Essentially what I was interested in here, was how I can create a really engaging audio format for a piece of existing content that I have created.

Hypothesis:

AI can take a piece of pre-existing source content, and create a logical, engaging podcast out of it

Process:

I took a chapter from my 2013 PhD thesis, which was all about worldviews. It’s a deep philosophical topic, with lots of very academic language, and it took me nearly a year to write. Every paragraph of this chapter took immense effort in terms of research, and was mentally taxing to figure out what I thought of it all, the implications for my research, and then to put it down on paper. It felt like a good challenge.

I copied and pasted the original text from the chapter into Google’s Notebook LM, and had it generate an audio from it. The style available for the chapter was deepdive – essentially a discussion between two presenters.

Output:

Here is the audio file that Notebook LM generated.

To say that I was excited by the output is a massive understatement. I was thrilled with it. The scripting and production quality is much higher than I could achieve on my own. It is faithful to the underlying ideas in the source material, and the only big error is that the “presenters” describe the source material as a research study, which it wasn’t. It was a chapter from a research study. Whilst this may seem like splitting hairs, it’s a significant difference because the focus of the study was not worldviews, the research was all about how the conditions for happiness and success are co-created in organisations. It argued for the legitimacy of an approach to research which uses individual and collective perspectives as data.

Was It Useful / What Did I Learn?

Very useful, sort of. I was initially really excited about the potential for “translating” my research into something that was really engaging, and which had high production standards, but the lack of ability to edit or control a significant error in how the AI had “understood” the source material is a limiting factor. However, it did get me thinking about whether this could be another useful approach to generating a “script”, which could subsequently be refined, and fed through another AI application.

It also got me thinking about whether AI had moved on to such an extent that I could now create a similarly engaging and exiting podcast, without spending a year researching and writing the source material. But that is another experiment.