Being a visual learner, I’ve never really felt the appeal of podcasts. I’ve even tried listening to them in the car while driving, but unlike music that I can just enjoy mindlessly, I find it hard to absorb a podcast’s content while watching out for silly drivers.
But the idea of listening to a 30-minute (or longer) podcast while doing nothing else is also unbearable to me.
Understanding people like me who have short attention spans, Melvin Poh decided to make his podcast episodes between 3 – 12 minutes long.
But there’s something more to his podcasts, something a little different than what we’re used to. Called Empirics Asia (Empirics Podcast) on Spotify and other platforms, his episodes are all generated by artificial intelligence (AI).
Doing a better job than humans
Melvin runs Empirics Asia, an open-access knowledge-sharing platform that’s aiming to build a knowledge hub of various topics via contributed written content. The podcast series is an extension of what the site does and was actually started to solve a key problem it faced.
Running Empirics Asia meant working with large amounts of data round the clock, with hundreds of article contributions from authors on a weekly basis.
Its team of volunteer editors then have to sift through these articles to publish ones with what they deem valuable sharing.
Thus, the Empirics Podcast began life as a proprietary AI that the team developed to assist them with their back-end editorial work.
Its core tasks were identifying knowledge trends, analysing the submitted data, and checking and editing the data through pre-set syntax parameters.
“Although initially we had planned to only utilise the AI in the back-end of our publishing activity, it soon became apparent that it was doing everything involved in managing media content on its own,” the Malaysian entrepreneur told Vulcan Post.
“Perhaps, it was even doing it much more efficiently than a human could.”
Empirics Asia therefore set out to develop and integrate the narrative capabilities of the AI, allowing it to “speak”. Today, the podcast is managed, produced, and narrated all by the Empirics AI itself.
When I gave it a listen, I was pleasantly surprised by its smooth speech and only-minor awkward intonation, but was still able to tell that it was an artificial production.
Fun fact: The team modelled the voice profile of the AI after British actress Gemma Chan, who they found had a soothing and clear voice.
But it’s still limited in its capabilities
Empirics Podcast produces explainer-type content spanning a wide range of social science that includes philosophy, art, business, science, sociology, politics, literature, economics, and more.
It collects data on what to produce by assessing user behaviours on the Empirics Asia platform, trending fields from approved journals and news portals, and audience responses to previous podcast episodes.
Thus far, Melvin said that he’s seen Empirics Podcast reflect global trends quite accurately.
For example, in Q2 2021, the podcast’s topics covered alternative investments and financing, following the rise of stimulus programmes, benefits, and grants to combat COVID-19. In Q3, as the world saw mass vaccinations, its topics covered starting businesses and entrepreneurship.
Despite the vast number of topics covered though, I couldn’t help but feel each episode missed something: a human touch.
I know the point of Empirics Podcast is for the AI to maintain a neutral stance when disseminating information. To me though, what makes podcasts even worth listening to are the additional insights hosts give through personal commentary.
So in a way, yes, Empirics Podcast is unbiased to a certain extent, but it also felt surface level, as if just digging up knowledge I could Google.
This is because of the AI’s limitations. “Although effective at determining the relevance and value of different insights, our AI is not equipped to fully detect bias on its own,” Melvin said.
Therefore, the team still needs to exercise strict quality control over how the AI aggregates topics and produces content. This means creating strict parameters for the AI and only allowing it to work with the content found on Empirics Asia.
At the end of the day, I see Empirics Podcast as a place where I can keep up with what’s trending and get an overview of the topic. Then to better understand it, I would turn to other media to get the whole picture.
I’m not sure if the AI will ever achieve what a human-produced and hosted podcast does for me, but I’d like to be proven wrong.
Further expanding the use of AI in knowledge-building
Currently, their largest podcast audience has been from Hong Kong, followed by those from Singapore, US, UK, and Malaysia.
“In our assigned category of Education in Hong Kong, we have ranked within the top 25 listened-to podcasts in 2021, and generally, we attract an average of around 1K – 1.5K listens on each episode,” Melvin told us.
He’s firm on not monetising the Empirics Podcast for the foreseeable future though, because it’s an integral part of the team’s commitment to the open-access knowledge movement.
Instead, he’s got other plans to monetise the AI. “I can reveal that we are currently working on a paid mobile application that will feature the Empirics AI.”
“[We envisage] that this app will be similar to modern voice assistants like Siri, but instead allowing users to ask knowledge questions that the AI will seek to answer in real time.”
Feaured Image Credit: Empirics Asia
