Med School Minutes

Med School Minutes- Ep. 47- AI in Medicine: A Talk with the Father of Virtual Assistant

• Kaushik Guha

Share episode

How is AI shaping the future of medicine and education? Join us as we sit down with the father of virtual assistant, Kevin Surace, to explore how generative AI is revolutionizing medical education and practice. 

Don’t miss this insightful discussion on how students and professionals can leverage AI for greater efficiency and success!


#AIInMedicine #VirtualAssistant #GenerativeAI #MedSchoolMinutes #MedicalEducation #AIRevolution #FutureOfMedicine #SaintJamesSchoolOfMedicine #AIForStudents #LearnWithAI #MedicalStudents #CaribbeanMedicalSchool #AIandAcademia #InnovationInMedicine #HealthcareAI

Speaker 1:

Hello and welcome to another episode of the Med School Minutes podcast, where we discuss what it takes to attend and successfully complete a medical program. This show is brought to you by St James School of Medicine. Here is your host, Kaushik Guha.

Speaker 2:

Ladies and gentlemen, thank you for joining us on another episode of Med School Minutes, where we talk about everything MD related, with a focus on international students, specifically Caribbean students. Today we have a very interesting guest. His name is Mr Kevin Serais and he is considered to be the father of the AI assistant, or the father of the virtual assistant. He is a Silicon Valley innovator, a serial entrepreneur, ceo and a futurist. He has been Inc Magazine's Entrepreneur of the Year, a CNBC Top Innovator of the decade and a World Economic Forum tech pioneer. His credentials go on and on, but today we're going to talk to him about how AI is really changing the landscape of education at large, as well as, more specifically, medical education, and how students can best leverage this.

Speaker 2:

So, without further ado, let's welcome Mr Suresh. Thank you so much for joining us today at Med School Minutes. So today we're going to talk about AI, and the biggest question is you know, when you go online, you know you look at Instagram threads. I'm going to be honest AI quote unquote experts are a dime a dozen. Could you tell us a little bit about your background and why we have you here and what really makes you an authority on AI?

Speaker 3:

Well, look, I think everyone and their brother who plays with ChatGPT all of a sudden says they're an expert, right? Um, I invented the virtual assistant back in the 90s. I've been, uh, I have 94 patents. Some of them are back here. Uh, most of them in ai and uh specifically applied ai. That's very different than core ai research. So there's ai research like I'm developing brand new models, like at open ai. Those people are way smarter than I am. And then there's applied ai, where we take these ai, whatever they are, all the way from the 50s to today machine learning and AI and Gen AI and say how do we apply these to real human problems? Right, how do we build something big around them? That's really formidable. So I think the virtual assistant is a great example of that, where you can leverage a lot of different models.

Speaker 3:

In the 90s, the models we had were much more simple than they are today, but everything that became Siri and Alexa and all of that came out of my team at a company called General Magic. So since then I worked on AI for HVAC, building controls, things like that, energy savings that became the train energy manager. It's used in thousands and thousands of buildings today. I've worked on these days, an AI system that finds bugs in our software not necessarily consumer software, but big enterprise systems typical enterprise Most people won't know might have five to 10,000 applications that they manage on their servers, and so you really need AI to find all those bugs. Humans there's just not enough humans to do it. So that's been a heavy lift. It's been fantastic work. Gen AI is helping in that, so I continue to march forward.

Speaker 2:

As they say, Awesome and I definitely want to talk a little bit about your TED Talk. You do have a TED Talk. We're going to link that when we put up your podcast so everybody can see and, just generally speaking, I think you definitely are an expert, but with that, we obviously are a Caribbean medical school and a large base of our viewers are either students who are in our school or students who are looking to go into medicine or typically, just generally students who are interested in medicine. Now the big question is when chat GPT or OpenAI came out with this whole concept of chat GPT, there was a big buzz around it and I think the buzz hasn't really subsided yet. If anything, we're getting used to the buzz, getting used to the buzz. But in academia in particular, there seems to be a lot of skepticism about AI. How do you, from your perspective, see AI changing the world of academia in general and maybe medicine specifically? I will talk about that.

