Making a Medical Podcast Using AI

We Tried to Make a Medical Podcast Using AI—See What Happened

Even massive companies like Spotify are experimenting with AI voices – they even have an AI DJ and “they” sound great. Have a listen, so you can see what we mean! 

So, if AI can work for a music DJ, can it also work for a medical podcast intended for clinicians?

We decided to use AI to make a cardiology podcast and find out! Here are some of the advantages and disadvantages that we came across during the process.

Technical challenges of an AI podcast

1. Accuracy

Factual accuracy is mission critical for any medical podcast, and obviously non-negotiable. So, for quality control purposes, we feel it’s important that someone on the podcast team has expertise in the subject matter. As a cardiologist, Franz knows his way around cardiology, so it was a natural choice for the podcast topic.

Can the AI do all our research for us?

Although there are AI tools that promise to search the medical literature to write scripts and articles on any topic you can dream up, that could quickly become an excessively time-consuming, fact-checking nightmare. 

Although a good AI is likely to be accurate most of the time, it’s inevitable that hallucinations (errors) will creep in from time to time. And we obviously can’t have factual errors in a cardiology podcast. 

So, here’s what we do to avoid that problem: Franz finds recent cardiology studies that have important and/or interesting insights. Then we decide (not the AI) which information from the study should be included in the podcast episode. 

This is much more time-efficient than letting the AI do all the research for us. The thing is, if the AI does its own research, we’d potentially have a lot more source info to go through when fact-checking, and the odds of missing a mistake would increase. Whereas when we pull out the most important information we want to cover in the episode and ask the AI to create a script based only on that, there’s only a relatively small amount of info we need to fact-check against.

2. Can an AI voice read our script without sounding like a robot?

The good news is that nowadays this is absolutely possible. 

But, it’s not as easy as a lot of the hype may make it sound. That’s particularly true in medicine where we constantly run into medical terms or names of medications that the AI hasn’t come across often enough to pronounce  correctly. Luckily, we have a workaround for that, which we’ll go into more detail about in the next article.

Also, to be perfectly honest, there are still things that creep into the audio to give away the fact that you’re not listening to a human. Sometimes the inflection or tone isn’t quite right, for example.

3. Will clinicians be interested in listening to an AI podcast?

The jury’s still out on this one. As with any podcast, it takes time to build an audience and get traction. 

NPR experimented with an AI-generated podcast episode and the results were quite interesting to listen to. But when Drew Carey tried it, fans reportedly got upset. However, one of the reasons people listen to his podcast is to hear him specifically. So it’s understandable that an AI replacement won’t appeal to many people in his audience. 

In our case, we’re obviously not replacing a human being that listeners already know. Also, our cardiology podcast is intended to be purely informative—and an AI voice won’t change the information provided by the script.  

So, we’re curious to see if listeners will be willing to give our AI podcast a chance. And if they do, we hope it’ll be an enjoyable and informative experience that makes them want to tune in for future episodes, too. Will occasional imperfections in the tone or inflection of the AI voices be something they’re willing to overlook? Or no?

Problems with traditional podcasting that AI-assisted podcasting can solve

1. The need for a podcast host and recording equipment

With traditional podcasting, we need an on-air personality to be the voice of the podcast. That individual needs to schedule time for recording, make sure they have good quality recording equipment, and so on.

In contrast, an AI podcast host is always available to “record” and never has issues with recording equipment. So in terms of flexibility, using an AI host is helpful.

2. Writing podcast scripts

If you find writing scripts to be a chore, or you don’t feel comfortable writing them, using AI can give you a head start. You’ll still have to fine-tune the output and fact-check it to make sure it’s free of errors, but for many people, that’ll be faster than writing the script themselves. And even fast writers may find time-savings in using AI to write the script, despite having to budget extra time for fact-checking the output.

Want to learn how to prompt AIs like ChatGPT in a way that saves time and produces output you can actually use? Medmastery has an entire course to show you how to do exactly that—and the best part is that it’s free! Check it out here!

Did we succeed at using AI to make a podcast?

We did indeed manage to make an AI podcast and we can’t wait to tell you exactly how we did it! In fact, we have several episodes live right now. We’ll share exactly how we create our episodes in Part 2 of this article.

References

Educational Resources

• Medmastery Cardiology podcasts [Spotify] [Apple podcasts]
• Guilleminot S. Making a medical podcast using AI – Part 2. LITFL
• ChatGPT Essentials for Clinicians. Medmastery
• Guilleminot S. AI in Healthcare. LITFL