Think AI Podcast - Healthcare AI: The Hype Ends Here | Ep. 17 with James Green (Cognome)

Episode Date: August 11, 2026

🎙️ Healthcare AI: The Hype Ends HereMost hospitals cannot answer one basic question: what AI is running on our network right now? James Green went from Pixar under Steve Jobs, to running the World Series of Video Games, to building the tech behind the NFL first down line, to CEO of Cognome, a healthcare AI company deployed in 20 hospitals including Montefiore and NYU Langone. In this episode, we strip the hype out of healthcare AI and talk about what actually works, what is genuinely dangerous, and the unintended consequence nobody is discussing: a billing system where nurses are expenses and AI is profit.In this episode: 00:00 From Pixar and esports to healthcare AI 01:46 Why healthcare is adopting AI 2.2x faster than other industries 12:15 AI Sniffer: finding every hidden AI on a hospital network 18:08 Ambient dictation and the wrong leg problem 21:55 The judge and jury system that kills hallucinations 26:17 Why missing context is the second most dangerous thing in clinical AI29:36 Silent mode: the right way to deploy AI with physicians 34:44 AI already beats doctors 99.9% of the time, with one catch39:42 Nurses billed like bed linen: the consequence nobody sees coming 43:34 LLMs are now healthcare's fastest growing cyber threat 46:52 The next 3 to 5 years: thousands of AI agents per hospital 49:32 What Steve Jobs was really like: seven reasons in 2.5 secondsIf this conversation made you think differently about AI in your industry, subscribe to the Think AI Podcast, drop a comment with your take on the nurse billing question, and share this with someone in healthcare who needs to hear it.🔗 Links & ResourcesCognome: https://www.cognome.comJames Green on LinkedIn: https://www.linkedin.com/in/jamesangreen/#ThinkAIPodcast #DaveGoyal #HealthcareAI #AIGovernance #ArtificialIntelligence

AI is better, like 99.9% of the time. It's actually already better than the human. But the problem is it doesn't know what it doesn't know. Welcome to the Think AI podcast. Each week, we talk about the most exciting AI research, tools, case studies and more. I'm your host, Dave Goir, and I've been working behind the scene in data and AI for over 30 years, whether you are an AI expert, skeptic, or something in between, this podcast is for you.

So today I'm sitting down with James Green. He has one of the most interesting career path I've ever talked about on this show. He started at Pixar Animation Studios working under Steve Jobs. He ran the world series of video games. And yes, that's a real thing. And he was the CEO of it. He built the company behind the first downline you see on every NFL broadcast. And then he sold a hundred million marketing technology company to Deloitte.

So naturally, he became the CEO of a healthcare AI company, isn't it? When you say it like that, it sounds totally insane. But yes, that is, I cannot. You're right, Your Honor, I plead guilty. So that's James. He runs Cognoma, a company that builds AI solutions used in real hospitals, not research labs, not pilot programs, 20 hospitals running every day.

And they started at the University of Texas in 2010. And today they are at Montefloor. Am I speaking it right? It's Montefloor. Montefiore, right? Yeah. Montefloor. And that's a health system in Albert Einstein's school at medicine.

And then they also had NYU Longon and several other hospitals to name a few year and academic research centers as well. James, welcome to the show. The key reason I wanted to talk to you is I'm tired of healthcare AI hype. People just talk about we are transforming healthcare using AI. And there is a excuse my language. There's a lot of bullshit around it. So here I am sitting down with you.

What do you have to say before we get started on it? So I'm going to say it's not all hype. And here's why. Okay. So healthcare has traditionally lagged from a technology point of view. most of the industry. So if you looked at health care five years ago, 10 years ago, and you compared it to any large hospital, I'm going to say, when I say health care, just to differentiate it from, you know, life sciences companies or vendors building tech,

but any large hospital system to a large retailer or someone like that, they tended to lag in their technology. And what people didn't understand was there are institutional reasons why that lagged. happened. It's regulated. There's your personal health data. There's all kinds of other personal information you have in there. Everything tended to need to run locally. So they were very late to the cloud. So healthcare in itself has been a laggard. But with AI, which can be run locally and which can sidestep a lot of the things that have been holding healthcare back, healthcare is making a little bit of a leap. Now, I don't want to say it's the market leader, but it is adopting AI.

I think the number is 2.2 times faster than the average institution. So that's a real number. You can fact check it. Right. So it is adopting AI faster than most industries. But it's a little bit like, you know, you're starting a race from behind. So you're adopting it because you're catching up, which I think is where you feel maybe the bullshit. you use. But it is, it's having an unbelievable impact. The other thing that I think people don't appreciate about healthcare is that AI is broken down into two types, broadly speaking, right?

