In the first episode of our three-part “AI for Healthcare Marketers” series, we break down the basics of AI so healthcare marketers can better understand how to use it effectively. Jenny discusses how AI has rapidly evolved—ChatGPT is just two years old, with 56% of adults ages 18 to 24 using it according to YouGov and Reuters Institute—and explains the types of AI, including Narrow AI (used today), General AI, and Superintelligent AI. She also covers how AI learns through methods like supervised, unsupervised, and reinforcement learning, and highlights its key capabilities, such as natural language processing and computer vision.
The episode introduces the most used platforms like ChatGPT, Gemini, Copilot, Perplexity, and Claude, breaking down what makes each unique. By the end, you’ll have a clearer understanding of the leading AI tools and how they can be utilized.
Connect with Jenny:
•Email: jenny@hedyandhopp.com
•LinkedIn: https://www.linkedin.com/in/jennybristow/
If you enjoyed this episode we’d love to hear your feedback! Please consider leaving us a review on your preferred listening platform and sharing it with others.
Jenny: [00:00:00] Hi friends. Welcome to today’s episode of We Are, Marketing Happy, a healthcare marketing podcast. My name is Jenny Bristow. I am the host of this fabulous podcast and I’m also the CEO and founder at Hedy & Hopp. We’re a full-service 100 percent healthcare marketing agency located in the Midwest and we specialize in working with payors and providers across the country.
Today I am very excited to kick off the first episode of a three-part series about AI and healthcare marketing. In this three-part series, we’re going to break down a topic that a lot of marketers really find overwhelming. Whenever you’re learning a new technology, that is such a seismic shift from how you work in your normal day-to-day you typically have a couple of different approaches.
You have early adopters, people who love learning new things, making themselves feel vulnerable, and shaking up their day-to-day. And then you have the people that, you know, adopt things in a [00:01:00] standard time frame. They’re not going to be an early adopter.
They’re going to wait and see which tools really shake out, which processes shake out, and how the industry responds. And then, of course, you have the laggards, the people who really don’t want to adopt new technology. I’m hoping this three-part series will help demystify AI and help give you, and hopefully, your team, if you choose to share this podcast series with them, a much better understanding of AI as it stands today.
So late 2024 going into 2025. And really understanding how can we begin using AI effectively within our internal marketing processes today? What does that look like? What do we need to put into place in order to make sure we’re doing it effectively? And you know, where do we start? So our three-part series, episode one is going to help you understand the primary AI platforms right now.
So I’m recording this in mid-December 2024. We’re going to talk about their differentiators and [00:02:00] examples of different tasks where they shine. Episode two is going to be talking about practical ways to really develop some framework around how you roll AI out to your internal marketing team. So we’re going to call it the six healthcare marketing AI tenents.
And we’ll provide some guidance around how you can work with your own legal and compliance teams, and be able to set up some frameworks so you can use them in an approved way within your organization. And the third and final episode is going to be real-world examples. I’m going to have you listen to this podcast, and have your laptop or computer up.
Pause me. Do the prompt on your other window, and then hopefully you’ll feel much more comfortable understanding what you’re looking at and ways that you can incorporate it in a meaningful way into your day-to-day workflows. So with that, let’s dig in. I think one of the most interesting things when you think about AI, I mean, ChatGPT really became available to the public in 2022.
So I’m filming this in 2024. It’s just two years [00:03:00] old. I mean, it is not that old. That’s basically the same amount of time that we as marketers have been worrying about all of the HIPAA guidelines changing. So a lot has been happening in our world. So first of all, take a deep breath and just say, to yourself, it’s okay if you’re not on the bleeding edge of AI implementation, you’ve had a lot of other stuff to worry about, but I do want you to think about, you know, the late nineties, early two thousands, you know, even if you like me, I wasn’t in the business world yet at that time, but I very clearly remember the shift to people being comfortable with using the internet and using email in the business world.
If you made that shift. You were successful. If you didn’t make that shift, it really impacted your career trajectory opportunities. And that’s what I am going to do. So a great quote is by Karim Lakhani, a Harvard professor. It is “AI won’t replace humans—but humans with AI will replace humans without AI.” and that’s very much my belief as well.
