Welcome to Real Talk on AI and I care with your host, Dr. Scott Morris. And I'm co host, Dr. Rayhan Ahmed.
In each episode, we discuss current news, add in a little AI education, and debate innovative AI tools and topics that will change the industry. Hello, everybody. This is Dr. Scott Morris.
This is Dr. Rayhan Ahmed. So today, Rayhan's going to cover in the knowledge bite something really interesting. We've been having a lot of conversation about and it seems to be the hot topic, which is the claw.
As I always call it, Rayhan Philcy. Yeah. So Scott, a couple of weeks ago, you talked about AI agents. And the reason I love these knowledge bites is that they continue to build upon each other.
And we're learning as we go through this. I mean, I learn every time we do one of these. We are sort of doing as moving so fast. So this is sort of stuff that's a cut off the presses and community on LinkedIn and all these messages.
And you board they talk about it. We'll be talking about applaud bot or open claw or has a couple of different names and notebook. But basically, these are AI agents and Scott, you did that great talk on a couple of weeks ago. But they they're the ones that live in your computers.
And the first time I actually saw one where I was in, I was talking to my coffee with a couple of friends and you know, we were talking about something and someone wanted to look something up there. Let me just ask my claw bot and he got his cell phone and he just texted. Now he didn't go on chat. She'd be tea didn't go on cloud or Gemini.
He just texted his own computer, his own agent to do something for him on his computer that was not with him at that time. So there's a ton of discussion about this. And Scott, you may have heard about this interesting story about her character of safety and alignment her name's summer you. Her entire job, I mean, we have day jobs as I care providers, but her entire job is making sure AI systems don't go rogue.
And in February, she had open claw, which is one of these AI agents, probably the most popular one. Go through emails and recommend what to delete. And she even told it, confirm with me before you do anything. She didn't.
The agent began deleting her inbox without permission, even after she told it to stop multiple times. If you, the article is hilarious. I recommend anyone to interest to Google it. She literally texting in all caps.
Stop, stop, stop. She couldn't intervene from her phone. And she literally, this is hilarious. God, she had to go physically, run to her computer and literally pull the plug.
It was like, it was like 2001 space Odyssey type situation. And this is her job. You know, this is her job. And then, so this is the world we're stepping into.
And I don't know. I so I use agents and I. And it makes me a little bit nervous, but. I'm my agent and is amazing because I have I literally had.
I thought I had a couple hundred files in my downloads folder, by the way. And actually eight thousand files on my Mac book downloads folder. And I asked my agent, hey, can you clean this up for me? And then said, you know, you have a lot of duplicate files.
Are you okay with me deleting it? And I pause it a little bit. I thought, gosh, well, I don't know. Are you going to go rogue on me?
But I really don't want to go through the trouble of me. It'll take you 10 seconds. And I'll have all these. So I allowed it and it turned out fine.
Thankfully. And it organized it beautiful. But we are living in a world that as AI gets more powerful and we're seeing its results that people. And I'm going to get more and more comfortable with them taking autonomous actions.
And that is the sort of the. The syn<|ca|> This is autonomous AI agentic tools dare I say intelligence that access or files or email calendar software and virtual. It's your virtual personal assistant, right? They're just that's right.
That's right. Instead of, instead of, you know, what what should I do as a person you're saying, you know, just go do it. Go do it. And so it's a totally different risk profile.
Because once it can act and act autonomously, you kind of make sure it stays within the boundaries you set. And it's not guaranteed as the storage with with the Facebook employee demonstrated. It's getting really interesting. So there is, if you've heard of Reddit, which is a form where people sort of have a message message board.
There is something called molt book, which is a platform. Not for humans, but for AI agents to communicate with one another. Oh, I don't know if I love that. I mean, this is not science fiction is happening now where they're comparing notes and talking about.
Whatever it is, AI agents talking about. You know, say you don't like, but think about think about and you know, we talk about good case. Before you go on, I just want to say that you know that that you know that their security officer quit, right? After that, she quit.
So you left Facebook. Well, that's probably not a bad idea. That's not a bad idea. But think about, I mean, it's doesn't think about all, especially now, let's let's go back to I care.
