Global Approaches to AI in Media: From Italy to Indonesia

AI Usage Notice: In preparing this introduction and the episode transcript, AI tools were used with careful human oversight and editing. We believe in transparency regarding the use of AI in our work.

Branislava Lovre: Welcome to Aimpactful. Today, we will talk about implementation of AI in newsrooms. Our guest is Mathias Felipe de Santos, a journalist, researcher and lecturer. Welcome Mathias.

Mathias-Felipe de-Lima-Santos: Thank you. Thank you for the invitation. I’m really happy to be here and share a bit about my experience and also my work at academia about AI and media.

Branislava Lovre: How did it all start, how did you make the connection between journalism and AI?

Mathias-Felipe de-Lima-Santos: My connection between journalism and AI, started a long time ago because, frankly I came from IT background, so I was a technologist before, and then I decided to move to media industry, and that’s where I try to combine my both backgrounds and back then. I was way, was concentrate on data science more and how this could be applied to journalism and somehow connect with data journalism.

And then I had the opportunity to join a Project, European project, called JOLT and this project, which was a research project, that’s also where I started doing research so I could have the opportunity to work also with, you know, media partners. And we had the possibility to spend some months with them and then developing project with them. And that’s where I did my first AI project for media.

So I work with La Nación Argentina. So, we developed a project that, we were trying to map the solar farms across the country because renewable energy was a hot topic back then, because they had another government that were really investing on that. But we wanted to see how those was growing.

So, it was very interesting back then because my idea was to work with data journalism and then naturally become AI. I remember also Florencia Coelho, who is the news manager, education manager there. She was doing a lot of, she did this article and if I’m not mistaken, at Harvard. And she was like studying, you know, the impact of AI in the media. That was really at the beginning. No one was talking about that. And she was one of the pioneers of doing that.

And that was really good because she came with a lot of ideas. And then I also had my own ideas and we could find intersection and do this first experiment. And from that I started working the other projects with the organizations, but also investigating the impact of AI in the media industry from the academic lens.

And that was very interesting because combining both was very unique and very challenge for me, because I always have to navigate from one to another. And it was very unique perspective because also, as people might know, La Nación is a traditional newsroom. It’s a large newsroom in Argentina. But, then I had also the opportunity to work with smaller newsrooms in Brazil and other countries, and that’s a different scope and different way of how we can bring these technologies to them.

So that’s how I started also looking for AI, from a critical lens and trying to understand how this would impact the media industry and, you know, create these, what I usually call AI divide. I know organizations that have resources and have staff and have time to dedicate to build these tools or experimenting with this tools, and other organizations that doesn’t have money, staff and then they are experimenting as, you know, opportunities show up, but some of them struggle to continue developing that.

Branislava Lovre: You’ve spoken to many large and small newsrooms around the world. What are the most common and useful applications of AI newsrooms?

Mathias-Felipe de-Lima-Santos: Among this project that I had the opportunity, to you know study more and interview the creators, I see a lot of projects that are really trying to help, you know, journalists to do their work faster and also, do part of what they’re trying to do with data journalism, but now doing with AI, so analyzing huge amount of data, having access to the data which was not possible or was not easy for journalists if they don’t have those skills. So some of these tools are in this way helping to, you know, get insights from the data.

But also there is a lot of tools that are giving access, to them, to other ways of analyzing information, that it’s sometimes really complicated and requiring a lot of time, among this project, for example, we have seen, you know, one that attracts a lot of my attention, it’s for example, in Costa Rica, there is a news organization called CLIP and they create, you know, like a platform that they connect different sources of data, and then they try to get insights to help investigative journalism.

Similarly, in Argentina, there is one organization, a large organization that create from group that’s called Grupo Octubre, and they create, you know, like, Visión Latina, which is a platform that give access, you know, information for journalists about the visual content that they have. So they could easily search and trying to find, you know, one video about someone, or you know, pictures about someone that they are like, doing some news story, for example.

And this is something that they are trying also to become a product that they could sell for other organizations. And that’s something that I think, I believe that’s going to be the future of these organizations that not only offering, you know, like information for the audience, but also offering products that is going to help maybe other news organizations but other sectors. And they could benefit from the revenue stream that these products could generate.

Branislava Lovre: Could you highlight some projects that have impressed you the most?

