From hours to minutes: How to revolutionise your newsroom with AI 🤳🏻
Hi,
My name is Sara Forni, AI Product Manager at Atex, and this is MyType, a newsletter dedicated to journalism, innovation, and artificial intelligence.
If this mail was forwarded to you and you would also like to receive future releases, you can subscribe to MyType here. You can now also follow MyType on Linkedin, Instagram and Twitter.
Now, let's get started!
Following the success of last month's interview with Jaemark Tordecilla (if you missed it, you can read it here), this month we went inside a newsroom that is very active in the development of AI solutions to optimise workflows.
A journey we took together with Sannuta Raghu, Lead - AI Working Group, News & Journalism at Scroll Media Inc. Scroll is a small newsroom in India, with under 20 journalists, known for their deeply reported longform journalism.
Among the various very interesting insights that emerged from the interview, I would like to make it very clear that developing AI products and innovating one's own newsroom does not require millionaire capital. Sometimes even with small teams and a lot of expertise you can find decisive solutions to improve your daily work.
I hope this interview can be an inspiration to many small/medium-sized publishers and journalists!
The subtitle of this newsletter is: AI and journalism, opportunity, or threat? To answer this question, we decided to ask experts in the field. So, what is the answer you would give to this first question?
I think we are at this juncture right now:
It’s not yet possible to zoom out and look at the big picture, and we are making decisions and assessing situations based on the information we have today.
So I would add a good dose of journalistic skepticism and take a ‘Yes, but…’ approach for both the opportunity vantage point and the threat vantage point. Yes, an LLM finds a needle in a haystack for my investigative report and saves me days of high-bandwidth work, but is it built with the human values my news organisation and I value? Yes, the problem of misinformation is real, but does increasing the supply of misinformation with AI mean that people will consume more of it? We are untangling and wading through many, many grey areas, and this is true with any new adoption of evolving technology.
In our small newsroom, for example, we are experimenting and building with cautious optimism. Our starting point and hypothesis is that the use of generative AI in our news workflows will make the lives of our journalists easier. But we inspect our output with a fine-tooth comb before we publish because we encounter bias, inaccuracies, misinterpretations, and misrepresentations very often.
The hope is that the payoff at the end of the end of the opportunity is much, much larger.
In your current job you lead the AI team at Scroll Media. Tell us more about this experience: When did you feel the need to have an in-house AI use case team in the newsroom? How many people work with you? How is the team composed?
In 2019, we had to let go of a big part of our video team. Many organisations worldwide have gone through this in the last 8 years. We decided we will not over-hire here and subsequently lay off colleagues and work on how we could make our video production workflow more resource-effective. For context: In 2018-19, when we were producing an award-winning news-explainer show for India’s biggest OTT platform, our team was at it for 14 hours every weekday for 14 months, performing high-cognitive load tasks.News agency footage in India is limited, which meant that we were relying on complex composite graphics to tell a news story effectively. At the end, we had *one* 12-minute video everyday. This was not sustainable.
After several rejections, we won a Google Innovation Challenge in 2022 to build an AI-based video production tool for high-quality short-shelf-life videos. This is where we started.
Over the last 18 months, because of our URL-to-mp4 tool, we’ve been able to gain experience in several areas that make up the workflow.
Our working group is composed of 2 engineers, 1 product leader, and 1 editorial leader. A larger policy input group helps shape the direction we want to go in. One engineer and I have been working on all our news and journalism-related AI experiments and POCs everyday.
We are moving to productizing our URL-to-video tool, and we now work with a project team of 14 (including frontend engineers, backend engineers and product designers) to ship this to market.
What are the main products you have produced for Scroll Media?
Currently, we are working only with generative AI. We are taking an ecosystem approach to building, which means we are prioritising tools and features which could sit within our CMS one day. When our value-add meets market demand, we are moving to productizing (like in the case of our URL-to-video tool).
The URL-to-video tool is our main product; it set us off on this journey. It is a traffic tool for video production which converts a news article into a video file, optimised for 9:16 social media formats. It cuts production time of videos from several hours to under 5 minutes. We have optimised it for news and journalism standards. It is facts and accuracy-first (Our editor tests and fine-tunes the tool *everyday*).
In our current phase of development (closed beta), the tool is capable of accurate output in English, Hindi and German. We are currently testing for Gujarati, Marathi, Bengali, Malayalam, Kannada and French.
Our other tools in early development and POC stages are: a repository-linked image tagger, two LLM-based research tools to aid our investigative reporting, and a style-guide-based compliance checker and corrector.
In your years of direct experience with the use of AI in the newsroom, what are some key lessons you have learnt?
There are so many lessons, but let me leave you with three:
Your output is going to be as good or bad as your input data; don’t expect magic. For example, if a model is largely trained on American news articles, one can’t expect nuances specific to journalism in India (which we take for granted) in the news copies we generate. Or that the word ‘president’ invariably throws up an image of Donald Trump or Joe Biden instead of the Indian president, Droupadi Murmu. For a small newsroom like ours (which can’t invest in building a tailor-made model), this becomes an extra hoop to jump through. (I would suggest using the Pareto principle to decide which hoops to jump through first: solving 20% of the problems will lead to achieving 80% of your expected results).
While transitioning to using generative AI in your news production workflow, the workload of your frontline staff carrying out real-world testing is going to increase significantly. They need to be supported, empathised with, and trained. (Regular check-ins to understand and solve pain-points and being directly accessible when needed to quickly troubleshoot or debug is working well for us.)
A use-of-AI policy built by a diverse group of colleagues is very helpful (especially with the different vantage points different teams bring to the table).
What advice would you give to an editor who would like to approach the world of AI but does not know how to navigate in this complex and multifaceted field?
With my current experience, I will only be able to speak about using generative AI for news production in a low-resource environment.
I would suggest starting with an important metric (not necessarily the north star metric) your organisation uses, or a problem (again, not necessarily the biggest one) you think needs to be solved in your news organisation, and auditing to see if AI can really help solve it.
If the answer is yes, I have found that creating customGPTs on ChatGPT serves as good POCs to quickly test if an idea/solution is useful. Depending on your results, you can decide to invest resources or approach senior management to secure buy-in.
Another suggestion would be to foster friendship between editorial and engineering teams. We often speak ‘different’ languages, and A LOT of things get lost in translation. Not working in the usual silos and collaborating everyday has helped me greatly in the last 18 months while building our AI tools.
Do you have experience in AI and journalism and would you like to tell a wide audience about it? Write to me for an interview on MyType!
Big opportunity!
Did the interview with Sannuta Raghu inspire you and would you like to learn more about how to integrate AI into your editorial work?
Contact our AI team or send me an email (sforni@atex.com) we can do it together!
There would be no 'digital joynalism' without sharing. If you like this newsletter, you could share it with colleagues, friends and family :)
Have a good weekend,
Sara