Machine learning models and then large language models, right? And today we say AI and it's like people don't really think about the difference. They think that they mean LLMs, but a lot of the AI we're experiencing is actually machine learning systems or some hybrid. And hospital systems have been rolling out machine learning systems in the hospitals for a decade. You know, we built a sepsis model at the Montefiore health system, I don't know, five years ago, using machine learning techniques five years ago before big LLMs were big, right. And so I think that also gets confused in the hype. People say, you know, there's a lot of machine, a lot of AI in healthcare and they don't separate out those two different types, both of which are pushing to the forefront. Well, that's very true. And I can relate and factually confirm most of it because I started in AI when it was not called AI, right? The machine learning thing that you're talking about in 28-30. years. Don't hold me to that. I'm dyslexic, so my numbers get fuzzy. Oh, I'm dyslexic too,

so all our numbers will be messed up. Perfect. That's very good. I do have a product for young kids called readon.a. with one of my business partner in Canada and dyslexia is a big deal. You know, any of these things where I'm just drifting away a little bit, but any of the things where kids can be helped, I loved it. And I struggled a lot. lot growing up, while they did good education, I think AI is very well suited for those kind of problems today. And so we build a product called read on that AI to educate everyone in the world. And we've been very successful doing so. But I want to come back to this. In a healthcare, I spent a lot of years with a large manufacturing medical devices company, which is again categorized

as a health care company. And a lot of compliance issue, a lot of security. security issues and a lot of other testing issues in terms of innovation can be directly solved with AI, both with the legacy machine learning models as well as the new age Gen AI LLM type of models as well. What still intrigued me, and that's the elephant in the room, and I'm going to jump onto it right away. So you went from Pixar under Steve Jobs to the esports industry to the sports, or the sports broadcast. cast technology to marketing tech to help carry it. Help me through this. And I've done these kind of crazy things too. But I want to know from your perspective.

Yeah, you pointed that out. Honestly, I haven't really thought. You're pointing it out. And not a lot of people have called me out on it. And so it took a little thinking about why that happened. And in the end, you know, I don't really have a good reason. And I have a very good reason. You know, there's not a connection between the two.

Like, I mean, that's why it's funny. right, there's not a connection between these things. But the connection that does exist is I am fascinated and drawn to things that are changing in our world, right? So if you look at Pixar, Pixar was revolutionizing the film business with realistic like animation that had never been done before. Disney had been the only company that had ever been able to make any money in animation. up until Pixar, and then Pixar came out of the blocks with Toy Story and changed the world. So that was like amazing, right? And then you talk about e-sports.

We've been watching sports all our lives, and suddenly people are competing on video games and watching it. That was a revolution. Now, you know, there's Twitch and all of these things that are out there, but I'm always drawn to these inflection points in society where, you know, interesting things are changing. And then you go to football and PBI. I'm very proud of that company. It did super well.

It was owned by ESPN now. And I wasn't there at the very founding of it. I ran it when it was a public company. I came in and ran it. But what the engineers did and a bunch of them came from Intel, inventing a way to make that game more accessible, I'm going to guess that more women watch American football because of that first downline technology than anything else

that ever happened in the NFL. So that's huge. If you look at television, well, I'll just use the year 2000 as an arbitrary cut, but if you look at TV before 2000 and TV post-2000, live TV,

the number of artifacts, artificial artifacts that have gone into television post this century has been unbelievable. You look at TV sports broadcast from the last century and they look really empty and barren. And so that was another huge cultural change.

And then you talk about the ad tech company, that essentially people would, this was when people first were going into the internet and then they'd look at a pair of red shoes and those red shoes would follow them around the internet everywhere they want. You remember that. That was another like, oh my God, how are they doing that moment? Which is essentially kind of what we were powering at the time, searchery targeting, right? And so all of these things are inflection points in society.

And so why am I at Cognom? there's, I'm not leading the charge with a frontier AI model. That's definitely changing society. I'm not doing that. But it is, healthcare itself is going through an enormous revolutionary change. I'm going to say that the industry that will change more than any other industry in the country over these next five years will be healthcare. And so I got to be in it.

I got to do something in it. So those are the things that caused me to be in these places because I'm looking at what's going on in society and what changes and what's interesting and what, you know, you want to lead an interesting life and have an interesting impact and be part of what's going on at the frontier, or at least I do. And that's the connective tissue behind these things. Yeah, and I'm so glad I got connected with you in all seriousness. I mean, the fun is the reason why it was funny for me.

It's an incredible journey. Everybody asked me the similar thing. So in my 30 years of career, I did the similar thing, build nine different companies, different types of companies. So you're cheating. You knew. You knew. You knew.

So I did build. I worked with a friend on building a, like an event management company. I went into media productions also. So I did healthcare work. Then I also came back to software engineering part of it. I studied as a CFA, financial analyst. So different profiles, right? But I always look for inflection points.

