Outsourcing to robots: Is Generative AI the death of creativity in podcasting?

How do we make podcasting futureproof? What needs to be done to challenge the industry to innovate and produce daring content? In the Revelations column, Meera Kumar pokes and prods the audio industry and its creations to reveal the shows worth listening to and their place in the zeitgeist…

I did ask Chat GPT to write this column for me, but it turns out that it can’t quite capture the pzazz and zest that I exude when writing about audio. So, to my dismay, I have had to write this entire thing myself.

New technology can threaten jobs, traditional methods of storytelling, and our sense of comfort in the way we have learned to engage with podcasting. Job automation is an incredibly important subject to consider, however I will not be addressing it in this particular issue of Revelations. Instead I want to focus on the applications of AI and how it might benefit the evolution of our artform.

There are parts of the podcast production process that may be expedited by AI. Take the development stage for example. You could use Chat GPT to generate your content calendar or a script around particular phrases or topics. However, I do not suggest using the ideas that Chat GPT produces as your sole form of idea development as there are four main pitfalls to doing so.

Firstly, the human brain is brilliant and may be inspired to connect different concepts in a unique and creative way that Chat GPT may not. Instead of creating exactly what Chat GPT suggests, use it as a starting point and ask yourself: “What else? So what? What am I missing here?”. AI may make this initial brainstorm easier, but there is currently no comparison to the unusual and complex connections that our brains can make.

Secondly, Chat GPT can produce repetitive results. I spent two hours using the software this morning and whilst I was impressed at the breadth of information I was able to learn (which would have taken much longer to obtain using Google), oftentimes it was saying the same thing in different ways, oftentimes it gave similar results, and oftentimes it just reworded the same answer. You get the picture.

Thirdly, other podcasters may be using Chat GPT to plan their content, and they may input many of the same prompts as you, and therefore may receive similar answers to you. In this way, the use of AI could lead to the homogenisation of the podcast landscape, which would kill the medium (and also your show) because everyone is making content based on the same information.

Fourthly, AI can reinforce existing biases due to the data it’s trained on; if an AI algorithm is trained on data that contains racial or gender biases, or information from one particular country, then it’s likely to produce biased information. A lot of data in the world is biased in these ways, so I’d be curious to speak to someone who works in the field who is able to explain how they are mitigating that.

The recording and editing process is another part of production AI can aid. For example, AI can be used to repair audio, remove filler words, and denoise your recording. Adobe recently released Adobe Podcast, which does pretty much everything. Most impressive is its ability to make the recording sound like it was conducted in a studio, which it achieves by altering the frequencies of your voice and reducing the background noise. You can use it to deep fake your host’s voice to generate an entire episode, or correct a single sentence that you don’t have time to re-record with the host.

Adobe Podcasts and AI tools like it could even be used to create a whole series without a host having to record a single line! The Times recently produced an episode of Stories of our Times about deep fake audio, with an accompanying video showing the host, David Aaronovitch, reacting to his AI voice clone. Aaronovitch identified a common issue experienced when using AI this way – mispronunciation. Some words were pronounced inconsistently throughout the podcast, and in a way that the host himself would never pronounce said word. But the world moves very fast, and creating a daily show takes work. We often want to cover more, but it’s impossible. German publishing house, Heise Gruppe, have solved this by using AI to clone their host’s voice and then used text to speech to create a second daily episode for their show. It increased their plays by 37%. But whilst the episodes increased engagement, feedback from listeners identified similar pronunciation issues in English and German, as well as an unnatural lack of breathing sounds.

There is a wealth of AI applications in post-production – we’re all used to using automatic transcripts already – but now that you can create automatic captions on promo videos you can use AI that utilises natural language processing to convert your podcast audio to text. This type of AI can be used to write an accompanying blog or the shownotes for an episode, it could write an entire Twitter thread for you, or you could use AI to create a filter for your TikTok fans. I think the priority here shouldn’t be to churn out as much content as possible, but instead AI should be used to increase our efficiency so we have more time to increase the value of each piece of content – something that is especially useful for smaller teams who may not have the beefy budgets.

I’ve recently downloaded Fathom, a more social podcast player that makes use of AI to allow you to save moments in a podcast and share them with your friends. The player also uses AI to recommend other shows, a general function that Chat GPT itself has told me could be a concern: it may be prone to the same racial or gender biases mentioned, and at the very least “As AI algorithms become more sophisticated, they It may favour popular shows and topics over less well-known ones, making it harder for new and diverse voices to be heard.”

However impressed we are at its current capabilities, AI’s computational power is doubling every six to 10 months, well ahead of Moore’s Law. The implication is that, what we’re seeing now is AI in its infancy. It’s scary but it might just be the most exciting stage of the internet – so far!

Still can’t get your head around what AI actually is? Listen to Meera’s AI podcast recommendations:

Hard Fork – ‘GPT-4 Is Here + The Group Chat Bank Run’

The New York Times weighs in on the latest developments in ChatGPT last week.  Listen on your favourite app >>

Short Wave – ‘Can you teach a computer common sense?’

NPR’s podcast Short Wave discussed the idea of computer thinking back in January, before the release of the latest ChatGPT. Listen on your favourite app >>

Between Two Mics – ‘This Episode Was Written Using ChatGPT’

The remote recording podcast has a play around with AI creativity, with an episode written by ChatGPT. Listen on your favourite app >>

Imaginary Worlds – ‘The Human Touch’

A show that focuses on fantasy and sci-fi worlds, this episodes looks at the way AI has been encroaching on visual art, the legalities of copyright (or lack of them) and what can be gained and lost through AI art. Listen on your favourite app >>

Talks at Google  – ‘Ep210 – Mo Gawdat | Scary Smart: The Future of Artificial Intelligence and How You Can Save Our World’

A conversation from Talks at Google about the bigger picture and concerns with AI. Listen on your favourite app >>

Synthetic Stories – ‘Episode 1 – Amelia’

This Is Distorted has made an entire podcast using AI, from the artwork, to the script, title – and even the Press Release we were sent. Synthetic Stories is a podcast, about a podcast, made by… the podcast? This recommendation was written by a human, however. Listen on your favourite app >>

Stories of Our Time – ‘Artificial intelligence: Bright new future or the end of humanity?’

Stories of Our Time with David Aaronovitch (possibly) asks if machines could be our surgeons, our judges and our artists, what would it then mean to be human? Listen on your favourite app >>