Mathias-Felipe de-Lima-Santos: The other thing that was very impressive and for me it’s a very interesting was a case from Econai in Nigeria and they create like a full platform to, you know, getting data from the audience, trying to create a community and everything, relate to sustainability, these natural disasters that they have a lot. So trying to bring data about that sometimes works really hard for them to have access. So having a platform that helps journalists, you know, audience to connect them and having this data and trying to visualize the data and have an environment that they could grasp more information. That’s something very interesting, very unique.

There are a lot of other projects. And even in Europe, for example, I was spending some time with RAI, the Italian public service media, and they were developing a lot of interesting projects. So they, similar what the Argentines newsroom did. They create also a platform that, you know, journalists could search term and get, you know, like news stories, but also brought videos from this person or an entity that they want to search.

And they also start experimenting. And I think that’s something that we’re going to see even more with generative AI so just large language models and for example they use that to generate titles, which is very interesting and I think it’s very important because then, can give ideas to journalist and also help them to become SEO driven. Because today what we are doing stories should be in a platform that people can find it. If there is a machine that can give us suggestions and then we can choose or even inspire us to create a title that will help the machine to find, but also the audience to find our stories. That’s going to be very relevant.

And the same is, different organizations, RAI and also others that I saw like Núcleo in Brazil. They were creating this summarization by trying to summarize the stories that help them, and help the audience as well, to absorb a huge amount of information in a very summarized way that, you know, diminishing the time that journalists need to, for example, look for information, confirm, but also give them a new idea and help them to create new stories.

Branislava Lovre: What are the main challenges newsrooms face when using AI?

Mathias-Felipe de-Lima-Santos: Smaller newsroom tend to have a lot of issues to develop AI tools for different reasons. As I mentioned, like budget is one of the main reasons, but it’s also what motivates some of these smaller organizations to do some of these experiments. As I was telling you, for example, the grants from these big tech companies, funding, they create grants, attract these organizations to develop ideas that using all these topics, the hype of AI and bring this money for the newsroom. That also is going to support the newsroom for the short time.

But on the other hand, what is really hard for them it’s like first trying to use these models in a proper way that works for the different languages because we are talking about smaller newsrooms that sometimes they cover a minority language or a language that’s not well trained, even though, for example, Indonesian is a language that’s spoken by a huge amount of population because Indonesia is a huge country but not necessarily the models are really well prepared for Indonesian, Bahasa Indonesian.

And so for example, and then we have countries that, you know, they speak something that is a hybrid between different language. For example, Philippines they speak Tagalog and English. So the systems need also to be prepared to deal with, you know, something that is hybrid. It is not even one language. So they have, they are struggling a lot with that.

And, for example, it is not only the models per se, but also the basics of technology. For example, one of the projects that I interview, which is amazing project that was trying to bring, you know, these news stories for people who are blind or have any, you know, visual capacity. So they couldn’t read properly. And so they are doing this AI that is converting to text to speech, which is something really useful. And nowadays it’s way more natural then used to be, which was the robotic voice. So it’s something that it looks like someone is reading for you. But problem is not that.

But when they’re trying to bring this technology and combine different newsrooms, they face a big issue that, you know, the CMS of these organizations are made of different systems and they not necessarily have, you know, APIs or ways to connect to this solution that they created. And then they create a platform or a solution, let’s call like that, that they cannot move forward to every single organization because there is different systems, different, you know, and making solutions that needs to a lot of adjustment, that needs to connect with different platforms is something that these organizations are not prepared.

Then there is, this brings me to the other issue that they have, because their staff usually they don’t have a lot of I mean, usually they don’t have technological staff on their organizations. And then when they decide to do a project and they get the money for that, then they hire someone. But who proposed this project? Someone that not necessarily know how things work and when they realize that what they propose is way bigger than the scope of the money that they have. They have to reduce this call, do another project.

Sometimes they try to develop the idea, but then they realize that’s going to take way more than what is offered for this grant. And another thing is to bring these people in because technological people, the staff that is, have technical skills, they cost way more than journalists. So these organizations are also not prepared mentally to, you know, have, you know, payslips that you’re not used to have. So they didn’t plan to have that when they developed the project.

