The world is full of news written by Ai bots, but how much of it is true and accurate, how much is disinformation and how much is stolen from reputable news sources? Irresponsible news publishers don’t care much about that, they just want to fill their pages and generate clicks

Responsible news organisations, however, take their work seriously and are cautious about the integration of Ai into newsrooms.

At this week’s SMPTE MetExpo, Scott Burns, the ABC’s Senior Product Manager for AI explained the careful approach being used at the national broadcaster, balancing the risks and benefits of using Ai.

The ABC’s main use of Ai at present is to give journalists the ability to more effectively search the whole of the national broadcaster’s archival and current content to create richer news reports and better program content.

“We thought carefully about bringing Ai into the newsroom because brand trust is crucial and we didn’t want to do anything to risk that,” said Burns, explaining the complexity of a large public broadcaster that has to balance its public obligations, its budget and its ability to adapt to new technology that will meet audience demands.

“People and Product are the 2 key things we have to keep in mind… We needed to carefully anticipate risks while empowering staff to use the new tools better…

“The aim was to curate what is being made now and also leveraging what was made in the past across the whole landscape of audio, visual, news, text and multimedia content.”

To achieve that aim the ABC developed ABC Assist, a tool to help users extract value from decades worth of trusted data.

“ABC Assist is not a tool to create or publish content, it is to enhance the content research capabilities,” explained Burns. With it a journalist can have “an interactive conversation” with the whole content of the ABC.

The difference between this system and standard search is that the ABC’s system uses semantic search in combination with large language models to look and listen within all the content, providing an index to all the content within radio shows and tv reports, not just the subject data and keywords that were entered manually into the system when it was archived.

“This answers the question from journalists ‘how can I find what exactly was in an unscripted audio or video file,’” explained Burns. “It is not a silver bullet, it requires close attention from people, but it can deliver what is needed faster.”

The technology team trains staff to understand the importance of prompts and how to use the system in all its nuances.

The expertise gained by years of life experience meant that the system was pushed to perform at its best by the journalists who first tested it.

“We had a relevance challenge when we trialled it. Journos told us. ‘I expected more, I know my content, I want to see if it knows the content as well as I do or better. It didn’t cut the mustard when they first tested it. They don’t want to be told things they already know, they want to find something new to further enhance their story.”

With that challenge facing them, the ABC’s Technology design team improved the product and upgraded the user experience based on the feedback from journos.

“Journos don’t take things at face value, they want to quickly find out where does this source come from and is it valid. They needed a quick check method in seconds.”

Semantics and key words can be taken out of context to produce poor search results, but the ABC system overlays its language model to find the most relevant content, giving journos only the most relevant results with the sources displayed so that they can make quick decisions about relevance. “The LLM functions as a judge-in-action to go beyond just the keywords, using reasoning to determine if this passage is relevant to this person’s prompt,” explained Burns.

radioinfo covered the initial implementation of the ABC Assist system here last year.

In the same MetExpo session, Jason Smith, from the workflow solutions team at Associated Press presented his work using Vibe Coding to improve journalistic experiences with news creation systems.

The AP Storytelling system “redefines a newsroom’s capabilities, making it easy to transition from legacy systems to advanced, intuitive solutions.” The cloud-based platform means journos can “focus on crafting compelling narratives while the technology handles the intricacies of digital workflows and multi-platform collaboration.”

Vibe Coding is a software development approach where developers leverage AI language models to generate code from natural speech instead of manually writing  code. “Non software developers can talk to an ai model to tell what wanted and the ai will develop the code,” Smith explained, as previously covered by audioinfo here.

Using Vibe Coding Smith developed a storytelling interface and a prompter interface for journalists to more quickly compose and air stories on tv.

“It’s easy to build, test and change things while building with Vibe Code.” His tips when using Vibe Coding are: Organisation is key. Know exactly what you want, don’t just jump in. You will have false starts, be flexible and change vibecoding providers if you are getting stuck. And finally, have fun doing it.

Contacts:

Scott Burns

Jason Smith

Reporting: Steve Ahern