Radio’s Most Cynical AI Experiment Just Got Exposed

ARN tried to cut costs with a fake host. They gutted their credibility instead.

Well, it doesn’t bring me a lot of joy to say that I was right about CADA radio host Thy being AI-generated.

Former colleague Sam Buckingham-Jones (from the Australian Financial Review) picked up the story mid-last week, did some proper journalism, and managed to get the whole thing confirmed.

By all reports, this was no easy task. I’m aware of a couple of different outlets which attempted to look into Thy – apparently ARN did a good job obfuscating the truth. But it was eventually revealed that CADA had been trialling Thy for six months without disclosing to their audience she was AI. ARN’s reluctance to come clean on this matter demonstrates, I believe, an awareness that they were operating in an ethically grey area.

“Thy”

The fallout from this story has been predictable. Thy was the only Asian broadcaster on ARN’s major stations – this did not go unnoticed by others in the media.

“ARN, did you actually think using “Thy” was a tick in the diversity box?” Presenter and voice over artist Teresa Lim wrote on Linkedin.

“Because for the very limited number of us actual Asian female broadcasters who have fought our entire careers for a more wider and inclusive representation in Australian media, this move is tokenistic and exploitative.”

So ARN blundered into a very avoidable and totally predictable public controversy. But why? What did they have to gain?

“At ARN, we’re always exploring how new technology can support great content and enhance the listener experience,” they wrote in a statement. “While the trial has offered valuable insights, it’s also reinforced the unique value that personalities bring to creating truly compelling content.”

I’m sorry, it took them six months to work that out? Surely one mocked-up radio show should have been sufficient. If I make a sandwich that tastes like shit, I don’t need to keep on eating it for half a year to be certain. I definitely wouldn’t repeatedly serve that sandwich to others.

Ethics aside, anyone with ears should have been able to tell how woefully inadequate AI hosts were. But even if the quality was poor, ARN probably saved lots of time and money, right?

Well, bafflingly, I don’t think they did.

Junior radio presenters are cheap and plentiful. Thy’s shift would have cost around $35k a year for a casual to cover. I suspect than the enterprise deal ARN has struck with AI voice-cloning software ElevenLabs would be in the same vicinity.

The process isn’t particularly efficient, either. ElevenLabs uses text-to-voice technology, meaning every sentence that Thy “spoke” had to first be written out manually by a person and then fed into the program. These sentences are then turned into MP3s, which need to be loaded into their radio playback system Zetta for use on air.

Arguably, it would be easier to put a person in front of a microphone, and have them talk off-the-cuff about what’s coming up, maybe share a few anecdotes about their day, or take a few calls and connect with the audience.

You know, basic radiocraft.

In being swept up in the AI hype, ARN have seemingly forgotten their primary purpose is to make entertaining radio. Audience trust is hard won and easily lost – the damage they have wrought on their own reputation for so little material gain is staggering.

Thy wasn’t just bad ethics – it was bad business.