We take a deep dive into the controversial world of generative AI and podcasts, focusing on the upcoming podcast Virtually Parkinson, featuring an AI-generated Sir Michael Parkinson.
In the last issue our AI application of the month was the 'podcast-generation' feature of Google's NotebookLM – simply upload an article, and with a few clicks you can generate audio of a podcast-like discussion between two 'hosts'. The 'hosts' in NotebookLM are two anonymous voices, and they're always the same. But what if you want to make a podcast with a particular voice? The voice of a celebrity, for instance – how about their mannerisms and their style, too? And what if they're no longer alive?
Then you might be interested to learn that Sir Michael Parkinson will return virtually for your ears in the soon-to-be-released podcast, Virtually Parkinson. Developed by Deep Fusion Films in close partnership with Sir Michael's son, Mike Parkinson, the podcast will feature an AI-generated Sir Michael interviewing real guests (real in the sense that their voices come from the guests themselves and are not AI-generated). The program will include a debrief with the guest, discussing their experience of the interview. Deep Fusion Films have trained their AI model on around 30 hours of material, with fine-tuning to get the AI Sir Michael to feel as real as the original Sir Michael.
This might sound slightly creepy at first, and it certainly raises questions around the ethics of AI (as do many things these days, as its use becomes more and more ubiquitous). For example, what data ought to be used for training? Does using AI take a job away from a human? And so on.
Benjamin Field at Deep Fusion Films has emphasised the ethical considerations that were borne in mind in this project:
- Informed consent – the project has full consent from Sir Michael's estate, and was produced in close cooperation with Mike Parkinson and the rest of the family.
- Licensed data for training – 30 hours of licensed archival material was used for training the model.
- Transparency with audiences – i.e. making it clear that AI Sir Michael is not actually Sir Michael.
Regarding concerns around 'replacing' humans, Field has said that he views AI Sir Michael not as a replacement for a human interviewer but a 'unique project', the purpose of which 'is to explore AI's role in media and human engagement'.
Virtually Parkinson has come at a time in which the use of generative AI in creative industries is hotly debated – with some championing its use and others speaking out against it. A year on from the writers' strike (in which generative AI was a key concern) sees the release of the film 'Here' which uses generative AI to de-age Tom Hanks and Robin Wright across a 60-year time period; on the other hand, directors Scott Beck and Bryan Woods included a statement in the end-credits of their new film 'Heretic': 'No generative AI was used in the making of this film'.
In the case of Virtually Parkinson, Field – who has a background in AI policy development – seems to have tried his best to allay the usual concerns when it comes to using generative AI, but it's a divisive issue, among those within creative industries and the public. Legally, the emphasis on licensed content for use as training data is good to see and copyright holders would undoubtedly welcome this becoming standard practise.
Whilst Virtually Parkinson has been developed with the consent of the real Sir Michael's family and estate and is not trying to trick anyone, there may be some podcast listeners who just can't get behind AI Sir Michael. But perhaps that's simply because it's new and strange right now – give it a few decades and we'll be happily watching James Dean in Fast and Furious 48.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
