is the
prognosis for this patient?
And you're going, this is whereAI appears like magic, right?
(36:10):
So it does all the other stuff,so it must be able to answer
that.
You're like, no, You humanscan't even answer that question.
Why are we asking the computer?
So that's why we said, no, it'scalled total recall.
It has perfect recall.
of what has happened with thepatient in the past, right?
That's what it does.
Perfect recall, right?
Don Finley (36:30):
recall.
That's it.
Timothy Chou (36:31):
that's, what it does.
Don't
Don Finley (36:32):
it's
Timothy Chou (36:33):
do
Don Finley (36:33):
in the name.
It doesn't, it's notpontificating on what is
actually like going on.
It's not connecting the dots,but it's just total recall.
I love that.
Timothy Chou (36:43):
Now remember, we now, the application now has
complete profile information ofthe patient, Cause it, looking
at the MR record.
So what we have come to learn isthat in phase 2, phase 3 drug
trials, the real data about whatthose protocols are, are sitting
inside protocol documents thatcan be 300 pages of PDF and they
(37:10):
are sitting behind the firewallbecause the pharma company
doesn't want to let everybodyknow what they're working on.
Okay, if you have a distributedinfrastructure, I park a server
inside Roche, Bristol Myers,Squibb, whatever, LLM, RAG, the
protocol document, and now multiagent, I can have the total
recall agent talk to theprotocol agent, and now the hit
(37:33):
rate has got to skyrocketbecause you have now complete
knowledge of the patient andcomplete knowledge of the
protocol.
And just to say it, This.
is not that hard to do.
Don Finley (37:47):
No, that's entirely you're, you're talking about
things that we do in otherareas, right?
Like there are
Timothy Chou (37:53):
bad.
Don Finley (37:54):
of this.
And in fact, we have ragsolutions running in our own
systems.
We distributed for our clients,right?
Like all that question andanswer stuff is.
is relatively, I mean, it's nota, a solve solve problem, but
they are solvable.
They require a little, it's likethe early days of Oracle, right?
Like, you know, it was adatabase company, but they were
(38:15):
still like building and like ragsolutions are context based.
For the most part, you got tofigure out how to get the
information in, in a way thatthe machine can process it.
But at the same time, likeyou're talking about stuff that
is, you know, day oneimplementable, not.
Two years, five years down theline.
I love it.
So Timothy, I gotta wrap thisup.
(38:38):
Is there any other like lastbits of wisdom that you would
like to share with the audienceof like any life lessons as far
as how to, how to live in this,you know, future AI world that
we're talking about?
Timothy Chou (38:49):
Well, it's something I, I said to you
during the pre meeting.
When I start out my Stanfordclass, I always say to the
students, I go, you know, thequality of a student is not
measured by what you know, butby the quality of the questions
you ask.
And maybe in a world of LLMs, itis even more true.
Don Finley (39:11):
Yes.
Timothy Chou (39:11):
It's the quality of the questions you ask.
And I think the, the more we'reable to be curious, the more
we're able to ask the rightquestions, the faster this is
all going to happen, right?
And
Don Finley (39:28):
I absolutely love it.
Curiosity is one of the thingsthat drives this world and
Timothy, thank you so much forbeing on the show today.
Like I really appreciate thetime we've had together.
Timothy Chou (39:40):
I have as well, Don, it's been a lot of fun.
Don Finley (39:43):
Absolutely.
Thank you for tuning into TheHuman Code, sponsored by
Findustries, where we harness AIto elevate your business.
By improving operationalefficiency and accelerating
growth, we turn opportunitiesinto reality.
Let Findustries be your guide toAI mastery, making success
inevitable.
(40:04):
Explore how at Findustries.
co.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Amy Robach & T.J. Holmes present: Aubrey O’Day, Covering the Diddy Trial
Introducing… Aubrey O’Day Diddy’s former protege, television personality, platinum selling music artist, Danity Kane alum Aubrey O’Day joins veteran journalists Amy Robach and TJ Holmes to provide a unique perspective on the trial that has captivated the attention of the nation. Join them throughout the trial as they discuss, debate, and dissect every detail, every aspect of the proceedings. Aubrey will offer her opinions and expertise, as only she is qualified to do given her first-hand knowledge. From her days on Making the Band, as she emerged as the breakout star, the truth of the situation would be the opposite of the glitz and glamour. Listen throughout every minute of the trial, for this exclusive coverage. Amy Robach and TJ Holmes present Aubrey O’Day, Covering the Diddy Trial, an iHeartRadio podcast.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.