Think about what this could do and I care. I mean, the possibility is our endless thing about an independent practitioner who has, you know, he may not have enough staff, you know, step maybe on vacation. Maybe you want to verify insurance. Maybe you want to follow up with patients in the last week, who, you know, maybe didn't make their appointment or something or maybe you want, you know, there's prescription refills.
The big pain point. What do you mean nice to have an AI agent do some of this? I mean, there's this world. We're looking at it.
And so, so I just. We've been working on all those things the last couple weeks and it's. It's going to be able to do. I mean, the agent's going to do all of that.
But then I think about what you just said. And I'm like, oof, what happens to the cut in and eliminated all of the OCTs on my files? Right. Yeah.
So there's definitely a lot of security concerns. Personally, I. I was this close to getting a lot of bought. I got it my own MacBook mini.
I was going to sandbox it meaning put it in separately for my files. I decided against it because there are just so many security issues there. I do encourage our listeners to keep, you know, we will keep you guys updated. I am, I am experimenting with other types of agents on my computers that are probably more sandbox and have a higher degree of security, which is.
Very important for us, of course, with hip and just work with the work we do. So definitely this is a topic we will continue to revisit. I think the agentic revolution is upon us. And really excited to continue talking about this.
Well, I'm going to take that we're going to go on. We're going to take that. We're going to take that agentic future and we're going to roll that into something we really haven't touched much on. And I am quite sure that I am going to get some love of mail on this one.
But, you know, we're not allowed to stay away from controversy. I want to talk about university 2.0. Right. I mean, I, I, and we've kind of hidden it a couple times like my kids in college or, you know, in grad school or whatever and what they say.
Now, you're talking about university, are you talking about optometry school 2.0? I think we're talking about a comedy school medical school undergrad 2.0. I think there's about to be a major AI academic revolution that's happening. And, you know, I think that's, I think there's a couple things that, you know, disruption is a, is a nasty and inevitable concept, right?
And I think that academia may have be more disrupted in the next five to seven years than any other industry there is. Any other industry, maybe with the exception, maybe FinTech, right? But I mean, it's, it's going to be up there. I think that so there's a couple of things I always look at and go, I think this is the ends, right?
I call it the one size fits all lectures format that we all learned. I learned, like, y'all go into a room, you know, as a 300 person or a 50 person or whatever lecture hall and you all listen to the same lecture, you all get the same information. And you can't really ask questions. And if you can, you probably don't get answers.
And everyone listens to the same talk at the same pace and same space. And is that really the best way to learn? Well, the answer is not even close, right? Is that we all learn in different ways.
We all learn some people have really short attention spans. Some people have shorter than that. If you've already tuned out, you were one of them. Right.
So I think the end of the one size fits all because, and you know, we used to think Google would do that. But now you look at Gemini and GPT and Claude and go, it could teach me way faster than I could listen to a lecture. I watch my younger son who's in school and he's in his third year of finance at CU and I said, well, do you listen to lectures? He goes, yeah, I listened to him.
But I then when I don't understand something, I go to Gemini and actually haven't teach me. And I'm like, oh my gosh, right? So where is the professor and the TA in this? And the answer is not there for some of these students, right?
It's not there because the students have figured out other ways to learn. And I wonder is our eye care. Academic professionals understanding that this isn't just going to be undergrad. This is if it doesn't already exist.
It's going to be very soon in the house. Think about my two student doctors. They later really super interesting case. And I said, well, tell me about it.
And then they both I had something else come up and I come back and they're both ones of Gemini and the other ones on cloud. And I said, well, here's what it is. And they just give me this laundry list of stuff. And it couldn't event 60 seconds.
Yeah. They didn't go look it up in the textbook. They didn't really go to the internet if you would. They actually asked, yeah.
And I thought, how does that change the one size fits all? And I think we're going to go from a world where everything standardized to everything's personalized. Yeah. I was recently talking to a first year medical student.
I said, you know, are you using AI and I said, absolutely. And what he loves what he does is that he learned something he tries to figure out then he goes to chat to you or Gemini and says, explain this to me like I'm a fiber old. Like go through this. And then it you press your test it you ask other questions.
You can ask what would a professor asking me on a test. What are some key confusing topics. And so I think, you know, that sounds great and all. But what happens and I'm being more about ophthalmology and optometry.