Speaker 3:

First of all, I'm on the board of Rochester Institute of Technology. We have over 20,000 students, so I am very deep in these conversations. I've been on the board for 20 years and we are an AI powerhouse in our own right. But of course, gen AI changes a lot of things. For example, many professors believe teachers, professors etc. Believe that students should not use, say, chatgpt to help them do their work. I say you're living in the dark ages. Of course they're going to use it, just get over yourself, right? So if you start creating your coursework and what you're expecting, knowing that students are going to use this tool just like they would use a calculator or Excel, well then you start rethinking how am I going to use this tool, just like they would use a calculator or Excel? Well then you start rethinking how are we going to evaluate the knowledge these students have and the critical thinking skills that they're bringing to the table? Right, they're going to go get their answers from ChatGPT. It doesn't matter, they were going to get their answers from Google. Maybe this is faster. Whatever the case is, the papers are going to be written this way. How do I, how do I evaluate? Uh, for example, you know how, um, how knowledgeable they are. So an example might be we know the students are going to go home on these medical questions and they're going to get their answers from chat, gpt and their copy and paste it and put it there. Great, in class I'm going to ask each one of them, or separately or whatever, okay, what makes this particular answer right or wrong and why? Okay, now, this is interesting because now we're getting into the way we're probably going to get used to these tools in the work environment.

Speaker 3:

Okay, and a med school, and I know a med school. You've got to get doctors through or nurses through that pass their court, pass their um, you know exams, ultimately to to to, to be able to practice medicine. But in the end, we also have to produce doctors, nurses and other medical workers that are going to work in this new world where Gen AI is just something we have, just like a calculator, and so they are going to use it, because there's no one that you can train at the St James School of Medicine who will know every single ailment that I might have, having just come back from you know, sub-saharan Africa, right, there's just no way, because they don't see that every day and you can't train them for that every day. That's not possible, right? Right, so they are going to use tools, and the old tools were go to some big medical book, pull something off and try to find something. And then we kind of had Google search and people have been using that. And now I have this thing that could be trained, and there are medical versions of, say, chat, gbts. They're fronted with a large language model but behind them we use a process called RAG to limit their responses to be within the medical literature.

Speaker 3:

Right, this is fascinating. So you won't try to be an expert on every possible ailment someone has. You're going to leverage the symptoms that they're giving you and all the background they can give you and say what are the possible ideas here? Right? So you start to use these as ideators or idea generators, or ideation. It is very, very, very powerful as ideation, right, and now you know you do. What a doctor does best is take these and say, boy, let me test for these three things. I've never heard of two of them myself, but we better go find out if that's what happened here. It's possible one of them could kill you. Blah, blah, blah. Right, this is powerful, and so I look at this as a way to allow doctors to be more productive, more powerful, more important, because they're going to interpret this information, but they're going to have more information instantly at their fingertips and use their experience and schooling to then interpret the best way to approach the patient with this, including the human sides of that, the EQ sides of that right.

Speaker 3:

You come back and the patient says, well, gee, this hurts and I traveled here. And they come back and say I think you have a worm in your brain, right, there's a good and a bad way to say that. Right, there's probably no really excellent way to say it, but let's let's talk about the potential that something could have happened like what a worm in your brain. Right, someone's going to have to say that to someone, or you know you, you're facing maybe only three months to live or these sorts of things.

Speaker 3:

These are the hard things the doctors have to do. So I would like to see doctors less worried about the ideation.

Speaker 3:

So I would like to see doctors less worried about the ideation they have a machine that does that, okay, it says it's probably these five things or these three things, or do these, following tests and you'll find out. And more concerned with how do I really have the time to show empathy to this patient that I'm treating? That has maybe a serious issue, maybe not a serious issue, maybe the family member has a serious issue, whatever, and I think we're all going to be better off. Number one. Number two there's a shortage of doctors, so nobody needs to worry about this. Right, there is a shortage, there's going to continue to be a shortage, and everyone I mean you go into family practice. You don't make enough money, there's not enough time in the day. You need these tools to give you some of your time back. I'll give you one example. I know you got a million questions, but I'll give you one example.

