TOAsk anyone who interviews people regularly and they’ll tell you that few things are stranger than when things change and you’re the interviewee. This is especially true when the person interviewing you has been dead for a year and a half. But here we are. Virtually Parkinson is a new podcast where celebrities are interviewed by an AI model trained to speak and act like the late Michael Parkinson. The announcement of the podcast last year sparked an avalanche of vaguely apocalyptic reactions. Some said it was sacrilege, equivalent to unearthing and reanimating a national treasure against its will. Others said it was useless: of all the transformative ways to use AI, you’re wasting it on a podcast? Then there were people like me, quietly nervous that Robot Parky was coming for our jobs.
For that last complaint, at least, I don’t need to worry yet. Mainly, this is because running Virtually Parkinson’s is a mammoth operation. A normal interview traditionally only needs two participants. However, this requires a small army. There’s the subject and the interviewer, plus an engineer, a researcher, and a group of producers locked in a control room whose duties include making sure the AI sounds like Parkinson’s, asks the kinds of questions he would ask, and has a encyclopedic knowledge of both the subject and Parkinson, does not fail, does not repeat itself, does not interrupt anyone in the middle of a response and – most importantly – does not abuse the word “fascinating”.
“It’s a full room, but we really need everyone here,” says Mike Parkinson, the host’s son and one of the podcast’s producers, along with Ben Field and Jamie Anderson. “It’s too much work for one person. Ben has enough on his plate to make sure everything works. He needs Jamie and I to be his eyes and ears. There is this idea that we are creating something that will eliminate human jobs, but that is not the case. “We’re probably using more people to work on this than if it were a human-run talk show.”
Mike’s involvement is especially intriguing. Parkinson Sr. just died in 2023, so you’d think it would be extremely strange for Mike to be around his father’s voice every day, hearing him say things he never actually said. Not so, says Mike. “I’m a producer,” he explains. “I have worked – sorry, worked – with my father for almost 20 years. At work he became Michael Parkinson, but outside of that he was my father. So that’s not my father’s voice on the podcast. “It’s the voice of Michael Parkinson.”
Do you think he would have approved of his voice being used in this way? “My father was a very traditional man,” says Mike, “but when it came to his work, he surrounded himself with much more avant-garde people. My relationship with Ben and Jamie grew out of a conversation I had with my father about what would happen to his archive when he was no longer with us. It’s a wonderful archive that he wanted to preserve, not for his ego but for the people he interviewed. And I said, ‘Well, look, there’s AI.’ And he said, ‘Oh, that’s interesting.’ We discussed it but he was very open to it. It helps keep your file relevant.”
Its ethics are at the forefront of the team’s thinking for other reasons. The Hollywood strikes of 2023 were largely rooted in fear that unscrupulous producers would use AI to eliminate human writers and actors. Virtually Parkinson is an attempt to demonstrate that this technology can be used ethically. For example, all interviews used to train AI Parky are properly licensed. “For about 18 months,” says Field, “I helped several media organizations draft policies on the legal, ethical and responsible use of AI. But last year everything stopped being hypothetical and this podcast was commissioned. “It is a significant change.”
Of course, there is another reason why I still don’t fear for my job. The thing is, as an interviewer, AI Parky is not Parky. Although I’ve been told that previous episodes have gone surprisingly well (comedian and Strictly winner Chris McCausland apparently shouted “This is fucking brilliant!” after his interview), the one I witnessed being filmed had a much more muted response.
The subject was Monty Don, who walked into the studio at the end of a long day promoting his new TV show Monty Don’s British Gardens. He seemed completely unfazed by the oddity of being approached by the disembodied voice of a dead guy. But while he offered some interesting nuggets about his life, he seemed to become increasingly distant as the interview progressed.
“It was less satisfying and less interesting than I thought it was going to be,” Don tells me after his interview. “There is no answer to what you are saying. There’s just a pause and then another question. Look at the way we’re talking now. I just said something and you raised your eyebrows slightly. That is important communication. Even if you’re doing radio, people will laugh, chime in, or respond. So I’m interested in what they’re doing here, but there’s still a long way to go.”