So 1996, 98, the inflection point was because of why took it, everything changing and going towards finance. So I was there. But then I realized that's not my cup of tea. It gives me heart attack when I lose money in portfolio. So even though I am very good in managing stuff. our portfolios and predicting that, that's not who I am, right? So finding the correct inflection point for you is what matters the most. I'm an innovator, an inventor, a founder, and that's what I love. I'm also a motivated, a leadership person. So I like to do those

kind of things. And I think that's the balance you found out for yourself pretty soon than what I did. I build nine companies and five of them were miserable failure. But I guess you were able to find out much quicker than I did. I was too slow in understanding it. So I commend you on what you did and what your journey was. And that's the testament to you. Thank you. I appreciate that. So let's get started on some of these other things. So let's talk about Cognom and what it actually does. Give me some specific examples, especially what they do in the hospital and who uses the software. And what decision does it change? Yeah. So, so, so, Cognom was built over 15 years now,

maybe it's even approaching 20, over multiple institutions. It's a huge tech stack. And at the bottom of the tech stack, we're doing data transforms and understanding how data works. And what that has caused us to do is have products that sit on top of that.

And really, the thing that is, so you have this big tech stack on top of that, we have built models that we think that's a commoditized space. We don't do that much anymore. then we have the ability, we have a tool called Explainer AI, which shows you how the model is performing and if it's exhibiting bias or any of that. But honestly, the thing that is getting, that is like white hot at the moment for us that people are calling us saying, you know,

we need to deploy this now, now, let's go, go, go, is we call it AI sniffer. And so, you know, and to understand it, you need to understand, this is true across all industries, but we're focusing on healthcare. So you need to understand health care. So first of all, healthcare is huge, right? It's about, I think, 18% of GDP in the country. It's a huge, huge industry in this country. And your average hospital probably employs a thousand people and that understates it because there's a bunch of smaller ones and then fewer larger ones. So really a hospital is going to more likely be employed like thousands, if not tens of thousands and sometimes even hundreds of thousands of people. They're multi-billion-dollar companies.

A lot of these hospital system, Montefiorey hospital system in the Bronx, which is like one of five large hospital systems in New York, is a $5 billion system. And it's the smallest of the five, right? The other five are even bigger. And that's just New York, right? So they're big companies, right? 18% of GDP, big companies.

When you have a big company, you have a hard time controlling what all your people do. And so what happens is some of these companies, they don't know what AI has rolled out in their system. They don't know. Has a researcher downloaded a copy of Quinn and are they running it on their laptop? Did they get approval for it? Did their vendor that sells nursing, scheduling software,

upgrade its software to have AI embedded in it? And is it now running on their network without them knowing? and for a healthcare system where AI is making decisions and can be a source of vulnerabilities, understanding what is on your network is like, that's like square zero. You can't manage it if you don't know what it is, right? Otherwise, you're managing a bunch of things

and the thing that you need to manage it aren't being managed. You don't even know they exist. So our AI sniffer technology goes out there and tells a hospital, even if they have no idea who's rolled it out, it goes and says, here's everything on your network that is using AI. I don't care if it's a vendor. I don't care if it's a scientist.

I don't care if it's someone using their phone to go out on the network to Claude or chat GPT or whatever. We're going to show you every single person who's using AI on your network. And once you've identified that, then you can move to managing it. And it's a very important part of AI governance. I mean, if you think of it as a patient, do you want to know that every piece of AI that could affect your healthcare is under tight control? That there's a policy around whether it should be giving advice or whether it should be making a decision and that everyone has thought about it deeply and it's going to result in the best outcome for you as a patient. And so that product, that AI sniffer product for us at this particular moment in time,

which is July 26, is flying off the shelves. It's like crazy. No, that's so true. And compliance is one thing, you know, to have a cliche thinking on it. But at the same time, that has some meaning. And the meaning is safety of the information of a person and even of their own. life because using their information, if you can track them on everything, let alone have a financial fraud on them, it can danger their life, their family life and so forth. So it's not just about

saving the private information. It has more repercussions than we can even imagine. And these are the things that caused healthcare to be behind in the first place. So these these structural issues that you just outlined caused healthcare not to adopt some of the tech that drove other companies in the first place. But AI is making them catch up. And now that they're catching up, they need to control it. And so we're part of that. As they say, a rising tide floats all boats.

We're part of that rising tide. Oh, nice praise. Loved it. Yeah. And so that prompts me to the next question also, which is, again, the continuation of what we are saying. So everyone excited about AI in healthcare.

Some are really finicky about it. Some are really scared about it. But most of us having a formal, if you will, right, want to have an AI in healthcare, but due to over-regulation, over-compliance, and I will use that word, there is not a good balance out there also. How do you really see the value?

you have done it for 10 years now, where it is adding value or where it is dangerous? I think in most cases, it's both at the same time. I'm going to give you the case, and I think this is widely accepted, where it initially added the most value, and then why that's incredibly dangerous at exactly the same time.

And I think that's often true in life, by the way. You know, like, you know, you were in the stock world for a little while, you know, you're going to get highest return with the highest risk. It's very similar kind of thing, you know, you're going to get the biggest improvement where you can also cause the greatest harm. It's, you know, you have to control it. So the first thing, and this is not at all revolutionary that really made a huge impact in healthcare was ambient dictation, just the ability for AI to listen to a doctor as they're going through their conversation with the patient and then update the chart and have that done.

