AI in Healthcare
This week I am talking to Rob Brisk, MBChb PhD Chief Scientific Officer for Eolas Medical (@EolasMedical). Rob has a fascinating background with experience in both healthcare and machine learning and artificial intelligence. Robe shares his journey from being a physician to venturing into the world of AI and emphasizes the importance of clinicians’ involvement in AI developments, highlighting the need for a cautious approach.
We discuss the immense potential of large language models like GPT-4 in healthcare, focusing on their ability to reason and process vast amounts of data. But all this is tempered with the hype surrounding AI and the importance of building trust and educating both clinicians and the general public.
Rob shares soem details of how AI can be used as a tool to connect clinicians with evidence-based guidelines, promoting a safer and more gradual integration of AI into healthcare. Listen in to hear the exciting possibilities and challenges in the intersection of AI and healthcare, and the importance for collaboration, trust-building, and responsible implementation.
Listen live at 4:00 AM, 12:00 Noon, or 8:00 PM ET, Monday through Friday for the next week at HealthcareNOW Radio. After that, you can listen on demand (See podcast information below.) Join the conversation on Twitter at #TheIncrementalist.
Listen along on HealthcareNowRadio or on SoundCloud
Raw Transcript
Nick van Terheyden
And today I’m delighted to be joined by Dr. Robert brisk. He’s the chief scientific officer at Ole us, Rob, thanks for joining me today.
Rob Brisk
Thanks for having me, Nick. Pleasure to be here.
Nick van Terheyden
So, obviously, we just have to clarify for users, listeners that are listening in, you know, they’ll have to separate out those brilliant dulcet tones of the British accent. Normally, it’s very easy to tell who’s talking. But in this instance, we share a little bit of common heritage. I hail from the same shores as you do. As I do with all my guests, I like to start out with a little bit of history, and how you got to this position. You’re one of those unique individuals, you’re the doctor, doctor of the world. Tell us about your pathway to this point. Today, if you would,
Rob Brisk
yeah, sure. So, I mean, as you know, Nick, something else we share is how we started our careers, I was also a, an out and out medic, when I started my professional life and pursued a pretty conventional NHS career for the first number of years, meaning I went through kind of general training and especially training and then got, what would in the USB, I think residency program place. And at that time of my life, I’d always been interested in computers, always kind of tinkering in the background had never anticipated computers would perform any sort of would fill any meaningful role in my, my professional work. And then, for a variety of reasons that we might or might not go into, in 2016, I realized, actually, I love clinical medicine, but it’s not all I want to do. I’m sort of at that point where I’d kind of done all the hard bits in all these competitive programs that you had to get on to and all these exams, you had to sit and then I was staring down the barrel that probably the next 3035 years of my career and thinking not only this, you know, I want to do this, but I don’t only want to do this. And so I began kind of picking up my interest in computers again, and was just at the right time to catch the kind of the first wave of their AI Wait, you know, 2016 was the AlphaGo year for listeners who are familiar with the world of AI. So that was the big landmark event. And I got wind of that and thought this is really, really interesting, like, like nothing I’ve ever come across. And the further I went into that world, the more I thought, this is the technology of our age, this is, you know, we’re going to talk about hype, I’m sure over the course of discussion now. But I beside I thought this is this is world changing at some level or another and particularly if that world is clinical medicine. So I, the rest is sort of history, I would have been drawn further and further into that world, I was able to go and do a PhD in computer science to really formalize that part of my of my professional life. I went and did a couple of years in Vidya, of whom most people will now have heard that at the time, people knew them as a gaming chip company. But of course, they’ve radically transformed themselves. And then now I’m, as you say, I’m homeless, where I’m working, I spent four days a week, working largely with people attending to court, large language models, but in a clinical context. And then one day a week, I’m still in the hospital practicing in general adult cardiology. And I think I’m really, really enjoying the confluence of those two worlds and being at that interface. And being able to kind of anticipate all these exciting things are going to happen, I hope within our lifetimes.