Speaker 3:

Where it's being used today is taking doctor's notes, right so throughout the day. Why would you ever take your own notes anymore? Let the darn thing record, let it transcribe and let it summarize and put it into your epic system at the end of the day Brilliant. You, as a doctor, are spending two hours at the end of the day trying to do this from your notes in your notebook, and now I have a machine that does it for me and it does it better, just like StreamYard today. I don't know if you're using this capability. We'll transcribe everything we said, It'll summarize it and it'll post it right there. And that's work you used to have to do at the end of the thing for two hours, and now you don't do it at all. You push the button, it's done. Maybe you'll review it and edit it, or just push the button. It's a summary, right? Brilliant use of technology, brilliant use. This is good for all the medical students. It's the best time I'd say it's the best time going forward to be in medicine.

Speaker 2:

Oh, wow. So you know, one of the big critiques that our faculty members not just our faculty members, but just academia in general has about AI is that one of the biggest principles of medical education is lifelong learning. Like, the students that we produce have to become learners for the rest of their life and you're essentially giving them quote unquote the tools to be able to function anywhere and learn things and adapt to their environments. Now, a big criticism about AI has been that students are not knowing the true principles of research. What would your response to something like that be?

Speaker 3:

Well look we have lived in a changing world of technology since the invention of the wheel, and when the wheel showed up, people who used to carry stuff up on their back at the top of the hill from the ship all of a sudden could cart it up right, and so that, fundamentally. And then some people would say, well, what's going to happen to their strength of their back and what? This isn't fair. Okay, look, Excel came out in the late eighties. If you were in finance, most jobs in finance were pencil and ledger book and it literally went away. Okay, Gen AI hit and people are already saying no one. Who who's going to learn to write by hand anymore when you have a machine that will write a blog post in 10 seconds and it's better than the one you could have written? Here's my answer to that we are all going to use Gen AI to write our blog posts. The days of writing that by hand are just not needed. I'm sorry. The truth is, we're going to try to still teach students oh, you need to write an essay. They will never write another essay in their life. This is a skill that's no longer needed. Just like we teach students long division, Okay, so that they understand the fundamentals of long division. They do not need to ever do long division again in their life. Yeah, Ever. The last time you did long division in your head was, I don't know, fourth grade or whatever. I mean these. So we cannot believe that humans have to have every skill and every understanding of every skill, because technologies come along and take those skills from us. We don't need that skill. So it's one thing to say. Here's how it was done in the old days. Like we don't learn slide rules anymore. It was probably a valuable thing in 1952. It's completely useless today. It's a useless skill thing in 1952. It's completely useless today. It's a useless skill.

Speaker 3:

Many of the things with all the respect to great people at your medical school but this is across all medical schools many of the things we are teaching these students today who will come out as doctors or surgeons or researchers perhaps, or whatever it is they will never use again and actually are quite antiquated. It's just that's part of the coursework and we have to do it. And this is true at RIT. We do the same thing in all of our courses. We look at it and go this is antiquated, but it's the thing that's approved. So I've got to keep teaching it until someone tells me to stop. We're not doing our students any good doing that.

Speaker 3:

What we should be doing is embedding large language models, transformers and other AI into our courses and saying when you leave here, you will be using these tools. Here's how you best learn how to use them. Here's how you best learn to critically think about their outcomes. Right, so I might put some things in and it comes out and you go. Four of these are brilliant. I wouldn't have thought of them, and one of them is dead wrong. I'm not going to go do that. Right? So you still need your critical thinking skills Very much. We need critical thinking skills, Right, but you don't need to remember all this information and your lifelong learning of. I am going to somehow remember. When someone comes in with that symptom, I might check for this Useless. There's a machine that does that. Now, you don't need that skill, but you need the critical thinking skills more than you had before. Right, Look at its output, right?

Speaker 2:

Right.

Speaker 3:

So let me switch gears a little bit and ask you about, you know, with the advent of chat GPT, which is obviously the most visible Gen AI model out there and accessible it's the best model overall that if I were building a legal data or a legal interface or a medical interface or whatever, I would use the front end of ChatGPT to have my English interaction and the back end I would say don't learn from this medical database or medical or whatever the case is.