You're seeing a patient you get an OCT. And what if it's just you hit the easy button on the OCT machine, which we know that all that stuff is coming. And it's better than it's interpretation, maybe better than 80 90% of specialist retina specialist because it just sees everything. It could give you a differential right away.
What do we do then? How do we know? I want you to think about this. Okay.
And I'm a little older than you. So, you know, when OCT first came out, we were all looking at near fiber layer with a 90 doctor lens or 78 or direct up. And then OCT came out and they give us all these numbers right. We're like, oh cool, but none of us had any idea it was the numbers men.
So what they do they give us a red green yellow dummy light right. And so what do we all do? We saw what's red is bad. Even though we didn't understand the numbers.
So we had a tool that we grasped onto and interpreted data for us. And we accepted it is true. Now obviously years later, still a year 90% of people look at the red green yellow dummy light right. We didn't really change.
But some people realize that's not true. I think we're going to go through the same evolution. So imagine you know imagine you're talking to a dean of a school and I've come to school or even medical school. What is your what's your advice to them, you know, they come to you say, hey, you're the egg.
But how should we how should we change what should we do? Should they start incorporating these tools and their classrooms and in clinic or should they maybe not and say no, we have to do it the old old school and what's your thought. Well, I think there's I think there's an argument for both directions. I think if you're at a med school or an obstacle and you're not at the cutting edge, you don't have a competitive benefit when it comes to recruiting the next next year, the next year, the next year students, because they're going to those students are comparing.
Go ahead, hey, your schools worth this much. This is reputation. This is your board scores. This is your technology platform or technology stack.
It's part of the competitive. I think you're still going, well, we're still doing golden bowl for imagery. You know, it's it's not going to help you from a competitive standpoint. But I also think on the other side, as you're right is that and we talked about this, I think two or three weeks ago is that.
Our role is going to change from data gatherer. And I think where it's already left data gather and I think now we're kind of in data interpreter. I think that's going to change too, because I think like you said is it's seeing stuff at a micro vascular level. Our eyes will never be able to see it's seeing pixels, you know, not blood vessels.
Yeah, and I think. It'll go from data gatherer to data collector interpreter to I think now maybe the human does what the human does best is the communicator aspect data trend. I was going to say we're a translator. Where how do we communicate with that?
And I still think we need human in the loop to go. Sometimes there's got right you look at an OCT of a mac and you're like, you know, there's more to this story right and that comes with experience. But how do you get that experience? There's the other part of academia that I think is going to change a lot and maybe not just academia, maybe continue education is because I think this hyper personalized learning is coming down the path when we're going to be able to run you talks about virtual twins right is that.
So are we going to build a have a virtual clinic where we're presented a case and we go down one set of algorithmic path and we get to the end and we're like, oh, that's internal. We want and so we can clinically back up one, two, three decisions. And then go down a different path without ever actually harming a patient right, but we have these models of. We start collecting data and that's a whole universal data system situation, but not let's say we're collecting data on thousands of individuals.
You really think they I can't put together virtual twins to go what happens if you go this way what happens to go this way what happens if the person instead of being a 55 year old guy is a 25 year old woman right and would puri get into geno, you know, before you get into. Type ready that kind of stuff or this person's on a blood pressure medication this person's this person's hypotensive this person's hypertension right and now all of a sudden the student can learn in a virtual setting and see hundreds of different patients in a day. Without ever stepping foot in a practice, whatever stepping foot in an interaction now that's good I think that allows for lots of reps right and I think i'm interested in to see your viewpoint on how that's going to affect from a surgical person. That's right from a surgical perspective like hands on, but on the other side of it I get also get worried going our humans going to lose the ability to communicate with other humans.
Because some of the most important things you learn in clinic are how to talk to people. Yeah exactly right exactly right I'm. I learned surgery stuff I mean I learned. So I 20 years ago I was a 30 year medical student at Vanderbilt and I remember being in a hall because everyone wanted to everyone who was in med school I.
Many of my friends all want to be neurosurgeoned and so the chair of neurosurgery was talking about what it takes to get into a neurosurgeoned neurosurgeoned medical residency and teaching neurosurgery. I remember a lion stuck with me he said I could. Neurosurgery and surgery general is not about technically performing surgery is that I could I could train a monkey to do surgery that's not the issue the issue is. I could train a judgment of when not to perform surgery and that's not always clearly teachable in a textbook that doesn't experience and speaking with not just the patient their family their spouse even today out of out of patient who was just really scared of surgery.
