Reuters is experimenting with using AI agents to speed up the video production process. Reuters hired its first AI video producer this week to help partially oversee the effort.
Typically, Reuters newsrooms have used AI technology to speed up text-based workflows. But Rob Lang, newsroom AI editor at Reuters, said he sees an opportunity for agent AI to function as a kind of supercharged video editing system. Currently, Reuters uses AI tools to create and process video metadata and cut various edits of video coverage.
“We asked LLM to cherry-pick and create what we call a wrap edit, and they’re actually doing a very good job. I mean, it’s not even multimodal. So if we can actually understand the AI as an agent AI system and be able to look at things and compile things, we might be able to have the AI build the edits for us. That would be very cool,” Lang said on stage at the Digiday Publishing Summit Europe on Oct. 30. 28.
To be clear, this AI agent is probably doing a rough job. Lang said there will still be humans involved making complete video editing decisions, such as Reuters’ new AI video producer Enrique Flores Roldan.
However, the idea is that it can be useful for more basic and time-consuming video editing. Ideally, Lang said, clips can be cut to account for continuity errors, such as when a subject puts on or takes off glasses, or how the sun peeking through clouds affects the subject’s lighting.
“It’s…writing instructions so that they can be understood and applied.” [by an AI agent]” he said.
About 60% of Reuters’ newsroom uses AI, Lang said. The most “skilled” AI users, around 50 to 100 of Reuters’ 2,500 journalists, use AI to “give flavor to the code” and aid investigative reporting.
“We’re at 60% and going up every month. Right now it’s probably 5% a month. We’re being told to get to 100% by the end of the year. I think we’ll be closer to 80%. Let’s touch wood,” Lang said.
The biggest hurdle to getting there, he noted, is that not all newsroom staff feel they need to use AI technology.
“I was talking to a photographer the other day, [who asked] How can we use AI? And I said, why don’t you use it to find a parking space when you come to a new city? It’s just simple things like that. There are ways to use AI that are not necessarily journalistic,” Lang said.
So why isn’t the Reuters newsroom using AI? Fact-checking, Lang said. AI systems have limitations, Lang said, especially when it comes to separating “junior” web content captured in training data from reliable information.
That’s why Reuters is building the RAG search database system. This is a technical expression that describes the archive of Reuters content that the LLM system can use as a basis for generated responses and to obtain more reliable output.
“You can use this as a solid foundation to fact-check with AI,” Lang says.
However, it’s still in its early stages, and Lang’s team is experimenting with what real-world use cases for this might look like. First, Reuters created a RAG database with a style guide for journalists to ask questions, such as how to spell Ukrainian President Volodymyr Zelenskiy’s name (Reuters spells it “Zelenskiy”).
“A lot of what I’m doing is actually saying, ‘What is this going to look like in the future?'” Lang said.
Lang said other AI tools Reuters has built to speed up newsroom production processes include:
- A “sandbox” where newsroom staff can enter prompts. The system creates a series of prompts to help you with the production process.
- Numerous CMS tools such as heading builder and bullet point features.
- Another CMS tool that lets you create the first draft of your story. For example, you can process press releases and interview transcripts and combine the two to create a first draft.
- An AI summarization tool called Fact Genie processes thousands of press releases each week and makes recommendations to newsrooms about what to cover.
- A tool called Federal Bot, created by Reuters journalists, pulls in all publicly available data and press releases from the U.S. government (three times a day), looks for new information, flags anything journalists should take a closer look at, and sends emails to the people closest to the relevant beats.