I don’t think marketing departments are [00:04:00] going to go away, but I do think we’re going to be expected to move faster and use technology to really expedite our processes and our output. So let’s learn how to do that. Another thing I will say is that in June of 2024, EMARKETER came out with a study, they did a survey to understand how many people are actually using AI in their day-to-day, not just at work, but in general, even for their personal life and 56 percent said of adults, ages 18 to 24, have used ChatGPT, but that number drastically decreases as folks get older.
In fact, if you look at the age range of folks, 45 to 54, only 28 percent of people have ever used ChatGPT. Now, that was about six months ago, so I’m sure that numbers are a little higher at this point, but if you haven’t really dug in yet. That’s okay. Let’s get started. So when I’m learning something new, I like to put a framework around it.
I’d like to understand what am I learning. How am I learning it. So that’s how I’m going to present AI to you [00:05:00] today through the lens of a framework. There is an AI upskilling framework that LinkedIn Learning put out that I’m a really big fan of. They, it’s basically a pyramid shape and each level of the pyramid is a higher level of specialty within this topic.
So the foundational level is understanding what AI is, having some overall literacy around it and understanding what responsible AI looks like. The next level up is actually applying it. This is where you start prompt engineering, start developing a strategy around AI implementation, and really focus on productivity with it.
Above that is where you’re actually perhaps building your own AI models or putting an overlay, a skin over an existing one to be able to build your own interface. Above that, you’re really becoming a specialist at this point. This is where you’re training and maintaining models. You might be, you know, building a machine [00:06:00] learning models, really getting into deep learning and neural networks. So you’re really a technical specialist. And then above that is really where you’re deeply specialized, where you’re at the level of education, where you can even do security and ops specifically around AI. So as we’re thinking about that level of special specialization that you can do with AI, we’re really going to focus on those bottom two levels.
We’re going to focus on understanding and then applying it. So, that’s our goal today. I do want to pause a little bit and give you a general reminder about compliance. I am the queen of compliance. I don’t want to be, but here we are.
I’m always talking about compliance on our podcasts. We’re going to talk about this a lot. As we talk about the actual ways that you can get your hands dirty and start using AI, but a general reminder, unless your organization has set up its own AI ecosystem that you know is private and secure within your own environment, you have to treat everything that you [00:07:00] put into AI as though you’re putting it on a billboard outside of your office.
So for example. Of course, we’re not going to upload patient names, but we’re also not going to upload information like our revenue goals for a service line or our organization’s name. And, you know, what areas we’re focused on for growth for the next year. Assume anything that you put in will be used to train the model, even if, many tools offer this, even if you opt out of allowing it to do that, you still have to make the assumption that unless you have signed a contract with them and you know that it is within a secure ecosystem, somehow the data could be leaked.
So proceed with caution. The other thing that I will say is that Europe is again leading the charge in the world of privacy, and they actually have developed an EU AI act. The estimated rollout is in 2026, and their focus is ensuring that AI is used in a safe and approved manner. And they’re specifically looking at things like [00:08:00] preventing the manipulation of human behavior to circumvent free will and the exploitation of vulnerabilities of a specific group of people.
So fabulous things to put in place. Some things that I know US-based AI organizations have struggled with a little bit because we haven’t put those sorts of guardrails in place. And it’s why we’re going to have a variety of tools to talk about today, because there’s been inner fighting and they break up and start another organization that they think will be more ethical.
But again, Usually, like we saw with GDPR, whatever starts in Europe will eventually make its way over to the U.S. We likely will see some sort of legal framework coming around AI usage in the future, but right now nothing exists. If you go out on the web and you just type in AI for copywriting, there are going to be dozens of tools that come up.
If you do AI for design. Dozens of tools are going to come up, but at the end of the day, most of them are actually powered by just a handful of [00:09:00] models, less than a handful of models. So what we’re going to do today is we’re not going to talk about the 40 cool design apps or this cool app that can help make sure I’m looking at the camera when I’m doing this podcast, because maybe I’m reading off a script, which I’m not.
