The latest innovation in artificial intelligence has made podcasting easier, but made differentiating between humans and artificial intelligence harder. A new text-to-voice feature of Google‘s AI-powered software NotebookLM can transform written words, such as news articles or blog posts, into an artificially generated conversation between two AI hosts.
The result is “so natural and realistic that there is no way you’d believe you weren’t listening to two real people talking”, said TechRadar senior editor Graham Barlow, who generated an eight-minute podcast from a blog post. After listening, “the world simply wasn’t the same any more; I no longer had confidence that I could tell what was real and what wasn’t”.
‘This is the devil’s work’
Google launched NotebookLM last year as a research tool, to help users summarise notes and other documents. It already uses Google’s AI model Gemini 1.5, said The Verge. This latest experimental feature is “like an audio version of that”, having AI hosts “summarise your material, make connections between topics, and banter back and forth”.
But the rapid emergence of AI-generated audio has sparked fears that it could upend the human industry. “I am not at all religious, but when I discovered this tool, I wanted to scream, ‘This is the devil’s work!’,” said ZDNET contributing editor David Gewirtz. The quality “verges on making creators and content producers like me begin to feel the heat”. The “voice fidelity”, the appropriate use of colloquialisms, the “completely organic nature of their banter and the fact that there even was banter”… the final product is “indistinguishable from a real broadcast”.
But it could also be a “watershed” moment in democratising the industry. It cost Google “billions of dollars to turn my article into a podcast”, said Gewirtz. “But it cost me nothing. It also took moments. That’s a huge reduction in the barrier of entry to content production.”
Future versions could become more realistic and increasingly personalised. Currently, NotebookLM generates a conversation only between a male and a female host with standard American accents. But imagine if you could choose the type of speaker, the style, the accent, and “maybe edit a little of the AI-generated script”.
‘One-to-one relationship’
NotebookLM isn’t the only software offering human-free podcast services. Just hours after exploding pagers and radios killed dozens in Lebanon, a “peculiar new show appeared on podcast apps”, said Wired. “‘Pager Protocol’ isn’t about the attacks aimed at Hezbollah members or the Israeli operatives believed to be behind them. It’s an ongoing AI-generated 10-part series created to rapidly turn those unprecedented events into a fictional podcast.”
Launched on the day of the explosions, the podcast was created by feeding information about the news event into Claude, a ChatGPT-like generative AI tool. Claude “spat out an outline of a story”, said Mark Francis, co-founder of Caloroga Shark Media, the company behind the podcast. The team wrote a script and “put it back into the AI for even more massaging”. The script was then fed into AI voice-generating software to create an audio version narrated by an AI host.
But others warned of harmful implications for the industry. The real power of podcasts is “the host-audience relationship”, said Jason Saldanha, chief operating officer of the non-profit digital radio company PRX. The most successful have a “one-to-one relationship with their audiences”.
An AI host might seem tempting in the short term, to save time and resources, but “flooding the market with content to get the lowest level of engagement” is not a “long-term strategy”, he told Wired. “It’s gross and it’s bad, and, ultimately, you’re cutting off your nose just to make an extra dollar.”