Nick van Terheyden
You know, I’m almost embarrassed to tell a little bit of my you know, counter story to that, because you’re right, very similar started out in clinical medicine. But in my case, it was a lot earlier than you. And I had to be excited by a 16k portable device called a sigh on organizer that just blew my mind. I was wow, this is amazing. Just think what this could do. And here you are talking about AlphaGo as your sort of inflection point, and I’m thinking, wow, if I’d seen that, but actually what I would have thought that and I’ve been on this journey, and it’s it just continues to expand. I mean, it’s truly exciting. I love it. Very similar. The only difference for me was the capacity to combine those two worlds just never really materialized. Certainly in the days it was very hard to find that, you know, and mostly in the clinical world. People looked at me oddly and said, Why are you doing this? And in the business world, they said, No, you focus on that. So I didn’t have that luxury unfortunately, but you know, I love Hearing this as you know, more and more people, because ultimately, we need physicians involved, because I think it is. It’s abundantly clear to me, tell me if you disagree, but I think this technology, and I’ll use the broadest term AI to sort of encompass all of the elements, and that’s large language models, deep learning all of the contributions is going to revolutionize our space. I think it will revolutionize lots of others, but certainly in healthcare, and you know, you’re at the nexus of this. I mean, it’s just, I’m so excited to hear a little bit about the in video experience, because that was still early days, they were, you know, still really graphics orientated. How did you end up there? And what what did you end up doing?
Rob Brisk
Yeah, no, great, great question. And just before I talk about a video, to your point about when what drew you into this world, I think the earlier you got excited about this world and realized it was going somewhere special that like, the more credit you can take, I think if it was AlphaGo, that turns you on to that world, it was more or less, you know, the writing was very, you’re
Nick van Terheyden
being kind, but thank you.
Rob Brisk
Yeah, so in video? Yeah, I probably was. I mean, I mean, you with Nvidia, the nice thing is, you can use their stock price to really track their journey and their transition. And I wasn’t right at the very beginning of that move, which, you know, to my financial detriment to an extent, guys who have joined even just 18 months before me, who they really saw that transition happened where Nvidia did a huge pivot, the point where I joined was the point where video had very clearly realized, as a company that this was, you know, their business model and the future of the entire digital economy. And all in all likelihood, I mean, probably just acquiring as a CEO, and a bit of a visionary realized it well before that, but it had been absorbed into the company ethos, and they were beginning to diversify and say, Okay, we need some expertise in various different fields. So, you know, we need people in FSI. So the financial services industry, we need people in autonomous vehicles. Health care was one of those domains that have been identified as one of the pillars of the future of AI. And, and so if any, were looking for a few people to who had crossover expertise, who could talk about, you know, the stuff that the silicon did, who could talk about the stuff in the software that ran on the silicon, but who also understood the impact of the of the applications that were being built on this stack, they use a technical term. So you know, really, from an oversight of the world, from patient to processor, and that was kind of the way we started to talk about it a little bit. So they actually brought in a number of people who had expertise in drug discovery and medical imaging, I was fortunate enough to be one of two physicians in the company. So there’s Mona Flores, who, who’s still here in the States, who, who’s a good friend of mine, she was a cardiothoracic surgeon before he joined a video. And then I was the sort of the physician covering the Emir region. My job was a real mix of things. And Vidya describes themselves as the world’s biggest startup. And when you’re inside the company, it does feel like that fantastic place to work always, always really highly rated on Glassdoor, and rightly so. But also very busy. And, and at times, you’re never quite sure what the next few weeks are going to hold, which, you know, if you’d like that, it’s great. If you don’t, it’s a little bit terrifying. I ended up sort of setting up and establishing a number of collaborations my job, and I ended up being sort of the global lead for partnerships in the life sciences space. So I would get to work with people doing really exciting stuff from companies like AstraZeneca, and GSK, and some very big and well known academic institutions, and a couple of big clinical organizations. And my job would be to work out what what are all these people doing? What are the key trends that are emerging from all this work? And how do we position ourselves so that we’re involved in projects that are ahead of they’re ahead of the curve, so that by the time the world at large, starts adopting these applications at scale, we’ve already been collaborating with people and be bringing in a deep understanding, which was an enormous privilege. If you’re interested in this world, and it’s probably clear already that I am then it’s sort of license to go and sniff around for what’s interesting and get yourself gender on it. So a brilliant time.