Speaker 2:

So with ChatGPT, we're looking at about Gen AI being accessible to everybody for a little less than a year. I think we're going close to a year now. If I'm not mistaken, yeah, a year plus A year plus. So the big question is we haven't seen any major shifts in education because of generative AI. It hasn't happened yet. So from what you're telling us, it seems like we need to make those shifts and from our conversation, my takeaway is that it almost seems like students have to be more efficient or will have to be trained to be more efficient to be able to use this Now. Do you anticipate at any point in time where our education system across the board not just medicine, but become so advanced that a six-year-old is being taught? I don't know? Now the regular curriculum becomes high school biology and our college level stuff becomes high school biology and our college level stuff.

Speaker 3:

Well, I'll answer that in a couple of ways. The first thing is that educational systems move glacially slow for a reason. So things in the tech world and in the commercial world and stuff you know, swing back and forth very, very heavily, very quickly, right. Technology's in, technology's out, this and that, and universities, in particular K through 12, also move at a glacially slow pace, and they've always moved slowly because they don't want to be reactionary every year to the latest thing that's happening on the web or the latest technology that's in over here, because it'll be out next year, right? So it was designed to do that. Now the problem with that is that technology is moving at a faster pace every year than it used to, and there's lots of reasons for that. For instance, chatgpt was available to 6 billion people on day one because we have an internet and we have smartphones, so everyone worldwide could technically use the free version on day one. What used to take 40 years to spread around the world now takes 40 minutes, right, or 40 seconds. So the educational system is still stuck in. We'll change every decade or two a little and it's possible that we're now getting to points, especially with Gen AI tools, where they really need to change faster and they really need to embed this because the employers are looking for people.

Speaker 3:

So I'll give you an example. Let's say I have a big doctor's practice. I've got, you know, 40 family doctors in this thing right. And I've got an opening for one new graduate next year right. And I interview two people. And I interview one that says, yeah, I've heard about that Gen AI, chat, gpt stuff, but I've never used it. It's probably unreliable and I wouldn't want to do that. I played with it once, but I don't think it's a good thing. Okay, tell me about your other medical history and what you came through and what you learned in school, et cetera. The other one comes in and says here's all the things I learned in school, but one of them included the use of Gen AI for my entire practice, and let me tell you why that's important, and I measured it made me 42% more productive. I was able to summarize things at the end of the day, I was able to diagnose things with ideation that I had never thought about before, and so I'm just going to be your most productive hire. Who gets hired? Yeah, number one number two.

Speaker 2:

Number two definitely number two, yeah.

Speaker 3:

So if you're watching this, you go. You know our job as schools is to graduate people who get hired. That's actually yeah, that is the. The customer is the student and finally, the people who hire them, right, the hospital doctor's office or whatever it is. And so you want to make the best, most hireable students and the most hireable students, the most wanted students today, walk in as student number two. I know all about this stuff. I you. I even used it when the teacher said don't use it, and I used it anyway because I learned all these other things from it and I used it to cheat. I learned so much more and I was able to critically think through those results. I go, I'm hired. Do you have more like you? I'll hire you. I don't want the first one, so think about who we're graduating, right right.

Speaker 2:

So I want to pinpoint something that you said how tech tends to change and tech seems to be, you know I mean relatively speaking a little esoteric, because you know, one day you have Blu-ray, the other day you have HD, another day you have both of them and then one of them is completely gone, et cetera, et cetera.

Speaker 3:

That is true. You're going back in time about 25 years because HD DVD got killed off in its first year or two. But I still have a player and I still have some discs. But Blu-ray won that right.

Speaker 2:

And I still have a player and I still have some discs, but Blu-ray won that right and this does happen. So, and you know, there are things that become completely obsolete, For example cassettes. Nobody needs them anymore. Nobody needs a cassette player anymore. In your view, Do you know?

Speaker 3:

I have to interrupt you. I just read an article last week that they're like LPs, like vinyl. There's a little resurgence in cassettes and these certain kids now, like they're teens, they think it's really cool to go and buy a cassette of Taylor Swift and put it in a Walkman and literally just listen to it the way it was meant to be listened to, and it's become this cool thing. There's actually some resurgence of cassettes, I kid you not.

