Can AI-generated podcasts boost science engagement?
Podcasts generated by artificial intelligence are shaking up how researchers interact with the literature and the public.Credit: Aleksei Gorodenkov/Alamy
“My brain starts to die when I try to read a journal article,” says Jessica Sacher, a microbiologist at Stanford University in California. Papers are written to be technically comprehensive rather than understandable, she says. “I just want someone to tell me what the gist is.”
Now, artificial intelligence (AI) chatbots can help to turn difficult concepts into clear, engaging podcasts, providing a simpler method for keeping up with the literature. Now, Sacher regularly absorbs the latest research by tuning in to AI-made conversations.
A flurry of AI podcasts has emerged (see ‘AI podcast tools’). Some tools require minimal input from the user. Others come equipped with customizable features, such as the option to change the voices or languages of the ‘hosts’.
AI podcast tools
Several AI tools are now available that can generate podcasts from research articles, books or other content. Here’s a breakdown of some:
Google NotebookLM: This tool can convert up to 50 sources, including PDFs, YouTube videos, websites and documents, into two-person podcasts. The podcasts are accompanied by other materials, such as study guides and glossaries. Users can ask questions and the tool responds with inline citations. The technology is in an early testing phase and does not currently require a paid subscription.
ElevenLabs: This platform allows users to create podcasts in 32 languages with a choice of voices and styles, from a British storyteller to a whispering American. The podcasts can be tweaked to focus on specific sections, with different speakers assigned to each. Generating an AI podcast of an article requires a subscription, but the free version allows users to generate a monologue based on up to 5,000 characters of pasted text. The speech can be adjusted in several ways, such as from dynamic to monotonous or to change the level of exaggeration.
Open NotebookLM: This platform generates podcasts in 13 languages using up to 100,000 pasted characters or a website link for longer texts. It allows users to suggest a topic of focus, decide between a fun or formal tone and choose between short (1–2 minutes) or medium-length (3–5 minutes) summaries.
Wondercraft: This tool allows users to generate podcasts in several styles, such as a newsreel or a guest interview, either using a prewritten script or by generating a dialogue with AI. Audio elements such as music and sound effects can be mixed in, including laughter. It also allows users to upload and edit podcasts produced by Google NotebookLM. Wondercraft is free to use, but a paid subscription provides more features, such as choosing the user’s voice or more than 500 others.
Learning on the go
AI podcast tools can boost engagement in several ways. Benjamin Gooddy, a neuroscience PhD student at King’s College London, uses them to digest his database of papers that are relevant to his work.
“I can chuck in 50 papers at once, and it can spit out pretty user-friendly audio,” he says. “I actually use it as a podcast to listen to on the tube into work.”
The podcasts can also help with exam revision. Mmesoma Francis, a student at Kyiv Medical University, says that the tools helped her to cram 60 pages of material the day before one of her classes.
“I just put the PDF in, and it made this podcast that was 17 minutes long,” she says. “I use it now on an almost daily basis because it helps give an overview of the topic, even if I plan to do in-depth studying after.”
In a study1 posted on the arXiv preprint server earlier this year, philosophy students who listened to AI podcasts performed better on tests than did those who only read textbooks.
Other researchers argue that the podcasts are better suited to public outreach than to digesting the literature. Pedro Beltrao, a geneticist at the Swiss Federal Institute of Technology in Zürich, says that robo-podcasters tend to explain basic concepts using multiple analogies at the expense of diving into details, which can make the tool less useful for specialists who want to keep up with studies in their field. Sacher works on phage therapy and says she wishes she could skip past all of the openings that introduce phages as viruses that infect bacteria.
Gooddy says that AI podcasts could help researchers to persuade news outlets to cover studies that might “have gone under the radar because perhaps they were almost too technical or just very dry, but the actual research findings might be quite exciting”.
Compared with standard podcasts, which are resource-intensive to make and tend to focus on buzzworthy studies that have the best chance of drawing large audiences, AI podcasts allow scientists to promote less eye-catching papers at a much lower cost.
Early-stage limitations
Enjoying our latest content?
Login or create an account to continue
Access the most recent journalism from Nature’s award-winning team
Explore the latest features & opinion covering groundbreaking research