Nick van Terheyden
What a fantastic opportunity. I mean, I think, you know, especially with that combination, and, you know, at least one other individual who has a similar background, that sort of, I think helpful in terms of settling into an environment I certainly found myself alone for so I can only imagine and, you know, from a company standpoint, I you know, I’ve been fortunate to land at a company that You know, very similar sort of experience, one of the best nicest cultures, you know, most dynamic that I’ve had the the sort of privilege of working with, which has truly been, you know, a real positive experience, I can really, you know, associate with that, as you think about where we are now. So, you know, let’s get into it, and talk about some of the hype, and you know, where the technology is, I think, chat GPT hit the waves. I’ve got to say, everybody that’s in AI goes, Yeah, well, of course, and stuff works. I mean, you know, what are we talking about? This is not as new as everyone thinks it is. But it was the same Siri moment that we saw with speech recognition. Everybody goes, Wow, this speech stuff really works. And that, for me was a positive because everybody’s saying, Well, this is good, you know, more accepting, and then, of course, comes all the, hey, we do this, you know, there’s an awful lot of marketing hype. I think there’s great potential, you’re obviously knee deep in these large language models. They’re, they’re almost magical. I mean, I feel like it’s a sort of Harry Potter kind of moment with some of the capabilities. And as you read the digestive folks that really deeply understand this, even for some of them, they say, I’m not sure how it’s doing that. But it’s achieving some really impressive. So. So let’s start with the positive and say, it’s clearly doing some amazing things. What are your thoughts about that technology? Certainly your space, your focus that’s really going to bring about, let’s say, some short term and medium term opportunities in healthcare?
Rob Brisk
Yeah, great question. So I think I think the first thing to say is is chat TPT. And particularly, you know, the underlying technology, which in the proprietary case of open AI is GPT model series, these generative pre trained transformers. But more broadly, what the technical term you do for these things is very large scale transformer based models, you know, sequence to sequence models, people, you’d say, call them LLM xlarge language models, that’s actually a bit of a misnomer, because they are by no means limited to language. People also call it generative AI, which is another misnomer. But we understand what we’re talking about when we’re talking about these things. The reason they’re so exciting is that the concepts that have been proven out by check GPT, particularly if you have access to the GPT, four powered model. So GPT four is by far and away the biggest, most powerful AI model that’s publicly available today, it’s a multi trillion parameter model. That’s huge. If you want to draw some, like really fast and really loose analogies. In terms of the number of synaptic connections, you know, I’m doing air quotes now, between these artificial neurons, you know, you’re talking about the size of a small mammalian Oh, behave. Yeah. But, but what what is clearly proven is that, you know, through relatively simple paradigm, you know, in the case, when you are talking about language based applications is next word prediction, and some some sort of naysayers about this, are fond of dismissing these models as simply next word predictors. To which I would say, well, we as humans, we’re simply next motor neurone predictors, that the mechanism of interaction is less important the fact that by training models like this, you can get them to learn these incredibly sophisticated, complex and abstract representations of the world around us that they can reason over. And we’re not sure where that journey ends, because to date, we haven’t seen the ceiling of the performance scanning of these things, you know, how far above multi-trillion parameter models can you go? And we know that they can bring a different sensory modalities, you know, again, I’m using the term sensor in a very loose sense, but they can make sense of language. And you’ve got GPT for vision, which has just been released under early access. So these things can reason in a way that’s very, very human like, and I think, you know, to stray, perhaps into the world of hype will be superhuman, in some respects, already is superhuman in some respects. So it’s not the chat DPT itself technically it is an application, right technology that under powers, it has such huge potentials across so many different domains and so many different applications that I think that’s what we’re all excited about. And you know, people who don’t understand this world, they might mom, for example. She’s not a technophobe, but she’s definitely not in any way into this amazing show a chat GPPs she’s impressed because it is like magic. But, you know, I’m telling her, this stuff’s gonna change the world. It’s gonna, it’s gonna change, you know, how will my kids experience the world around them? And I think a lot of people are struggling to make that leap. But it’s, you know, what we can do with the same technology that powers that and of course, what I’m thinking excited about is what we can do in healthcare. All right.