I always go off-script. Just ask my podcast manager. It drives them crazy. But we’re not focusing on that. I want you to understand the foundation of how AI works because then what you can do is look at any tool that a team member presents to you and you can say, yeah, but what model runs it and then you can understand structurally how it operates.
So I think we’re going to start with understanding the categories of how AI works, learns, et cetera. So the first question you want to ask whenever you’re presented with AI is what kind of AI is it? And there are really three kinds of AI. You have narrow AI, which is also known as weak AI. General AI, also known as strong AI and super [00:10:00] intelligent AI.
The only kind of AI that’s available to the general public today is narrow AI. So any sort of AI you’re interacting with right now, ChatGPT, Siri, or Alexa, all of those are really narrow AI. Strong or general AI is a platform or an AI tool that can really do anything a human can do, like learn new things, solve problems, and understand emotions.
They say this is not yet developed. I believe it likely is created, but it is not yet available for the average consumer. And then the third, which I really hope does not exist yet is super intelligent AI. And this is an AI tool that is really smarter than the smartest human in every single way. So picture a super brain that can solve problems that we can’t even imagine.
That is really what horror movies is made of. So that is a third kind. So all of the AIs we’re going to be talking about today are considered the first category, which is narrow or weak AI. The next question you want to ask when you start thinking about AIs [00:11:00] and how they categorize themselves is how does it learn.
There are again, three ways. But how it learns, you have supervised learning, unsupervised learning, and reinforcement learning. So supervised learning, for example, is when you show, you want them to recognize it, to learn how to recognize a cat. So you show lots of pictures of cats, labeled cats, and then it understands what a cat is.
Unsupervised is when you actually give it lots of photos of different animals, and then. Without telling it what it is, it kind of figures out based on similarities and context clues. Oh, these are all cats and these are all dogs. Reinforcement learning is more like training like a dog. So if it does a trick correctly, it gets a treat.
If it doesn’t, it doesn’t get anything. So it wants to learn in order to be given a treat of some kind. The next question that I always ask when you’re confronted with a new AI tool is what can it do? So you have four different categories of generally what AI [00:12:00] technologies can do. First is expert system.
So this is an AI that’s like an expert in their specific field. So think like a medical AI that can help doctors diagnose diseases. So they can perhaps look at test results, they perhaps can look at an MRI scan, and they can identify certain things. The next, which is what most people think of when they think of the word AI, is Natural Language Processing, or NLP AI.
So this AI understands and talks in human language, so this is a chatbot that you can talk to, think, ChatGPT. Again, that is an NLP AI. Computer vision AI. This is where AI can actually see and understand images and videos. So think like when you’re flying internationally and TSA doesn’t need to see your passport because it just took a picture of your face and it knows exactly who you are.
That is computer vision AI. And then the fourth category is called robotics AI. And that is when an AI [00:13:00] controls robots to do physical tasks. So think like an assembly line in a factory where it’s assembling cars or the Amazon warehouses where it’s actually picking items for shipment. That is robotics AI.
And the last section that we want to talk about whenever we’re thinking about AI is what kind of interface is it. So you have two. You have an out-of-the-box solution, which is what we’re going to talk about today. So you have ChatGPT, Gemini, et cetera, or you can build your own interface using one of those existing models.
So, when we think about all of those cool tools, for example, like that tool that can make my eyes look at the camera, even though I’m looking off to the side. That is an interface somebody built using one of the existing models to power it, and you can do the same thing. So once you get good enough at AI and understand how the different models think you could, for example, create a platform to allow your marketing team to interface with [00:14:00] a model that is segmented off into your own secure ecosystem, and it’s trained on your brand voice, and it is trained on your service line priorities and your marketing goals for 2025 and your team can do Q and A with it and get really specific with it. Really exciting and not really that hard.
So, in order to do that, you have to do a couple of things. You have to create a dev environment, decide which model you’re going to use, set up an API. So you can do that. Create an interface. How do you want to look at it and type with it? And then launch. And of course, you have to think about the security through the lens of of course, HIPAA and whatnot.