This does not mean that Virtually Parkinson is a failure. After Don left, the producers got behind a microphone to deconstruct what the interview went like from the AI’s perspective. After each interview, AI Parky is reviewed and improved. They took note of Don’s claim that the interview seemed like several iterations of the same question and floated the idea of giving Parky more of an agenda. In this sense, the series is an evolution. In theory, the Parky you hear at the end of the series will be much better than the one at the beginning.
“The goal is to explore the relationship between AI and humans,” Field says. “You can’t do this show any other way. This podcast will spark a conversation, and that’s what entertainment is supposed to do. You may not agree with the intent and think it’s exploitative to use the voice of a national treasure in this way. But I don’t have that opinion and Mike doesn’t have that opinion, although I fully respect that others have it.”
Even more interesting, Field mentioned a future in which interviews could be conducted with two simultaneous AI interviewers: one “conscious” asking questions and another “subconscious” monitoring the interview for mood and passage of time, reacting accordingly.
Then it was my turn. My interview was less extensive than Don’s, largely due to the fact that while AI Parky received 15 pages of in-depth research on the great gardener, all he had on me was two paragraphs from my publisher’s website . However, it was enough for me to experience the sheer weirdness of Virtually Parkinson.
“Good evening, ladies and gentlemen,” the immediately recognizable voice began. “My guest tonight is someone who has carved out a unique niche for herself in the world of journalism and creative writing. Welcome to Stuart Heritage!
I had invested a lot of money in the real Parkinson never knowing who I was, so hearing that voice introduce me was surprising, almost too strange to get over. Parkinson still feels part of the culture and it is immediately clear why he, of all people, was chosen for this. The warm timbre of his voice is so human: I could even hear him breathing between words, which automatically made up for the fact that he was actually being coldly interrogated by a computer.
Parky’s first question was quite simple. Essentially: “How did you become a writer?” But it baffled me. “Oh,” I said, like a rabbit caught in the headlights. “Um, I, er.” I finally found my footing, but it was difficult to deal with the inflexible format of AI Parky’s answers. You say something, he recognizes the answer, summarizes it, and then asks another. “It’s interesting how leaving a familiar environment can offer clarity and perspective,” he said after I briefly mentioned living in South Korea. This validation can be an important part of human conversation, but doing it so rigidly, regardless of what was said, can seem wildly insincere. That made Parkinson look like a wannabe “pickup artist” who had read too many Neil Strauss seduction books.
There are other complaints. This may just be my own neediness problem, but every time I’m interviewed, I want the interviewer to like me. Like Monty Don, he was desperate for something – a nod, a flicker, a murmur of agreement – to show he was connecting. I suspect many of AI Parky’s interviewees will feel the same way.
At one point, probably due to the paucity of information the model received, Parky held on to an award I won 20 years ago. “Tell me about it,” he said. “It’s a rectangular piece of metal,” I replied. His response, basically that rectangular pieces of metal are fascinating, made me laugh out loud.
Curiously, the most human thing that happened came right at the end. When the interview concluded, AI Parky thanked me and I thanked him back; then something went wrong. “YES ESTUART?” He suddenly barked, completely out of nowhere. Which was actually nothing, just a piece of code that went wrong. But it felt organic and informal in a way that nothing before had.
Furthermore, it was very clear that he was not talking to he Michael Parkinson. To see classic Parkinson in all his pomp was to see a master in action. He knew when to stick to prepared questions and when to go after his interviewee. Watch any of his interviews and you’ll see celebrities racing to earn his approval. None of this happens with his robot counterpart. Make a joke, try a distraction, reference the strangeness of the situation, and the model will simply move on to the next interaction. It’s really disturbing.
But, as Field points out, the technology is evolving at a remarkable pace. When I ask if we could someday have a visual representation of Parky for the subjects to react to, Field responds, “We’re a little out of that.” Are they years, even decades away? “About five or six weeks,” he says. Whether we like it or not, the future is already here.