Nick van Terheyden
So for those of you just joining, I’m Dr. Nick the incrementalist today. I’m talking to Dr. Rob brisk is the Chief Scientific Officer at Alas, we were talking about large language models. And again, forgive me I do simplify deliberately, because this is a, you know, brief 30 minute show, getting into the details, you know, and I’ve listened to some of the AI podcasts, what’s interesting, you could track the same as that stock price, the number of AI podcasts is gone. And I do not have the listening time for all of them. But they’re all very long, I mean, the hour plus shows because the detail that’s necessary to really, actually get to it. So, you know, for anybody that’s listening and says, Well, no, that’s not exactly right. We are sort of oversimplifying, certainly me because my understanding is not as deep as Rob’s but that’s part of the reason for having you on the show. We were talking about LR lands, and you know, the Generation One of the things that you said, and I want to sort of focus a little bit on this, because, you know, there was a lot of talk from three point, I think it was five, and then, you know, when four was released, there was what’s the size of the model, and so forth. And in fact, to date, I haven’t seen anything published to say how big four is, number one, but as you looked from, you know, two to three, there was this sort of big increase. It’s not as simple as that. I mean, yeah, sure, you can add all this, but that actually doesn’t create necessarily better models is my understanding. Is that a fair assessment?
Rob Brisk
Yeah, it, that’s certainly not all there is to it, or though the scale that so as the models get bigger, their potential increases? On a, almost as far as we’ve seen in the public sector on an almost linear scale. Now, this is where some technical people in the audience are going to be going. That’s not right. Yeah, broadly, broadly. That’s what we’ve seen. And to your point about No, I haven’t seen it published either the model size, but I was at a panel discussion that Peter Lee was sitting on earlier this year, the VP from Microsoft Research for any listeners who don’t know, and he let slip, he let slip, he used the word, this is a multi trillion parameter model we’re talking about. So no, there’s there’s a lot more to it than that. And these things are fantastically difficult to train, you need huge amounts of data, you need to be very careful. I mean, in fairness to open AI, you know, and open AI has a controversial organization. And what I will say about them is they are anything but open. Despite
Nick van Terheyden
Yes, very good boy.
Rob Brisk
Yeah, that’s, you know, they were conceived of as a as an organization, democratize AI, and then they’ve built this huge model, and it’s proprietary, and it’s very close. But that, you know, they’ve done an incredible job of training this thing. But as, as you know, Nick, and I think as, as everyone knows, at this point, you know, there, there’s a lot more to work through. And one of the big issues that gets talked about in this space is hallucinations, which I think is an awful term. But agree, it’s commonly understood what we mean by these things make stuff up, they confabulate because they’re not, you know, they give the appearance of human like intelligence, but they, they’re not human, they’re AI is trained for a certain job that they try their best to do. And sometimes they think what you want to hear is an answer that they’ve plucked out of thin air. And therein lies the major, or one of the major pitfalls, these things in healthcare. And that’s a challenge, I think, as a community we’re very much wrestling with at the moment. And you made the point earlier, that there’s there’s power in the combination of, you know, having background and a couple of different domains. And and I think, to your point about the importance of physicians getting into this space, this is one of the key reasons we need physicians to be, you know, conversant in the language of modern AI. Because only only frontline healthcare professionals can can just instinctively see, okay, well, we could use the AI in this place, and we can exercise some oversight. Whereas if we use an AI in this situation, that’s going to be incredibly dangerous. You know, the technologists that tend to be amazingly bright, I’ve never felt so stupid as I quit working at Nvidia because the people I was working with just could, you know, think circles around me a lot of the time, but they haven’t done they’re there, you know, 10s 10s and 10s, of 1000s of hours on the clinical coalface and you just have to do that, to understand the world of clinical medicine.
Nick van Terheyden
So, limited amount of time I think it’s important to get to some of the hype and you know, the places for people to focus or think about because, you know, we’re both excited but I also want to, you know, give some caution on this whole concept. It’s not the be all and end or tell us your top areas of ionic Call it over over excitement height that, you know, inappropriate that you should look out for specifically around health care where I think we we need to exercise caution and clinical oversight.
Rob Brisk
Yeah, great question. I mean, I mean, the answer is probably everywhere. And we need to exercise caution and clinical oversight.