So your team would have to help you with that but it is not as technically advanced, you know, or scary as it may seem before you get to know how all these models work. So whenever we are talking about all the AI tools today, we’re really talking about are these couple of core models that have chat [00:15:00] interfaces on the front.
So we’re going to talk about 5 models. We’re going to talk about ChatGPT, Perplexity, Copilot, Gemini, and Claude. So these are the top 5 based off of our team’s usage and trends that we are seeing in the industry. Let’s dig in. The first, one, ChatGPT. This is the tool that when everybody says, do you use AI, this is what they are talking about.
Right now they have a 65 percent market share. It’s pretty astounding. They were the first one in market, they officially launched in November 30th, 2022. So they just celebrated their two-year anniversary, which is really crazy. It was developed by a company called OpenAI. And there’s a couple of specific things around it.
So, you can use the tool without logging in. Which is really nice if you’re wanting to just kind of understand the tool without creating a free account or signing up for a paid account, which you can do. Paid accounts or anywhere from 20 to 30 a user per month. And you also can get a [00:16:00] company account, where then you can create notebooks or the ability to have shared threads with other team members, which can be really helpful.
The next one we’re going to talk about is Gemini. So Gemini is actually a combination of two tools developed by Google. We have Duet AI and Bard. They lived separately and then they were, it was officially rebranded on February 21st, 2024. So this one has really only been around less than a year and it kind of feels that way.
It feels much younger. When you’re talking to it, it also feels like training guardrails have been put in place that make the answers feel a little bit more generic, but then also biased towards Google. So you may have noticed in recent months that Google is starting to include more Gemini responses in its search results.
So if you do a search, for example, saying like, who’s the best orthopedic surgeon in St. Louis. It will try to answer that with AI above the search results. So when people are thinking about and talking about like AI SEO [00:17:00] or AIO, that’s what they’re talking about is how do we get listed up in that area? It’s been a pretty slow rollout simply because of how long it takes to generate the answers.
And users are very impatient when it comes to searches on Google. So they are only rolling it out for some topics and for some users, but they will eventually have a pretty wide rollout. And then Gemini is also been the source of news headlines because of its apparent bias with its tool, especially when it comes to image generation.
If you think back whenever there was that big controversy around a query, such as give me a picture of a land owner. And it was an older white man, like there’s just some biases and the answers that it provides. And there also are biases in the written answers. So, for example, one of the queries we asked all of the platforms during our comparison and testing was information about if Google Analytics 4 was still safe to use with the new HIPAA guidance. And it wrote back and was like, why yes, you can still [00:18:00] use GA4. Whereas all the other platforms were like, proceed with caution. Here’s some information you need to know about it. So the tool is very clearly designed to keep you within that Google ecosystem.
But there are some built in integrations that make it really easy. For example, it’s already built into some of the tools you use on a daily basis such as Google drive, Gmail, et cetera. I will say my 14-year-old son is a huge Gemini fan and he uses it for a lot of his schoolwork. He creates private notebooks or gems for each of his classes.
And then he has it create study guides for him. So he uploads the documents that teachers give him, asks them to create study guides and to quiz him. So it’s really easy that it’s built-in with the existing infrastructure that many of us are already used to. The next platform is Copilot it’s owned by Microsoft and it runs on OpenAI’s GPT 4 large language model.
So what [00:19:00] again is interesting here is once you start seeing, okay, there’s all these different brands, but wait a minute, Copilot operates on OpenAI’s GPT. So you start to see overlap in the technology. It launched in February of 2023 and it replaced Cortana and Bing chat. It’s limited to only five queries a day without logging in.
So if you do want to kind of understand the tool without creating an account, you can dip your toe in a little bit. And then there’s a paid version that integrates with Microsoft 365. So your company may already have activated this. This could be an easy one to play with depending on your company’s tech stack and ecosystem.
It was definitely lesser known overall in the generative AI market, but they’ve done quite a few recent, well, throughout 2024, huge marketing pushes, including a bunch of Super Bowl ads that really increased its awareness. And we also really like that the formatting and prompt responses are pretty easy to read and consume.