Speaker 2:

Anyway, I know your point, but that's a nostalgia factor, absolutely, and I've seen that in LPs and vinyl records and whatnot. But in your opinion, do you foresee that Gen AI is potentially replaced in the next three to five years with something even more advanced that is completely different from the way Gen AI is used? For example, I mean, hypothetically speaking, a chip that we implant in our body that automatically gives us all this information? Again, I'm no. Yeah.

Speaker 3:

So let's look at it this way Look, a transformer model is a great model of a language English or any other language right.

Speaker 3:

And it's read trillions of sentences, let's say trillions of tokens, and therefore you can ask it things relative to the things it's read and it will come back and give you some interesting information about that, right, that's the bottom line. So the concept of a transformer or whatever, replaces a transformer. It's correct, it's, it's, it's. It's an amazing way to learn everything that's ever been written and build a model around it, right? So that's pretty interesting, right? So I don't think that goes away.

Speaker 3:

How we interface, look, I invented the virtual assistant. So I've been saying interface with voice since 1997 or so. Voice is the most natural way, it's the fastest way to interface with anything, including computers, including humans, right? You don't see me writing things down, typing them and handing them to you. I can't type that fast, right? I might as well talk to you and you talk to me. So I think we will continue to move towards an era where more and more of our interface to the world is through voice. When it's appropriate to use voice, right, can't do it in a crowded place all the time, and so you know, clearly, ultimately there's a lot of work going on in brain-computer interface. I, you know from what I'm seeing. We are years and years and years, maybe decades, away from having a very high volume kind of interface. Today it's like maybe you know we can get the A to get typed out the thing you know by thinking about it right, but it's very crude today right, we're at

Speaker 3:

a very, very, very crude level. Someday, clearly, we're going to, you know, implant a chip, maybe by the end of the century, and we'll just be interfacing, but still the transformer will have to be there. Right, we've got to understand the language, understand everything that was there, build models around it. So, whether it's a transformer or some other kind of neural net, as Sam Altman said recently, everything we've done with transformers and chat, gpt and everything else and gemini is paying homage to the fact that neural nets actually worked right. Deep learning work period full stop. We weren't sure it'd work, but deep learning, which the math was kind of done in 2012, it works, and that's all this is. We're leveraging those deep learning models and building out these huge, huge, huge models that are thousands, millions of layers deep right.

Speaker 2:

I do want to touch upon a lot of this. Ai stuff is very intimidating to a lot of people youngsters as well as people who are more advanced in age and they keep thinking that, oh my God, this is like computer programming. What sort of technical details or what kind of skills do you think a person needs to have to be able to really leverage, uh, this, this information? Number one and number two a follow-up to this question, essentially is there is so much I mean everything out there seems to have ai nowadays, whether it's your phone or whether it's even your like vacuum. But the question is and it may not be the real AI or generative AI as we talk about it, but the real question is how do you sift through all this information and do you need any technical skills to be able to do that and become efficient at using Gen AI?

Speaker 3:

Yeah. So look, I think there are hundreds of web applications and mobile applications out there that can do interesting like. Hyperwriteai has already kind of pre-set up all kinds of prompts for you so you don't have to think about prompting correctly. But here's what I tell people stop playing and start doing so. I do 40, 50 keynotes a year in front of audiences sometimes 5,000 people, Right and I'll ask the audience raise, raise their, raise your hand. If you played with something like chat GBT, everybody raised their hands. Now, raise your hand.

Speaker 3:

If you're doing real work with it every day, everybody's hands go down. Like two people put their hands up, right, and then I say that's a shame, because all of you are wondering gee, I wonder what I can do with this. I wonder how do I use blah, blah, blah. Okay, you will never learn if you don't learn to do actual work. So I use these models every single day, all day. I generate images, I generate diagrams, I analyze spreadsheets, I ask it how I should reply to this email. I generate all kinds of content blog posts or technical data or whatever. So I don't generate a thing without first asking a large language model or a transformer right or a multimodal, Okay.

Speaker 2:

Why.

Speaker 3:

Because it's a better generator of content than I will ever be. Period it read more than I can read in my life. I can't change that. That's how it works, and so what I want to be great at is asking the right questions and getting the right answers, and then going through those answers and picking out what's important to me and how I want to change it if I want to change it at all and critically think about that. So the skills that I need going forward the skills we all need going forward is doing that for every single thing during your day.

