We had Google’s NotebookLM AI podcasts and now we have ‘Virtually Parkinson’, an AI podcast resurrection of the famous interviewer Michael Parkinson (aka "Parky").

Virtually Parkinson is interesting because it’s a very convincing recreation of the original human voice. In AI-Parky's first episode ‘he’ interviews musician Jason Derulo. The questions are typical Parkinson, digging into the guest’s life experiences.

The comedian Jenny Eclair has heavily criticised AI-Parky, pointing out that dead people competing with live podcast hosts for clicks is troubling (she used stronger words than I have). The publishers of Virtually Parkinson countered that the show has given work to 15 people to create it, the point being that such shows are more than the host. So Virtually Parkinson has created work for real live humans, even if it's hosted by a dead person.

So what’s it like? I listened to the first episode and it’s so-so. The voice is spot on, impossible to separate from the real human one. And yet, something’s a bit off. AI-Parky seems reluctant to spark off the guest’s comments and comes off a bit robotic and never interrupts or laughs. At one point Jason asks AI Parky a question, but that’s clearly something the model’s not yet designed for. The show is strangely devoid of humour, which was a key component of the original.

As a result, Virtually Parkinson doesn’t quite have the flow of a real conversation. Not terrible, but just not quite how a real interview would go if the characters riffed off of each other a bit more. It’s more a series of questions and answers, less a natural conversation.

The show was created by Michel Parkinson’s son (Mike) and at the end of the first episode contains a discussion amongst the team of creators. They explain the process involved, which is super interesting. Mike acknowledges the slightly formulaic nature of the questions, so this is pitched perfectly and recognises the reality of what they’ve created. This isn't a static thing and the plan is to improve AI-Parky across the episodes. So each episode should get a bit better than the previous. The team clearly acknowledges the need for a more natural conversational flow and that's a priority.

Strangely, the more natural conversational flow is something that Google's NotebookLM already does well and is one of the things that so impressed me about it. Perhaps that indicates a differing level of AI between the two initiatives. It almost certainly hints that AI-Parky can get better, perhaps a lot better.

I'm not sure I would have prediced podcasts hosted by dead people, but we live in nteresting times ⚰️🎤

Do you think this is interesting, or a terrible idea? Imagine historical figures discussing current events. Or a personal hero talking about things they could never have experienced. It feels to me that there's potential here, but also risk. Not least, the risk of putting words into the mouths of dead people and getting them to comment on things when we cannot know for sure what their opinions would have been. AI-Parky is pitched about right, being an open exploration of the possibilities and limitations. But others might go further and it feels like "there be dragons" might be an appropriate response when it comes to AI resurections 🐉