So it’s notably different the way that it. [00:20:00] And the formatting it often did bulleted lists versus paragraphs of text. So that’s just an interesting difference in the way that it’s choosing to communicate. It’s also connected to the internet in the free version, which is a very helpful feature. But you have to be careful as it doesn’t have the maturity of ChatGPT, and it could be misleading with its accuracy or confidence levels.
So it hallucinates and we’ll talk about hallucinating a little bit. In the next episode, all these platforms hallucinate but Copilot definitely does. Another thing that I didn’t mention about ChatGPT is. And that’s going to come up with copilot also is that they right now are being sued by multiple different parties as a result of the way that it trained its model.
So, it basically just let it loose on the web and let it digest hundreds of thousands or millions of documents and pages and websites and books and all of these original content pieces created by other people without permission or [00:21:00] compensation. So, lots of lawsuits are happening with ChatGPT.
CoPilot, so, it’s powered by GPT. What’s interesting is they actually have advertised that they will protect any of its commercial customers from these lawsuits based off of their uses of Copilot. So, if ChatGPT ends up kind of going down or being sued for 2 billion dollars, will that roll downstream to its users?
We don’t know yet, but we do know that CoPilot through CoPilot, Microsoft has made a promise to allow it to roll down to its customers. So we’ll see if they hold true to that. The next two are interesting. So we got Perplexity. Perplexity is privately owned by four co-founders. It was launched in 2022 and it leverages OpenAI’s GPT 3.5model and Microsoft Bing’s search engine.
So it is really kind of positioning itself as a search enhancement tool rather than generative AI tool. And one of the things that we really like about Perplexity is how it really cites its [00:22:00] sources for any searches that you do. For example, if you’re going to be doing market research or anything where you really need to understand where the data is coming from, we really like perplexity for those.
Purposes, but we will see how long Perplexity is around as these other models continue to get more sophisticated and people just get used to asking for these models to cite their sources. It may not become as big of a differentiator. And the last platform I want to chat about today is Claude. So Claude is actually my personal favorite platform.
It’s owned by Anthropic, which was started by former members of OpenAI. So at the very beginning of the episode, I mentioned how there was a little bit of infighting within OpenAI because of the lack of consensus around if guardrail should or should not be put around the training models, the kinds of responses and questions that you can ask AI.
And so four people that helped found OpenAI just said, forget about it. We’re leaving. And they started [00:23:00] Claude. They use a unique approach called constitutional AI. And that means it’s focusing on making the models helpful, honest, and harmless by having them self-critique and revise their responses based on the company’s guiding principles.
So we think that they wanted to develop an AI that had more guardrails from the start, as opposed to the direction OpenAI took. The Claude AI models have been developed with a strong focus on safety and ethical AI practices, and they’re designed to be transparent and how and why they share the information that they do.
An interesting thing is that Claude can actually analyze both text and Images and you can even understand complex diagrams. We know that in June, Claude 3 Sonnet was released, but they’re rolling things out continuously. All these platforms are. So I’m not going to talk about specific models for any of them, but they’re continuously releasing new models and updates.
The other thing that I think is interesting is that [00:24:00] the concept of constitutional AI sounds good on paper, but there are some queries. For example, our team asked it, how can I kill all Python processes in my Ubuntu server? That’s a normal thing a developer would ask, and Claude refused to answer it because of the word kill.
So is that a problem with AI model because it didn’t understand that or is that a problem with our prompting and we need to learn to use different language. So something to think about. But again, Claude to me is the most personable and the way that it communicates. So, thank you for tuning in today.
This is the first of three episodes that we’re going to be talking about AI. Today we really covered AI as far as how to understand what you’re looking at from a model or technology perspective and the six core platforms that are used the most right now. Next week, we’re going to get in and talk about the six core tenets of using AI and ways to convince your organization to [00:25:00] integrate AI and allow you to use it.
And then the final episode, we’re going to be talking about specific prompts and give you some tips about how to actually incorporate it into your day-to-day. I hope today’s episode was helpful, and we will see you on episode two of three of AI 101 for healthcare marketers with We Are, Marketing Happy. See you soon.
