- il y a 2 semaines
From Patients To Experts, How AI Can Improve The Journey To Better Healthcare
Catégorie
🤖
TechnologieTranscription
00:00Sous-titrage Société Radio-Canada
00:43This session is moderated again by Benoit Georges from Les Echos.
00:48I'll let him take it from here.
00:50Please give a warm welcome to our speakers.
01:07Hi again.
01:09Well, everybody knows how internet has become important for patients.
01:14Sometimes maybe a little too much because we are all hypochondriac sometimes.
01:19And even if we joke about it, we all love to search about everything every time we need some advice
01:26or every time we have something that's not so, so good.
01:30And usually, for everyone of us, it starts with Google.
01:33That's why I'm very happy to have on stage from Google DeepMind, Joel Baral, Research and Engineering Senior Director at
01:42Google DeepMind, which is one of the best experts, one of the best companies to talk about artificial intelligence.
01:48And also, to welcome also Stanislas Newks-Chateau, co-founder and CEO of Dr. Lib, everybody knows.
01:57In France, the search also starts with Dr. Lib.
02:00In Germany also, maybe not for the other countries, but it's coming.
02:04And both of you, you are the way that most many patients in France, more than half of the patients,
02:12use to interact with their doctors and health professionals.
02:18First of all, I would like both of you to give us your own vision of the future of medicine
02:24and the future of interaction between patients and doctors in, say, like, five or ten years.
02:31I'll let you give your vision, Joel.
02:34Thank you, Benoit. Thanks for having me.
02:36Hi, everyone. It's really great to be here today.
02:38Five to ten years is tricky, right?
02:41Like, always looking ahead.
02:43I like to take maybe ten years because that's the duration of med school in most countries.
02:48So when we try to predict, you know, ten years from now, we're trying to predict how will healthcare look
02:54like for students that, you know, are entering their medical studies today, right?
03:01And when they go out, how will the world look like?
03:04So at Google, we profoundly believe that generative AI will have one of its biggest impacts on the world in
03:12healthcare.
03:12And we want to have a bold and responsible approach to do that and also really work with many partners.
03:21So we believe healthcare will change things for the scientists, for the clinicians, for the patients.
03:28I think we're just on a panel with Colin, who is part of our isomorphic labs endeavor.
03:35So that tells you, I think, the scale at which science is going to be profoundly changed.
03:41If you remember AlphaFold, I'm sure you talked about it.
03:45AlphaFold, now we can predict the 3D structure of every single protein.
03:48That's one example in molecular biology, but you can think of the sciences being profoundly changed in the exact same
03:55way, you know, very broadly.
03:57So scientific discovery really is going to change rapidly.
04:01Then for healthcare professionals, generative AI is really a tool that has the potential to augment every single person in
04:13your healthcare journey.
04:14Not only physicians, nurses, you know, anyone, specialists that, you know, you may want to interact with.
04:23And that's, I think, something that humanity, like a tool that humanity now has at its disposal that is completely
04:31revolutionary.
04:32You know, if you think about it, you can have the expertise of all of those professionals or maybe all
04:37of the oncologists on earth having the expertise, you know, one oncologist is having the expertise of all of them
04:43combined.
04:43Because really, generative AI can look across, right, all of the information, all of the knowledge that we've accumulated everywhere
04:52and distill it into what matters to you as a patient, right?
04:57What will really help you get cured?
04:59And then for users, you know, you mentioned it, access to information and high-quality information is very important.
05:07And that's why we start most of our healthcare journeys by trying to find information for ourselves, right?
05:14Like trying to see what might happen to us or understand the information that a healthcare professional is giving us.
05:21And so, again, I think we're seeing that we're entering an agentic era where, as a user, we're helped with
05:27agents that can really address our own questions in a much more personalized way.
05:34It's, of course, really important to have a responsible approach when we are entering, you know, such a future.
05:41And that's something, you know, that is really at the heart of all of the research we're doing.
05:45And then maybe the last thing I would like to bring up is the fact that in healthcare, we never
05:49do that alone.
05:50Like, we really work with partners.
05:51That's really important because healthcare is quite complex as an industry.