And there's a couple of, you know, a bridge and some other people were sort of really leading that charge. And now most EHRs like Epic is the big one here are sort of making that a native part of the EHR itself. And that's hugely time savings. There's a one of the problems in health care is that the documentation part of a clinician's work is a huge percentage of their overall time. And you want them to provide care to you, not documenting the kind of care. So being able to automate the documenting of that care is great. However, you know, AI does silly things. So one of the things that's just going to probably make the hair on the back of your neck stand up is, you know, you know, you.

you could be there with a patient and you could be saying your right leg and you touch the leg and you're like, everyone knows what the right leg is. And maybe that's for a piece of surgery. And in the thing, it says the right leg. But actually, that's not how you, for a surgeon, describe a leg, you know, because what's right, is the right when I'm looking at you or is it right, you're right? Like, which is the right leg? And that can result in surgery on the wrong leg, you know?

And that is a hugely risky and consequential outcome. There are the right ways to do that. And so here you have this thing that comes in that just seems fantastic. But even something as mundane as the conversation that was happening with nothing wrong being said can result in errors in the chart. So then you have to have something that goes through that and normalizes it highlights maybe for a nurse or for another AI. saying, you know, if they're saying this, this needs to be updated in the chart so that it's recorded properly. And so these kinds of things are highly risky and from, and yet they bring big rewards. So there's an example. I have more.

That's a really good example. And that also prompts me in that same continuation is that how do you really handle the hallucination in clinical settings? I mean, I know, I know. I know. know, there are standard operating procedures, there are protocols, there are how-to guides, there are a bunch of other things which is already there. You can restrict AI too much and then it will only spit out what you're saying. So it will become apparent in a manner, right? And maybe in some settings, that's perfect. But in some settings, you also want some decision making some insights generation, some, you know, course correction as it goes along. And clinical setting requires, a whole lot like you're saying. So how are you handling that? So it's a good question because,

I mean, also it's great because we have a product in this space called Explainer AI. So I'm going to, I'm going to give you that as a disclaimer. You know, we have a product called Explanor AI. Helps with product hallucination. I'm now going to talk about it generally rather than specifically about a product because I know this isn't about me selling Cognom. But the, the, the, the, The way that, and we have this to technique, but most people handle technique, the techniques that we handle to prevent hallucination in any setting, not just healthcare, not just healthcare, is really twofold. One is you have another AI system that looks at the system that's making the decision. And instead of asking the same question, you know, does this patient have sepsis or does

should, you know, but instead of asking the same question, you, you fuel it with a prompt that says, does this AI have enough information to accurately make this decision? Right. So it's a totally different question, right? Is there enough information there to make that decision? And then, you know, in the cases where the AI doesn't have enough information, you know, to make the decision, but it's making a decision. Then you can route that through to another system. It could be another AI system. It could be a human. And you can then have a training

mechanism so that, you know, but that's one main system that you use to do it. And then the second is just pure sort of statistics. You actually create a sort of judge and jury system. And if you have a, I'm going to use round numbers. I hope no one has a 10% hallucination rate, but if I say 10% I can do the math quickly. So I hope all hallucination rates are, you know, at 1% or less, but for this example, we're going to say 10%. You can have multiple systems that are running on that. So you can, you know, the first system has a 10% error rate. You can run it through a second system. That'll have a 10% error rate, which will take it down to 1%. You can have it sort of through through third system, that'll take it down to 10.

So there's a sequencing of decision making that you're going to have. And the combination of those two strategies can really help make sure that there's nothing going wrong. What won't work is what most consumers have, which is, hey, Claude, hey, ChachypT, hey, hey, Grock, what do you think about this? And it gives you one answer, that's not going to work. You know, you have to have these, you have to have really, It's about agenic design, it's overused word,

but you have to have these agents sitting around doing these extra work to make sure there's no hallucinations. Yeah, and that's perfectly said. And these words exist for a reason, and that word might change as big AI companies might change it to something else, but the concept's still the same. If you are creating a human-like resource with AI, you have to consider all the human-like traits of it.

They can forget, you know, they can make judgment issues. And how do you handle that, like you're saying? And second thing, I come from finance background, like we are saying, the education is there. The checks and balances, right? You never validate your balance sheet in one way. There are always two methods to do that.

You know, credit comes in, there it goes out. So you have to figure that out and then balance it out like that. And AI is no different. And hallucination is there. Because if you ask someone, what do you want to eat without knowing whether he's a vegetarian, non-vegetarian, if he has a dietary issues, it's going to give one blanket answer. Pizza is the most famous thing in U.S. or in a particular region, let's say New York, you get amazing pizza, which I love there. So it's going to just say

eat pizza, and he may be a lactose intolerant person, and he's not going to be able to do it. You can't say that's a hallucination. It's how you ask something. But you can say that New York has the best pizza just for the rest. I love that. I would say it out like, and I would actually take some notes, you know, where to go when I'm there in New York. But you, I was going to mention this earlier when I talked about Ambient AI. But context is actually the second most dangerous thing. So I don't know if you, for anyone who's watching this who's seen the television show The Pit, which goes through an ER room and sort of, you know, it's one day.

