It has been nearly two years since we set ourselves a big challenge: With appointment viewing dead or dying for TV news, how could we use Artificial Intelligence to reinvent our approach to television and get more eyeballs for our journalists’ work?

That’s when a small cross-functional team in Mediacorp, led by our newsroom AI strategy and solutions team members PT Alagan and Wang Yin, embarked on a project we affectionally call AI SmartCut.

Putting AI to work 

We decided we would train AI to clip out CNA news bulletin highlights and upload them to our Web site and YouTube channel in more timely fashion.

We collaborated closely with a vendor to develop a solution tailored to our needs. We used AI to identify audio patterns in our prime-time news bulletins — the jingles, the news presenter and journalist voices, and so on.

The solution transcribes the bulletin elements and uses natural language processing to group related video elements together: the introduction and the ensuing news package, an entire live interview with a guest, or a ground report by a correspondent. These are uploaded automatically onto our CMS with a suggested title and keywords.

We started out with, at most, 30%-50% accuracy in getting usable clips and trained the AI to about 85% accuracy over the course of a year and a half. This resulted from labour-intensive data validation, but we knew we wanted to minimise the grunt work of having someone watch a bulletin just to mark in and out for TV video reports.

The project was launched in the last quarter of 2023 and results have been transformational.

The tedious job of finding and uploading TV clips was previously spread across four teams: media managers, digital journalists, sub-editors, and an audience growth specialist who worked to optimise video titles and thumnbnails.

It is now down to two teams: media managers who inspect the AI-cut clips and the sub-editors, who check the text describing the videos.

Streamlining processes

Now that AI has made it so easy to clip out TV reports, TV journalists write titles and descriptions for their videos online. This frees up sub-editors to focus on optimising video titles and descriptions and creating better thumbnails.

This task used to be done by digital journalists or sub-editors who were unfamiliar with TV stories. They had to watch the videos, interviews, or live hits repeatedly to work out an accurate summary. It made sense to have TV journalists, who were closer to the product, take ownership of this job.

“It has made a tangible difference to our workflow,” a senior member of the digital team said.

The AI solution and its resulting workflow have significantly reduced processing time, and TV reports can be uploaded onto the CNA Web site and YouTube within minutes of broadcast.

In the long run, this solution opens possibilities for news bulletins to be reimagined as modular and personalised. Could we offer a “build your own bulletin” experience based on the latest, most relevant updates and interviews on a particular news topic?

We are excited by the idea and heartened that we are making TV news less ephemeral, giving each journalist a stronger digital footprint for their work.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .