|
A funny thing happened on the way to delivering an audio version of the Free Press.
Several months ago, we engaged with a company called Instaread on an initiative to allow readers to listen to the journalism our newsroom produces. With just a click of a mouse or swipe on your smartphone, a calm and collected voice would read our news stories to you, so you could relax or do another task while taking in the day’s news.
The initial feedback was more than positive. Visually impaired readers told me how important it was to hear the stories they wanted to access. Our site metrics showed many readers were tapping into the voice-version of our stories, a trend that continued to grow. Based on that early experience, we began considering what else we could do with audio news.
Advertisement
 Why this ad?
All that came to a screeching halt, though, when we realized the computer-generated news reader was making mistakes.
Suddenly, the reading app was misreading Premier Wab Kinew. For reasons that my intelligence can’t explain, the artificial intelligence behind the transcription reader began saying “Premier Pallister” or “Premier Whitecloud” when our copy referenced Canada’s first First Nations provincial premier.
As we listened more, we found other problems, too. In short order, we had heard more than enough — so last week we temporarily hit pause on the audio version of the Free Press until we can work with the company to determine what was going wrong and how to prevent it from ever occurring again.
I don’t know if this is an example of an AI “hallucination,” where large-language models suddenly start spewing forth misleading information posing as facts. But I do know that we spent weeks before this feature was launched in extensive testing, down to ensuring its chatbot could read “Wab Kinew” correctly (along with Métis and Wiebe and Portage and Hellebuyck and many other words important to Winnipeggers).
If there is a cautionary tale here, it’s the need to be wary of the siren song of AI. Leveraging machine-learning to read the news to Free Press subscribers seemed like a safe bet. Until it wasn’t.
We recognize how important trust, credibility and accuracy are to our brand, which is why we instantly shut down Instaread – and also why I am disclosing what happened to you.
But what about brands that don’t care as much about those values?
If you ask one of the new AI chatbots to tell you what a famous person did or said, or about a local court case or what was in yesterday’s provincial budget, can you rely on it?
The Free Press is keeping an eye on the potential of artificial intelligence — but you can trust us to have your back and apply human intelligence when it comes to the perils of this new technology.
|