I think there's only been two seasons, so there's only been two days, but they're very dramatic days. And it's a great example to me of why context matters. So if you look at the systems and what they're capturing, they're capturing what someone actually wrote in the chart. But what happens in that show all the time is things happen outside anywhere that's documented. But people generally know, like person A told person B or someone saw something, happen and so they're reacting to it, but it hasn't yet been documented. And so this is the other huge risk about AI is it just simply doesn't have context. So it's going to look at everything that's written down and that's its entire universe. And it doesn't know that the patient

has some other thing that wasn't documented yet. Sometimes that can be as simple as progress notes in hospitals have to go in at the end of the day. If you're asking the AI to make a decision before the progress note gets added. It's got to make a mistake. So context is a huge problem for a system that doesn't necessarily have all of the context. So true. And along with the context,

you also need to have short time and long-term memory so that the combination will give you a human-like insights with the decision-making with your own data in that setting. Correct, correct, correct. Yeah, the AI doesn't have a memory that works like our memory. You know, we, everyone has faults with the human memory, but it works fundamentally differently. So we can remember things from very, very disconnected things and recognize patterns and store them over long periods of times over various different context windows. That is not how a large language model works, you know.

True. At all. So the two complement one another when the systems are working well together. This is great discussion and I'm loving it. And especially in the healthcare setting where people are generally scared about AI. I actually worked with or did a course with Harvard on health care in AI. And a lot of doctors were doing it. And this next question is because of that.

So they are asked about, you know, can I use this AI? It's sort of seem worthless. And that drives the thinking, if you're building an AI product, what's the real metric? And to me, that's adoption by doctors. And how are you solving it? What do you see in terms of adoption from the doctors? What complaints you see?

What the things they ask for which you build? What's the pattern looks like? So when we've been in the model building business, which we historically have been, and now we're not doing so much, we're providing sort of data systems for other people to build models and then monitoring those models. But when we were more in the model building business, the number one thing that we did when we built a model was we wouldn't build any unless there was overwhelming demand and a clinician. who was championing. So if it wasn't getting pulled by a desire of the frontline staff to do it, we wouldn't build it. Now, that's not how startups work. Startups aren't in those contexts.

We happen to be born inside a health system. So we had this massive unfair advantage. You know, we would talk to, we would talk to clinicians who would be like, I have this problem. I have, you know, I'm trying to predict whether when someone has cancer surgery, what sorts of things goes wrong. It's a real model that we built. But, you know, we're doing that with the clinician. And then when we build the model, we roll it out in silent mode right alongside what they're doing regularly. And unless the silent mode produces better results, it never actually gets deployed into

production. So that's the right way in a healthcare setting to roll out a model or an AI solution. Doesn't always happen, but that's the right way to do it. You said it well, and that's the perfect way to do it. And I have some parallel story there. So along with healthcare, we also work with manufacturing and some of the other industries. I have their dashboards that management love. We work with CXOs only, but nobody used. And the difference was that workflow integration wasn't there.

If the insight doesn't show up where the decision is made, You know, if a line level manager or a director has a different problem, management wants certain high level numbers, the KPS, that's great. And they should have it by every means. But can it help to lower level decision makers, supervisors, managers on the floor? If it doesn't help them, it's a waste of effort. And, you know, they will be a yes-man in the group meeting or a board meeting, but outside of it, it's not in use.

And same thing here, what you're saying. in healthcare, if it affects them, if it affects their decision making, if it gives them inside while they do their job, it'll be worthwhile, right? Yeah, yeah, yeah. I'll give you a sort of a concrete example of that. We built a sepsis model at Montefure. And at the beginning, it was still super accurate. Nothing changed in the accuracy, but it flagged whether or not a patient have sexist.

And I think it was at the time, it may still be, I don't know if it's true, but at the time it was the most accurate model in the country or maybe in the world. So the results were great. But sometimes the model was wrong because it didn't have context. So, for example, in sepsis, you can have a high white blood cell count, but you can also have that if you have leukemia. Right. So what we added to the model in the workflow was the ability when the patient was there

to click one button and see why the model made the decision. What were the variables that the model was looking? looking at to make the decision so that the clinician could kind of have a conversation with it and be like, oh, mate, I'll use the same example, that white blood cell count is, I know they have leukemia, so actually they don't have sepsis, you know, thanks for the advice, but no thanks. But without that kind of building it into the workflow and giving the transparency of the decision making, then the clinicians would have found it very, very hard to use because sometimes they'd be like, it's wrong, but I don't know why, you know.