05:57Many stakeholders, a lot is at stake.
06:00And we work with fantastic leaders like Stan and explore, you know, potential collaboration with Dr. Lieb.
06:07We work in France with the Curie Institute, which is, you know, world-leading in cancer research.
06:12We bring our very powerful AI tools and see how we can marry them with, you know, world-class research
06:18to really help with women's health in particular.
06:23So, again, you know, very, very exciting to be at the beginning of such a revolution in healthcare, coming with
06:31a responsible approach from the start and working with the best, working with a lot of partners to really bring
06:37that to everyone.
06:39Stanislas, when I was preparing this, I found out that Dr. Lieb was launched 11 years ago.
06:45I couldn't believe.
06:46To me, it's so, in France, it's a ubiquitous company.
06:49I mean, people use it all the time.
06:51I couldn't see my doctor if I didn't use Dr. Lieb today.
06:56But what do you imagine in 10 years?
06:59You changed a lot.
07:00The medical professionals, health professionals are sometimes seen as very conservative people towards technology.
07:06But they embraced yours.
07:08I mean, teleconsultation and all the tools that you provide.
07:13What's next?
07:14What will it be in 5 or 10 years?
07:15I don't know.
07:16I'm, of course, completely aligned with what you said.
07:20Joel, I'm as positive as you on the future of healthcare.
07:24And I think it's important for all of us to realize that thanks to technology, thanks to AI, we will
07:31together reinvent the work of health professionals.
07:35We will reinvent the role of a patient.
07:41Today, patients, they have a problem.
07:44They have an opaque patient journey.
07:46They need to go to see a doctor.
07:47They have an appointment in 15 minutes.
07:49Everything is decided.
07:50Tomorrow, patients will be more experts, more autonomous.
07:55Tomorrow, patients will have access to unlimited medical knowledge, clinical assistance, in order to help you to prevent a health
08:04problem, to detect a health problem.
08:06But also, in order to help them to adhere to a treatment, to guide them.
08:11So, for sure, AI and technology is a revolution on how we transform patients into an expert and an autonomous
08:22player of their own health.
08:25And for health professionals, it's as radical as it is for patients.
08:28As you all know, health professionals, they have deteriorating working conditions over the past decades.
08:37And her job at Doctolib, her job as AI expert technologist is to help health professionals.
08:44To help health professionals to have zero minutes of administrative tasks, of financial tasks, of repetitive tasks.
08:52And yes, AI and technology can transform that, can help health professionals to focus only on their vocation, providing care.
09:03And yes, AI and technology can help, as you said, health professionals to have unlimited access to medical knowledge, more
09:12and better patient data, but also clinical assistance to provide care differently, to provide better care.
09:19So, our job at Doctolib is to reinvent the job of health professionals and the role of patients.
09:27And yes, I believe technology and AI will have a massive impact on the health of patients at the end.
09:34And we should all work to make that happen together, wherever we come from.
09:41Doctors, scientists, technologists, of course.
09:44We talk about AI agent or assistance for patients.
09:47You already provide your customers, your doctors, an assistant to synthesize and based on AI to do a synthesis of
09:58what's the consultation.
10:00Can you explain more about this?
10:01And is it accepted well by the doctors?
10:03Yeah, but we have a couple clinical assistants, financial assistants, administrative assistants already used by health professionals.
10:10Just to give you an idea, we have launched a consultation assistant.
10:13And I suppose that's what you prefer to, but we launched a consultation assistant six months ago, some other.
10:20We moved from burning constraints into free and augment health professionals.
10:25And for patients, it's the same.
10:27For patients, it's the same.
10:29Unlimited access to medical knowledge thanks to AI and technology.
10:33But there's a great responsibility also by talking about us, all of us, looking on the Internet about the thing.
10:41We used to, today we have 10 links or links that are from very clear sources.
10:48And Google did a big work on finding the best sources.
10:52But the world of search is changing a lot these days.
10:56Now, you don't only have links, you have AI-generated answers that are more complex, probably more detailed.
11:04How do you ensure there's a big responsibility on providing them to patients?