I had to just sort of talk through that and some of where their concerns coming from and it's just kind of a long conversation about historical things about mistrust of the medical establishment and mistrust of medications and mistrust of the pharmaceutical industry shows. I mean these are what I'm really hopeful for is that that's a big reason for I think where I lot will win into medicine in the first place is the connection with other humans. And so that's where the humans in this AI future will shine is those who can make that connection and talk to patients in a way that makes them comfortable because all the gathering of data and the interpretation of data and even. And the replacement of what you learned is all there I mean the answer to their you the question is how do you train someone to synthesize all that information because if there has to be something there and that's what I struggle with.
But you still need someone to be able to communicate in a way that demonstrates their experience and judgment and again when not to do something when maybe the computer saying eight technically that's the right thing to do is like well that's not what the patient or the patient. Or the surgeon or anyone. Yeah you got to you can read body language and you know their head shaking their head shaking yes but their body saying no right so they're saying yes but their body saying no. You know but I also think I will extend it one step further than that you know and it was really interesting listening to my older son was he was studying for his LSATs for law school.
And he took this program where it wasn't about taking an exam anymore it was very much very short pieces did you pass did you fail and so instead of going to take in a four hour exam practice exam or let's say let's let's convince the stacademia you know we for the last couple years people been saying well people are going to use AI to cheat well. The internet to cheat they use no cards to cheat they use audio pens to cheat they use their cell phones to cheat. People are going to cheat right but maybe now we start integrating in how we provide education as well as how we grade it because now all of a sudden instead of having these really big high stakes midterms or finals where all your grades based on one or two tests what happens if instead it was a 60 seconds every day right or two minutes. At five different intervals during the day just boom boom boom boom where it wasn't about cheating was about learning right instead of grading and saying you have to.
You know you have a hundred point midterm you have a 200 point final and your grades 300 points now maybe instead you have a 2000 point semester. But it's three four five points every day right and so and can we learn what people learn and what people remember and what they regurgitate and do can we track gap analysis we track learning I think the exams right and I'd say exams in a school and maybe maybe and I can't speak for ophthalmology but for optometry the board exams you know I thought we're doing the board exam wrong for the last 35 years and that clues when I started when I took up you know is that. Are we really testing in a two hour test everything you need to know to help patients we're not we're just testing pieces of it. You know I I think that instead we're going to see a continuous assessment we're going to see much more.
Individualized augment authentic kind of assignments where instead of writing some stupid generalized essay paper case report maybe instead we'll have a way to just use a virtual twin and do a very customized one that the egg and upgrade. And you know then maybe I think about I read a lot of Solomon Khan stuff his books about education he's a fascinating guy and he's like no it's all going to be about a human AI collaboration of how do we teach how do we grade how do we learn and it's going to require AI and humans to human using AI tools to reinforce learning but I don't think it's ever going to be a lecture format. Yeah I think you're going to use humans to tutor you when you don't know the answer I think I think you're right and I think modes of thinking will will change I'm lateral thinking or being able to think across disciplines such as thinking like you know having compassion speaking in a way that is showing empathy empathy you. And then also connecting it to knowledge and then having a agent there for making sure the clinical aspects are correct super passing discussion Scott I think something like agents we're going to be talking about over and over again so I encourage our listeners in the academic circles to please reach out to us we'd love to have you know fellow deans and administrators and professors.
Join us so please. Yeah and faculty really want to get in and learn this and be the conduit of AI information for the student body I mean I know not every professor not every dean and not every president's going to care and that's okay you know but there are members within the organizations when each academic institution you know. You know that that want to take on this charge that just need help and resources we're here to help if we can. Exactly right so please reach out to us i'm Rayhan at AI and I care dot AI and Scott Scott with one T at AI and I care dot AI all our material is again they'll go at AI and I care dot com and Scott great conversation that look forward to speaking with you next week sounds good Rayhan everybody have a good day.
You've been listening to real talk an AI and I care you're weekly podcast to keep you informed about AI technologies revolutionizing I care.