you don't have to have that happen very long before they just start ignoring it, and then you're not being helpful at all. Yeah. And this is a great example for if you really compare with human wisdom, so people, doctors especially, who have worked in the industry for a very long time, they've seen it all. And they can really quickly assess it as an example. A heart burn is not a heart attack, and they have a pretty clear ways of identifying it and so many other things, right? But that comes through their wisdom, obviously through their education as well. But if AI can harvest that knowledge to wreck, I didn't want it to throw any fancy terms there,

but if it can harvest all that knowledge into a system where AI, through a chairboard workflow, can help. I don't want to call it a junior physician or a junior clinician, but you know, who may not have the life experience as much as the others, but they can use it. Same set of works in legal and paralegal, same set of works in CPA and other things. And this is where AI is becoming so, so, so powerful if you use it in the right setup.

If you want AI to write your email, yeah, go ahead and do it. But that's not where you get the most value out of it. These are the places where you get the most value out of it. Yeah, yeah, yeah. And just to be provocative in a case where the same information is present, so a human and an AI, same information available in both. cases, AI is already better than the human at making these diagnoses. Well, the difference is that the human may have other contexts and other knowledge that the

AI doesn't have. That's the difference. Otherwise, the AI is better, like 99.9% of the time, it's actually already better than the human. But the problem is it doesn't know what it doesn't know. And so, you know, that causes problems still. So you can see where the world is going, though. It's going to be a huge part of healthcare for sure.

Yeah. And that's where I want to talk to you on the next, but I want to make a quick comment on what you just said. I've been saying I did data and data intelligence and architecture work for 30 years. I wrote a book also called real-time business intelligence mastery. And I keep saying that data before AI, AI is useless if you don't have the right data. Now, in certain areas, because it has learned, digest the internet, it can give you certain. answers with a lot of hallucination because, again, it doesn't have the right context of what you're asking. But then the second part, what you just mentioned, you know, if you have trained everything from a bunch of physician, but there are still things that they know which AI doesn't know. But with time, it will start catching up more and more and more and more and more. And eventually it will become 98, 99, maybe even 100% better than that. I'm going to push back on that. I don't know. Like I, you know, again, you know, I have this image of someone coming into an ER, right, where you see the person.

When you look at someone, when you're looking at me, you see that I have glasses. You see that I have, you know, thinning hair. Seeing that I'm wearing a shirt that says, like I do. You see, you see immediately in seconds, you see all these things. You know, AI, unless you tell it, doesn't. see these things, sure, you could have a camera, it could be capturing it, but in general, that gets way too expensive to have a video camera capturing all of this stuff. It's just, it's prohibitively expensive. And so I think, I think it will always be, or at least in my

lifetime, perhaps because I only don't have another 50 years to live, you know, have less. it'll always be some combination of a person who can absorb all of the things that are going on and interpret what the AI, at least in healthcare, interpret what the AI is saying, because the AI is already better if it has the same information. I just doubt that it'll ever in healthcare have everything that someone who's looking at the patient and talking to the patient and seeing everything holistically, it's super hard. You know, it's very easy for it to take your vitals

and, you know, predict something out of that, but it's still not seeing your whole self in a way that a person can see your whole self. And I think and hope that at least in healthcare, it'll always be a mixture of those two things and the AI will be relieving people from the things that people don't like to do and leaving people in the roles

where they do want to do, what's the doctor want to do? They want to provide health care. They don't want to be sitting in front of an EHR going, right? So I hope AI is helping with that. I hope it will continue. But I'm not so convinced

that it will continue to just replace the doctor. Yeah. And I agree with you. I wasn't meaning to say that and I appreciate that you are mentioning that. I don't think in any form, in any business, like music, same example,

there are patterns. I love music, like to compose music. And there are patterns there, but there are certain things that hit your emotional, right? And great musicians understand that. When they go and perform and they, you know, they change how they present and perform in the live than what they record because they don't have people in front of them.

And when they see people, the reflection is different. AI won't be able to produce that, at least in our lifetime, like we are saying. Or if it can, it will require a lot more training. But my point initially was that AI can get better and better and better if you provide enough data and things like that. But by no means AI should ever replace human. I mean, that will be something crazy like what we see in movies, and we don't want that. Well, you know, let's just be provocative again. and I've been trying to write a blog post about this for the last 24, 48 hours,

and I haven't written it yet because I haven't had time, but I'm fascinated by it. This is about unintended consequences. So healthcare in the United States, this is very United States-centric because we have a for-profit healthcare system, so it doesn't really exist to the same level in Canada or England, and places where I have some other experience, and anywhere where there's a national health provider.

But in the United States, you have this tension between a provider of healthcare who then has to use something called CPT codes to go and translate that into a bill, which typically an insurance provider then pays. That's a very US-centric model. The CPT codes, there's a lot of procedures and a lot of things that are there for CPT codes.

And now there are some AI procedures that are getting CPT codes. And the AMA is just debating at the moment whether to allow AI procedures to have codes that you can bill for. What I didn't know, and it still blows my mind, obviously they're doctors that have things with CPT codes, but nurses don't.