11:07How do you ensure that it's not going to hallucinate or it's not going to misunderstand what the patient is
11:14asking?
11:15Yeah, I think that has always been very important for us, right?
11:19You mentioned it earlier, some people with symptoms will search for information.
11:27And we have a big responsibility to provide them with high-quality information relevant for the country they're in.
11:34So we've worked a lot with the different institutes in different countries to make sure we provide authoritative content, right?
11:44And if you're in France or if you're in the U.S., you'll get a different answer because the way,
11:49you know, your symptoms might be treated might be different in those different countries.
11:53So, you know, you have the cards that explain exactly, et cetera.
11:56So that's something we've done over the years, right?
11:58And we have, I think, always been, you know, extremely thoughtful in the way we are bringing health-related information
12:06to users.
12:07That's true for search, but that's true for YouTube.
12:09If you think about it, in many countries, people turn to YouTube first, for example, in India, to search for
12:15high-quality answers.
12:16The other thing that's really important is very often people turn to search not because they have a health problem,
12:22but because they have a human problem.
12:24They want, you know, they will need to cope with a disease.
12:27There might be stigma associated with it.
12:29They want to find people like them that, you know, have done that before.
12:32And, again, it's very important that we, you know, we bring up and amplify the voices of people that are
12:37best capable of helping them.
12:39And so, you know, we have that long history.
12:41And now comes Gen AI.
12:43So one thing that's really important is that, you know, we're not here to provide a diagnosis, right?
12:48So if you're asking, you know, which medication should I take, we'll not tell you what you should take.
12:54We'll tell you, well, you should go to your doctor, right, if that's the right thing to do in that
12:58location.
12:59And making that very smooth, actually, is very important.
13:02We work together, right, to make sure that in France, if you're looking for a doctor, you will very quickly
13:07be able to, you know, get to your doctor lead page.
13:11And maybe you can just explain that a bit more.
13:13But I think that's one of the ways, you know, again, to make sure that people get to care and
13:18have the best information to make the right decision.
13:21What's very interesting with generative AI is what you're describing, which, you know, we're experimenting with AI overview, something we
13:29call AI mode.
13:30And those things allow us to do the searching for you, if you wish, right?
13:35Rather than you going and then looking at various links, et cetera, we can summarize that information with generative AI,
13:41still giving you the sources.
13:43So allowing you to click.
13:45And for health care, what we see is people actually really want to know where the information is coming from,
13:50right?
13:50So it's really important that we are giving them the sources and we're explaining why this is the way we've
13:55summarized the information for you.
13:57And it's also very interesting if you take a step back from a generative AI perspective more broadly, you know,
14:04outside of health care.
14:05What we're seeing with all of our large-ronged models is that the speed at which you're giving an answer
14:10is not always the only thing.
14:12For things like health care, you know, people are okay to wait a bit more, right?
14:17And have the models do some thinking before they give you their best answer rather than just their fastest answer.
14:24And so it's also a very interesting application of generative AI in that sense.
14:28The ability of our models to think before, you know, before they get back to you, if you wish, is
14:34really, really helpful for health.
14:36You have a very interesting project also, a research project, I think, on diagnosis and reasoning and conversations between patients
14:45and doctors.
14:46It's called AMI, Articulate Medical Intelligence Explorer, if I get it right.
14:51Can you tell us more about it and how you can ameliorate the conversation between doctors and patients with that
14:57tool?
14:57Yeah, so I'm part of our foundational research teams where we really explore the cutting edge of research with clinicians,
15:05with foundational researchers, engineers, but really looking at the art of the possible, right?
15:10So this is really, you know, disconnected from any product or any product surfaces.
15:16And with AMI, as you said, the Articulate Medical Intelligence Explorer, we are looking at whether large language models are
15:24capable of having a dialogue as would a physician with a patient.
15:28And we've already published two papers in Nature, and we're continuing to have a series of papers to go from
15:35something that is purely capable of dialoguing with text to adding eyes and ears, having a model that can take
15:44medical documents, et cetera, to, you know, little by little, getting closer to a real physician-patient interaction.
