HMN 2025: How AI-generated podcasts open new doorways to make science accessible

The first study to make use of synthetic intelligence (AI) expertise to generate podcasts about analysis revealed in scientific papers has proven the outcomes had been so good that half of the papers’ authors thought the podcasters had been human.

In analysis revealed within the European Journal of Cardiovascular Nursing (EJCN), researchers led by Professor Philip Moons from the University of Leuven, Belgium, used Google NotebookLM, a personalised AI analysis assistant created by Google Labs, to make podcasts explaining analysis revealed lately within the EJCN.

Prof. Moons, who additionally introduced the findings on the Association of Cardiovascular Nursing and Allied Professions (ACNAP) convention in Sophia Antipolis, France, mentioned, “In September 2024, Google launched a brand new function in NotebookLM that allows customers to make AI-generated podcasts. It made me take into consideration the way it could possibly be utilized by researchers and editors.

“When I did a primary check case with one in all my very own articles, I used to be flabbergasted by the top quality and the way pure it sounded. At that brief time period, I spotted that such a system may have the potential for use by journals for science communication. Of course, the standard and accuracy wanted to be evaluated. Therefore, we designed this study to guage its potential.”

Prof. Moons, who’s Professor of Healthcare and Nursing Science at KU Leuven and editor-in-chief of the EJCN, and his colleagues chosen 10 several types of articles and contacted the authors to ask if they might agree to participate within the study. The contributors weren’t knowledgeable that the podcast about their study could be generated by AI.

The researchers despatched the AI-generated podcasts to the authors to guage for engagement, trustworthiness and AI detection. They adopted up with a questionnaire (generated with the assistance of ChatGPT) and a 30-minute interview through Microsoft Teams.

The authors reported that the podcasts captured the important thing factors of their papers in “quite simple, easy-to-understand phrases,” had been well-structured, had steadiness by way of size and depth, and that the podcast ‘hosts’ had been skilled, with some authors even assuming the audio system had a background in nursing or drugs. The conversational interplay of the hosts was a big asset.

Most of the authors mentioned the podcasts had been dependable sources of knowledge. However, some commented on the American accent and magnificence, together with some hyping of analysis findings with the usage of phrases equivalent to “wonderful,” “groundbreaking” and “completely.”

They mentioned there have been some inaccuracies and misrepresentations, there was generally lack of context, incorrect use of medical terminology and mispronunciation of medical phrases. The podcasts would should be checked rigorously for accuracy earlier than their launch.

AI detection

When they had been requested: “Would you be stunned if I instructed you this podcast was generated by Artificial Intelligence?”, 5 authors mentioned “sure,” together with one creator who had performed analysis on AI beforehand. They mentioned they had been “shocked,” “amazed,” and even “having an existential disaster.”

Five others mentioned “no.” The contributors mentioned that if AI podcasts had been launched, they need to make it clear that they had been AI-generated, acknowledge the authors and reference the unique article.

All the authors mentioned they thought that sufferers and most of the people could be probably the most acceptable goal for the podcasts, primarily due to the tone and the flexibility of the podcasts to clarify the articles in an easy-to-understand method.

However, some commented that the podcasts may additionally be helpful for enabling health care workers to remain updated with the most recent analysis, entry it extra simply and make the unique analysis article extra seen.

Some of the authors prompt that podcasts could possibly be tailor-made for specific audiences in line with age, pursuits or ethnic background because the expertise develops and improves. At current, it’s not potential to vary the voices or the language of the AI “hosts,” however these options will most likely be enabled sooner or later.

Prof. Moons mentioned, “It was hanging how correct the podcasts had been basically. Knowing that we’re simply originally of this type of AI-generated podcasts, the standard will develop into higher over time, most likely throughout the subsequent few months. Another necessary facet is that these podcasts appear best suited for a non-technical viewers, as an example, most of the people or sufferers.

“If podcasts could possibly be generated by AI, that might actually be a game-changer. Podcasts could possibly be made with little or no work, simply by importing the article and possibly a little bit of prompting. This could possibly be a sustainable model to get the message out to individuals who don’t usually learn scientific journals.

“I feel that this can be a expertise that may permit editors, journals and researchers to speak about science to most of the people. It is not going to make human podcasters redundant.

“There will all the time be a marketplace for human-made podcasts, most likely as a result of not all matters are able to being addressed precisely or appropriately by AI. I may even think about that there can be hybrid podcasts, by which human podcasters and AI come collectively in several sections of a podcast episode.”

Now, the researchers plan to discover additional potentialities of those podcasts for science communication, together with seeing what sufferers and different members of the general public consider them. They additionally need to examine if it will be potential to make use of AI-generated podcasts for scientific convention periods.

“For occasion, making a podcast on the content material of periods on the ESC conferences, as a abstract for individuals who didn’t take part and need to get a blurb of it,” mentioned Prof. Moons.

The researchers have made an AI-podcast on their analysis paper about AI-generated podcasts. A hyperlink to it’s contained within the on-line revealed paper.