There are no CPT codes, no billing for a nurse, which I didn't really appreciate. And for those of us that have had any kind of healthcare, nurses are a big part of that healthcare. I'm going to just go out on a limb and say, nurses are more important than AI. Like, you know, if you don't have nurses in a hospital, like, God help you. But what I am concerned about is what I'm trying to say in the blog post and I'm being a little

long-winded because I haven't got it yet. I haven't figured out how to say it succinctly. But if we go into a world where nurses are expenses and AI are profits, I worry about unintended consequences in that universe. And that's a big, like, what? True. What? I started thinking about it already. Like, yeah, how would that shit double in the whole context?

Yeah. You know, nurses, I'm going to be. provocative again. Nurses are charged the same way that your bed linen are charged. It's just wrapped into your room charge. I think nurses have less in common with bed linen than AI does. I put more AI with the bed linen than I would nurses. But that's not how we're thinking about it. So there's a lot of things that could be explosive and we could get very wrong in AI with very unknown consequences. I worry about what's going to happen when charge of. like that are not allowed for humans, but are allowed for artificial intelligence.

I haven't understood all of the consequences of that down the road, but my gut says that's a problem, and we should not be doing that. And you know, that definitely brings or leads to a point that, you know, a lot of people talk about guardrails. Then they also talk about ethical AI. What you're saying does fall into ethical AI, but like you're saying, being provocative, I think we need to have a different manual altogether, not just a think of that's a different thinking. Here, what we are saying is what roles and rules AI can apply and not, or should not apply,

not that it cannot apply, it should not apply, so that we can still harvest the benefits of AI and keeping their correct roles alive in the manner they need to be alive, right? You're 100% correct. And again, just. to show you an example of how correct you are. Historically, when you, so in healthcare, healthcare has a higher instance of cybersecurity incidents than any other industry. It just does.

And historically, the number one way in has been sort of social engineering. You know, you get in through someone, you find a password, and you then find an exploit and you go in. the number one growing way in today is LLMs. I think 20% of cybersecurity instances were using LLMs last year, and that's just going to go through the roof. And if you think about, as you said,

you're talking about the guardrails around that. If people were the main risk in the past, and you're putting a lot more into AI, and AI is now going to be the largest risk of cybersecurity threats, you know, that's, you better, you better know what you've got, what AI you've got running on your network and you better be monitoring it, which is why, you know, we're in such a hot space at the moment. But, you know, just think about what unintended consequences could happen there

if you don't have the right guardrails around it. Yeah. And going back to the risk of AI without naming any big companies, here, we all live in this era of AI. If somebody says, oh, I'm out of AI and my data is completely secure, no, because you have a phone which is listening to you and is doing a bunch of things that you don't know.

And yeah, you can disable those things once you're smart enough, you can research on it. But then you lose certain features. So it's like you enable it and take the risk or you just get disconnected from the world. And those are the things to think about and have a lot more regulations, you know, not regulate AI, but the consequences of AI, how it is being used in what settings AI is being used. And that sometimes worry me, but I'm a pro-AI. I'm a pro-life.

I'm pro-you-building something for mankind. I'm a disabled entrepreneur, and I came to this country with the hope that either me or somebody can build braces. So there are, Pyonic legs, who doesn't have a leg, amazing technology. But people who have this muscle at trophies and things, you don't have anything, you know, and you see a lot more areas. And AI is amazing. I have a patent in AI and neuroscience. And AI is amazing in these spaces like dyslexia.

I wish I had back in the days, but now from 6 to 13 years of age using, you know, this gaming methods and gazing and the AI in gazing, we can solve that. and we have solved it 100% for a lot of kids. And that gives me a lot of pride and joy doing so. So going back to that thinking and the futuristic way, where do you see healthcare AI goes next, maybe three years or five years? What do you see it should be doing based on your own prediction

or what you proposed AI to do? I think that's not a very long time, but AI is moving at crazy speeds. So one of the things that happens in technology a little bit is that things fracture and then they consolidate. They fracture and they consolidate. There are these expansion things where they fracture and then they consolidate. And I think in the next three to five years it'll be more in the fracturing. And what I mean by that is that AI, there'll be just so many AI solutions rolled out in very specific areas like your bionic lake.

example, like a surgical example, like at the moment, I think the number of AI solutions in a hospital is measured in the low hundreds on average. Like, it's still a very big number. But in the next three to five years, that'll be thousands of agents and other things in AI splintered into all of these very, very specific things. Somewhere in that three to five year horizon, it'll start to get consolidated and come back, but I don't think that'll happen. in the three to five years. And one of the things, and one of the reasons that's driving that, you know, our solutions, as an example, and I only say this because our clients require it, you know, our solutions all reside local on in the healthcare environment. That could be on-premises. A lot of healthcare still run machines on-prem, or it could be in their private cloud, which is then isolated from anything else, right?