Speaker 3:

So, for example, you could have said should I have Kevin Serais on my podcast? It will opine on that and you can decide whether it's good or bad, but use it as an ideator. Who else should I have on my podcast? How might I contact them right? What should we talk about? What might be some of the topics that students might be interested in, or or professors might be interested in, or teachers might be interested in right, or administration might be interested in?

Speaker 3:

So any idea you can think of, it's an ideator next to you and you can have 100 brain power using that thing, or you can sit there and have one brain power, which is fine. It's a great brain, you got a great brain, I got a great brain. But the person next to me has got 100 brain power. Well, how am I going to beat them? They got 100 brain power. If I'm a doctor and I get 100 brain power, I'm crushing the doctor next door who's only got one brain power, right, you need to think of it that way, and once you start thinking about it that way, you go. I want 100 brain power too. Okay, then stop playing. Get a real subscription at $20 a month and use ChatGPT we use GPT-4.0 and connect it to live web information and then start really using it for every single thing you do, and within weeks you'll go. I'm getting really good at this. Now I see the power. Otherwise, you don't see the power.

Speaker 2:

I see, and so I mean the gist of it should be that just jump in and not worry about what you already know and don't know. And then you know and I use chat GP for a lot of daily tasks and I would say that, at least in my organization, I'm probably one of the more advanced users, so to speak. But and this is a conversation I have a lot with our faculty members it's like oh, I don't know how to leverage this, I don't know how to use this, oh, I'm very intimidated by it. It's like oh, I don't know how to leverage this, I don't know how to use this, oh, I'm very intimidated by it.

Speaker 3:

Start Start at your start. Start doing real work. You've got to get out of the play mode. It's like the difference between playing a video game and writing a video game right. Playing a video game, you're not learning anything other than just playing the game. Writing a video game game, you're learning everything about what it takes to write a video game. So stop playing the video game, stop playing with chat GPD.

Speaker 3:

Get a subscription to GPT-4-0 right and actually work with it and do real and say I'm going to do five tasks today with this. Then I'm going to do 10 the next day. I'm going to take every email and have them have my response written by gpt40 and I'm going to edit that. I might not accept it, I may have it, rewrite it. And then you start to get good at asking the right questions. So, um, hey, I had an argument with my spouse this morning. How should I reply to him or her in email? Now that will save our marriage, trust me. The results that come back from that darn thing you, you could never beat. You can't beat it, you can't beat it. It's better than anything we can write. Why? Because it read every single thing that's ever been written about that subject.

Speaker 2:

Right, right, that's amazing. Well, thank you so much, mr Suresh. I know we're a little pressed on time, but this conversation for us could for me at least could go on forever. There's just so much to learn from you and I really appreciate it and genuinely, it's been an honor and a pleasure to have you on our podcast. And again, thank you so much for the time and thank you so much for literally laying the groundwork for us having a better life.

Speaker 3:

Yes, well, I think it's. Look, it's a great time in technology and we've seen so many big changes in technology the internet, smartphones and now generative AI. These are huge game changers and there's more to come, so we'll be back on the show another time, take care. Thanks so much for having me.

Speaker 2:

Thank you so much, mr Sarris. We really appreciate the insights that you provided about AI. As you know, we always see hundreds, if not thousands, of AI experts, but we're really honored to be able to have a real expert and the father of the virtual assistant on our show tonight. But once again, thank you so much, mr Suresh and everybody out there, make sure you polish up your AI skills and work on prompt generating and prompt engineering to be a more efficient student, and always remember there is no shortcut to becoming an MD. If you like the content that we produce, please give us a like, follow and share. It goes a long way, especially for our production team, who put a lot of work into producing these episodes and, if you feel free, to download our other episodes from Spotify, google or any other platform that you prefer. Thank you so much and thank you for your support.

Speaker 1:

Thank you so much for tuning into our show. We hope you enjoyed another episode of Med School Minutes. If you like our content, please follow us and receive notification when a new show is posted. This podcast is brought to you by St James School of Medicine. For a video version of this podcast, please check us out on sjsmorg slash video.