AI, journalism and podcasting: The good, the bad and the ugly with Paul McNally of South Africa’s Develop Audio

Journalist, trainer and podcast producer Paul McNally was born in the UK but has spent a good part of his childhood and professional life in South Africa. He’s now based in Cape Town. He started his career in magazine journalism.

He started listening to podcasts in around 2010 on the American public radio network NPR, finding the podcasts similar to magazine features. He produced his first podcast episode in 2013 and moved further into podcasting after beginning to work in community radio in Africa. He’s the founder of Develop Audio, which offers training and produces programs with people who have compelling stories to tell.

As artificial intelligence began being used in more and more applications, Paul decided he needed to start exploring what it could and couldn’t do. He wanted to know beforehand what might happen, unlike when he was in magazines and the move toward digital completely hollowed out South Africa’s magazine industry to the surprise of many. To this end, he started learning how to code and began playing around with creating entire podcasts with AI. You can follow his experiments with and thoughts on AI in general at his Substack newsletter, Develop AI.

Paul McNally of Develop Audio and the PodcasTraining team

Paul is also a trainer on the PodcasTraining team, wrote the MethodKit Knowledge Guide and has just released a podcast series on disinformation called Challenging the Truth.

Tell us about your experience with creating a podcast only using AI tools.

So, you can't just tell an AI chatbot, “Please give me a podcast episode” and it’s done. Even though, this is rapidly changing. In my case, I learned to code with Python and broke production up into small steps and eventually the code that I wrote created a podcast. The idea was to have a weekly show about African politics that scraped Google News and was done completely through AI, host voices included. It was complicated and took a lot of fiddling around and tweaking the prompts for the AI tools. Right now I’m not sure if people would even listen to this kind of content. But it was an experiment to find out where we’re at this moment.

In your newsletter, you wrote about building an application that could produce a podcast episode from scratch without ever needing to record a human voice. It sounds like you did it. To me, it sounds a little dystopian. How do you see it?

I don't think it's a good thing. The interesting thing about the experiment was that the API cost made it unfeasible. The business model didn’t work. But that’s where we are now. The tech is going to get better, the cost is going to drop, and then I think people's resistance to this kind of thing will also fall. My questions in doing it were: could it be done? and what would that mean?

What is your overall view of AI? Is it something exciting or something very worrying?

It’s different for different people. You do speak to some, especially people from very small organizations, who are very excited about it. It allows five people to do the job of 15. For example, organizations in Malawi which exist in a very fragmented mainstream media ecosystem can use AI to greatly amplify the work that they’re doing. That side of it is very exciting. But when you talk to people in organizations with big newsrooms and classical setups, then it becomes difficult to argue how AI will make things better for many of the people there. For example, I’ve employed people in the past to do translations for podcasts, and it's difficult to make an argument that that job isn’t going away.

When I saw Sam Altman (CEO of OpenAI, which developed ChatGPT – eds.) at the World Economic Forum, he talked about how AI is not going to cost anyone their jobs, but that it will make people more high functioning. And, yes, that will be true for some people, but not everyone can be a manager. AI enthusiasts might say things like, “Yeah, this is great because I spent all my life making PowerPoint presentations. Now AI can do it.” And sure, that’s great for you. But if your job in a big company or corporation is to make PowerPoint presentations for executives, well, you’re probably freaking out.

I suppose AI has people fighting disinformation very concerned.

I'm doing a series on disinformation now. It's nothing new, people have been worried for 10 years about disinformation. But I don't know how people are going to deal with what AI could unleash in this area. No one's coming up with any real ideas about how to to react besides just not looking at your phone. If people get to this point where their day is just better if they don't consume anything, that's the worst outcome in terms of people's engagement with what's going on in the world.

So, we see lots of areas for concern. Let's look at AI with regard to podcasting. What's the upside for podcasters?

It can be huge in terms of making your workflow much simpler. We've talked about making something completely synthetic like I did, but that’s still hard to do. What you can do now is clone your voice and use it for ad reads. Or people can loan out their voices even. People with certain types of voices or accents will be able to make a passive income from their speaking voice. I think that’s quite exciting. Instead of people just stealing your face or voice, maybe you can sell the rights to use them.

The translation side with voices and dubbing in different languages is very interesting, even though it’s still incredibly limited with African languages. You can do a podcast in English and dub the whole episode into Spanish. It’s your voice, but in Spanish. That creates the ability for your podcast to be accessed in a way it couldn’t have been before.

What’s been your experience with AI audio editing?

It works but not well enough, in my opinion. I mean, I think if you were starting today and learning on an editing platform like Descript, you might have a different opinion. But if you’re kind of dyed in the wool with some things, like I am, I have that feeling that I could just do it quicker. The other positive side is post-production work, which is getting better and better. Ten years ago, you couldn't really clean a lot up if you had bad audio. You’d just have to re-record. Now it’s improving at a very quick pace and that's positive in terms of recording interviews if you don’t have a quiet place to go. In general, the leaps in post-production capabilities over the last year have been huge.

Are these tools affordable?

They are expensive, and they’ll take time to kind of trickle down. When I was in Ethiopia, people were saying that they were sure these tools would reach them, but by the time they were free or even affordable, people there will have been left behind in terms of tool creation and making inroads into the revenue side of things.

Early this year, you wrote that you were bullish about the outlook for 2024 when it comes to content creation and AI. Has that changed?

I feel that you've caught me on a negative day around this. When I wrote that, I was thinking that in terms of people being able to produce more content from the journalism that they're doing. We know journalism hits limits because it's linked to language, to being on websites without much traffic, etc. I think content creation tools are going to unlock a lot of that. I like this idea of, for example, taking content that appears on community radio and trying to get that into the mainstream press. I think AI tools, if they're used well, will supercharge that.

Give me an example.

Say you do work at a community radio station, and you've got an hour-long interview about a water crisis in a township. In theory, you can record that and with AI transcribe it and then translate it into English. And then take that huge document and compress it into a 150-word article. Then a newspaper will publish it or put it on a website. Without AI tools, it would take someone quite a while to do that.

Any final words?

Regarding experimentation, for podcasters, it’s worthwhile for them to explore AI tools to see if they can automate certain processes. But they should think about how much they want to hand over to AI. I've really struggled in terms of making podcasts with AI that I’m fully engaged with. I found it difficult to automate some bits and do other bits myself because, well, I just don't really want to work that way. But it’s important to figure out how much you want to give over and how much you want to keep for yourself. That’s key to this whole thing, I think.