Nick van Terheyden
I, you got to you, you’re playing a terrible
Rob Brisk
answer. What a terrible answer. The I think. I think it’s a difficult one. Actually, I think that I think a lot of the hype, the problem with a lot of the hype at the moment is a question of timescale. Honestly, I think that’s, that’s probably been the case for a while. I mean, I hate a hike has been so detrimental to progress in AI and healthcare, where AI could do such good and, you know, particularly in the UK, you know, we’re in trouble. You know, we need, we need innovation, we need change, we need transformation and disruption. But people are very scared of AI and with good reason, because a lot of what they hear is a load of nonsense for people trying to sell products. I think timescale is the big thing. I think I do see a world and I do think I would like to see it in my career where there’s a constant AI second opinion, you know, maybe maybe it’s true, like, you know, the, whatever the equivalent is to airports in 10 years time, and you just constantly have, you know, maybe an AI that can see the world through a camera that’s on your phone, and it can hear the world through a microphone, and it can just process everything that’s going on and talk to you like in the film her. Scarlett Johansson. Yeah, I you know that I the roadmap to that level of technology, I think is very, very clear. At this point, I don’t see anything. That’s a major, major stumbling block to getting that within our lifetimes. And I think it could be a really, really good thing. But I think the question is, if someone tells you, that’s going to happen tomorrow, like absolute rubbish, it’s almost sort of sell you this tomorrow. I think what we need to do is we need to build trust, we need to educate people. And we need to show them how you can use this technology in a safe and oversee of all way. And actually one of the things we’re doing at Otis is, we’re using generative AI to make clinical suggestions that are based exclusively on, you know, human created evidence guidelines, peer reviewed papers, that kind of stuff. And you know, in the idea is, you’ll be able to go into our app. So some of these features are still in development, you’ll be able to go into our app, and you’ll be able to ask a question, any clinical question. Now, how often do I? How often do I image a patient with a thoracic aortic aneurysm? And you’ll get an answer returned. But it will be from a guideline that’s been written by humans. And I think if we start using this technology in this kind of way, where we’re there’s kind of generative AI in the loop. But actually, you’re still practicing medicine in a very conventional way. So the generative AI is just hooking you up with the evidence, will begin to build trust will begin to build experience will begin to educate the next generation of clinicians. And then one day, we will progress to a point where we can rely more and more on what the generative AI has to say. And, and in the mean time, the technology will improve, and it will warrant the trust more. But yeah, to your original question. I think one of the important points and probably a reasonable point, as we come to the end of the discussion to just highlight is the fact that the only way we’re going to, we’re going to understand how AI can be used and where we need that oversight most is by having clinicians in the loop and understanding both sides of the house.
Nick van Terheyden
Yeah, I think great point. And I like that timescale. The one thing I would add that you you didn’t call out, but you sort of talked about was the trust issue. And that’s the whole, you know, making sure that a we’re delivering information that is at least reliable, I think one of the things that I’ve seen repeatedly, is this sort of binary, you know, it’s an answer, it’s it’s right or wrong. And that’s really not the way that any of this works. It’s not the way our brains work. It’s it’s a percentage sort of approach. And importantly, the trust, and you talk about it with you know, the earpiece air pods, I’m, you know, dating myself and say that I still have a pair of Google Glasses. I’d love to sell them some high high value considering how much they were at the time, you know? Yeah, I think we’ll see something with that imagery. We’ve seen chat GPT for you know, it is an exciting time. But as you rightly say, the importance of bringing clinicians in like yourself with experience, but also our colleagues who don’t need to be edgy, we’ve got to elevate the whole ocean to elevate the understanding because we need the clinicians who are working at the coalface to be able to contribute and deliver the insights as well as The appropriate trust and understanding of all of this to all the people that are essentially using it. I got to say, in so many respects very unsatisfying, because this conversation was not long enough, as we do always we’ve run out of time. I’m going to commit you, if you will, to coming back at another point in time to sort of continue the conversation if you would, but unfortunately, at this point, I have to say, Rob, thanks for joining me today.
Rob Brisk
Thanks for having me. Yeah. Been a pleasure.