15:51And what's really interesting is when we evaluate very carefully those models, first, with patient actors, as medical students would,
16:00right?
16:01They are always evaluated in those scenarios where patient actors interact with them.
16:05So we evaluate our models the same way.
16:07We see that they are very performant on what you would think are pure cognitive tasks, right?
16:13Recalling knowledge, being precise in the information, et cetera.
16:16But they're also quite performant in understanding the patient, if you wish, right?
16:21Showing empathy.
16:23They're also infinitely patient.
16:25So if you don't understand what they told you, you know, they can repeat, they can rephrase for a five
16:29-year-old differently than from, you know, a 20-year-old, et cetera.
16:32There's an aspect of personalization that the models are actually really capable.
16:36Yeah, interesting.
16:37And talking about the future of health in every country, even develop or not,
16:43one thing is there's a shortage of physicians, doctors, specialists, that's true in France, that's true in all around the
16:52world.
16:53Can tech help that in tools like Dr. Libs or in artificial intelligence answers, like you said?
17:00You are working on that?
17:02Yes, for sure, technology and AI are solutions to medical disease, solutions to access to care.
17:09Because as I said at the beginning, there are missing 10 million health professionals in the world.
17:1410 million.
17:15In our countries, it's hard to access to care.
17:17But in developing countries, it's even harder.
17:20And so why technology and AI can change the game?
17:24One, because with Health Companion, we can turn patients to be more experts and autonomous, so we can help them
17:31to prevent a disease, detect a disease, guide them.
17:36Second, because by powering health professionals, we are freeing them.
17:40And so concretely, they have 20% more time, 30% more time, and so they have more time to
17:45take care of patients.
17:46But also, they have more time to take care differently.
17:51The encounter visit, the patient visit, if you want, it's not the future, it's part of the future of health
17:59care.
18:00And of course, because of all these new technologies of communication, we can also give access to patients to more
18:09fluidity in understanding their care journey,
18:12in having access to the supply of care resources.
18:16So for sure, technology is a key role to play to improve access to care.
18:21And we have proven that.
18:22In France, we reduce by 10 days the time to access to care.
18:2710 days.
18:28After 11 years, and today, we have more than 50 million French people using Doctolib on a daily, weekly basis.
18:37So we have proven that by powering health professionals, by helping patients, we can find a way to have a
18:45better access to care.
18:46But also, prevention, detection, adherence to treatment.
18:50Of course, technology is not the only way to improve access to care.
18:53So we need also to train more health professionals, and we need probably to reinvent a bit their scope of
19:00work.
19:01And AI will also help health professionals to have a different work tomorrow.
19:06Because with more medical knowledge, with more medical assistance, we could imagine different types of health professionals treating different types
19:15of patients.
19:17And do you have also experiments or works on regions that are underserved by medical professionals?
19:25Yeah, absolutely.
19:25We've done a lot of work globally to help bring better solutions where there's a lack of physician.
19:31For example, with diabetic retinopathy, which is the leading cause of preventable blindness worldwide,
19:37often people get blind just because they didn't get screened first and treated second.
19:43And so with a project named Arda, especially in India and Thailand, we've deployed AI technology to 700,000 people.
19:50We're going to expand that to 6 million over the next 10 years to really help all of those people
19:55get screened.
19:56And I think what Stan also alluded to, which I think is really powerful with generative AI,
20:02is the fact that we have now foundational models that can be open-weight, that can be distributed in the
20:08community,
20:08precisely to help in those places, those countries, all of these use cases that are quite specific to a region
20:14and a disease, for example.
20:16So we recently released MedGemma, and we are really thrilled to see the community build enormously with MedGemma already,
20:23where there might be developers in the room, we are going to have a hackathon here in Paris early July,
20:29and really we feel that this, you know, infusing the technology broadly is how we are really going to be
20:35able to disrupt healthcare at a global scale.
20:37Well, thank you so much. I wish we had more time to talk about this.
20:40Unfortunately, we have to leave the stage. That was a fascinating discussion,
20:43and thank you for your patience and the way you explained this so clearly to all people.
20:49Thank you so much.
20:49Thank you.
20:54Thank you.
Commentaires