But we don't, most modern software companies, and we're a software company, have some kind of SaaS multi-tenant thing, you know, where you log in and everyone's logging into the same interface and you're all sharing. That's very centralized. That is not the way healthcare works. Health care is very decentralized. And I think that it will just continue to be decentralized and fractured over time before it comes back and becomes centralized. Now, that's a great insight, and I can clearly see what you're talking about. And I'm really hopeful it will play out the way we are thinking about it. And that leads me to my last question. Then I have one more thing to ask. I'm switching gears from AI. We've talked a lot about AI. Like I said, it intrigued me when I heard you worked under Steve Jobs at Pixar.

So what's one thing you saw in that environment? that most people would find surprising. What's unique about that? I assume you really mean about Steve Jobs, and so I'm going to sort of tell you a story about him. I bet there were a lot of amazing things about Pixar, you know, some of the work that Ed Catwell did there, and just some of the technology

and what they were able to do were really kind of incredible. But I assume we're going to talk about Steve Jobs here. And so I'll talk about... Either way, Steve Jobs or Pixar, because if you're... If Steve Jobs is impacting, it will be Pixar. Same thing, right? Yeah.

So the thing that struck me about Steve Jobs, and I think it's an insight into how just clever he was, how he was just a really smart person. I'm not a huge believer that everyone, that money is correlated with intelligence. Like, this has not been my experience. I think I'll be a little rude. I like to say there's an idiot quotient and it's it's independent of wealth. There's a safe number of idiots who are billionaires as there are who are homeless.

Like they're just the same. So from idiot quotient to income quotient, it doesn't correlate as what you're saying. Yeah, it doesn't correlate. It doesn't correlate. There's other things. But Steve Jobs was really smart. And one of the things that happened to me multiple times with him is I would come in and I'd say, you know, I think we should do this. And when that comes in in writing, that's sort of easy to respond to. I think we should. And he might say, no, I don't think we should. And here's a bunch of reasons why.

But what he would do in real time is he would say, like this. I'd say, you know, I think we should do this. And he'll be like, here's seven reasons why you're wrong. Now, just think about that for a second. He's known this information for two and a half seconds, right? and he said, here's seven reasons. Now, I challenge most of you to anyone who's listening to think of seven reasons for anything and hold that in your head and then list them, right?

So in two and a half seconds, he's decided there's seven reasons, and then he'll list them to me. Here's number one, his number two, here's number three, he's number four, he's number five, he's number six, and here's number seven. And that's why you're wrong. And being able to do that with that speed and that accuracy, To me, you know, that talked about the volume of information and his ability to organize thought in his head. And I don't think people talk about that.

They talk about his, you know, a distorted reality zone. All of these are kind of other things about him and about how he interacts with people. I don't think people really talk about how he's just, he was just unbelievably smart. Like, in a way that I couldn't do that, you know, I might, I might be able to do two or three. I couldn't do seven. I just couldn't do it. I'm not that smart. He was just a remarkably

intelligent person. I mean, he had his faults. I'll give you an example of one of his faults. He was convinced when I was there that Pixar should have a different logo and I kept telling him, don't do it. And, you know, the logo for Pixar today is the same as it was back then in the 90s.

He was wrong. But, you know, the entire time I was there he was captured in redesigning the logo, he's got to change. Really never changed. So he is definitely not right about everything all the time. But he was still unbelievably intelligent, had huge horsepower between his ears. Yeah, no, that's an amazing story. And like you said, him being wrong is that he is also a human, isn't it?

Yes. Yeah, yeah, yeah, yeah. So you can afford to be wrong. He was definitely wrong about a bunch of shit for every beautiful, amazing, thoughtful thing you see about him. Yeah, there's a whole mess that he was wrong about. But that, you know, we're human. It's okay.

So, James, that brings me towards the end, and I'll give you an opportunity. But thank you for that conversation. I think our audience got a rare window into why healthcare are you actually going to work. And when you strip away the hype, you can clearly see what's behind it and you are making it in action through Kogna.

And where can they find you? And what last, message you and I leave it to the audience, especially you're your own users like physicians, clinicians, etc. Yeah, I mean, I don't know how many people who are working in healthcare are listening to this podcast. If you are in healthcare and you are listening to this podcast and you don't know what AI is running in your network, you should go to www.com and send us a message. We can help. We are native to your world. And even if you don't want what we have, we can help you learn about how you can achieve what you need to achieve. That would be the message

that I would give to people who work in healthcare, who are listening to this podcast, both of you. Thank you, James. And I'm Dave Goyle with Think AI. And if this conversation made you think differently about AI in your own industry, whether healthcare, manufacturing, whatever it is. Follow me on LinkedIn, subscribe to this podcast, Think AI podcast, and always share this with someone who needs to hear it. That's a wrap. Thank you. Thank you, Dave. You have been listening to Think Yeah, podcast with Dave. Take one idea from this episode and turn it into action.