And there is a lot of competition because, usually when they offer a position for someone to work with that, this person needs to learn to understand the business of journalism, how the industry works. But once they get it, they get offers from other industries, that pay way more. For example, banking, fintechs and these professionals move to other industry and then we have the project that are being developed and then they need to find someone. So there is also just learn curve that it’s really important and they cannot, like continue with this person because they need to pay more and they don’t have money for that.

So there is a couple of issues that these organizations, mainly the small ones, struggle a lot. Unfortunately, those are not the only problems. There are more. I think another thing that’s really important, the data, the data to train these models because not necessarily there is a lot of data available. And mainly when we are talking about, you know, like these countries that have freedom of information, that is really is I mean, it’s really bad or open data is also really bad. All this data that could used to train these models are not available.

It makes it harder for them and one of the organizations that I work with in the past and they struggle to find data to train their model that we’re talking about, you know, that visual content because there is no data available for that. So we need to collect this data. That means, you know, also to pay for this data or using in a proper way that is not going to be, you know, against the laws and it’s going to also respect the boundaries that is important for this organization.

So there is a lot of issues that are interconnected in development of AI that organizations generally are not aware. And mainly when you are small organization and you have also, you know, like 5 to 10 people in there, in the newsroom and there are, one or two people that are really interested and they move forward with this project. And when they don’t have support of their news managers or someone who is leading this team, it’s really hard to move forward.

So I think there is all these issues that it’s really important and these organizations have to be in mind when developing, you know, these projects.

Branislava Lovre: When we talk about the data, it is also important to mention disinformation and misinformation, which is also topic you focus on.

Mathias-Felipe de-Lima-Santos: I think this disinformation is really a topic that has been hot and you know, a lot of people discussing about that for the last years. And with generative AI. The situation is going to escalate for more level that we haven’t imagined. And just to give you an example, I was, you know, walking the street and then I was waiting for something and then I heard a conversation. I plan my trip using, you know, ChatGPT and for them, was like, that’s the most reliable search. It is like Google. So they could find all the information. But the point is, they are not aware how these tools are created. They are not aware how this data works.

And we know that generative AI and particularly what we are talking here, ChatGPT, create, you know, results that are not necessarily real, but they look like real. You have, you know, like a plan, someone plan a trip for you saying you should go to this spot because the best time is April. And then if you don’t check this information in the Internet, they’re going to realize that sometimes the best time it’s November, December. And people don’t have this literacy because they don’t know how these information is produced.

And when we use ChatGPT, I mean, very few place that we can indentify that the information is not real. Sometimes the tool tell you, oh, maybe this information is not real, you can check. But most of the times they just reply you as a conversation and then you get the information that you want. But you are not sure if you are not familiar on how this systems work. So I think that’s the biggest issue.

So people tend to create a every time more on this generative, content that was generated by AI. And I think there is also another issue that it’s going to be challenge for us practitioners, researchers to first, identify this content. This AI generated content is going to be every time you know more realistic looks like real content and we or tools need to escalate on the same level to identify this content.

And I think, I’m not saying that we shouldn’t use this tools, but we should be aware and how to use this tools and how we can engage better and also be aware that, you know, the content that they generate is not real sometimes.

Branislava Lovre: Your primary field is academic work. We often read your papers recently you achieved an outstanding result with more than 500 citations an incredible achievement. Congratulation.

Mathias-Felipe de-Lima-Santos: Thank you. Yeah. No, it it’s. Yeah. This academic milestone sometimes it’s really interesting because for example, I reach these 500 milestones, as you mentioned, but for me, what matters more is not that people are citing my work, but they are able to read and get some information.

And for me was really important because I’m a scholar. So I have been in academia for four or five years, is not that long and having the opportunity to reach different people. And when I look for my citations, I see that it’s from different regions of the world. That’s what matters for me most because I see that in my work and also people read that and people are interested in that. That’s why I believe more than just the number.

And, but It was, like, really surprise, to read that in a short, in such short time. I just want to thank you for everyone who have the opportunity to read. I know there are some, you know, and I think it’s really important in academia, people realize there are different types of work, work that, you know, you put a lot of effort and you believe that is the best one. And then other ones that you say, okay, let’s do this case instead. But sometimes it’s way better, than the one that you put a lot of effort into.

And I think what matters most is not what you think, but what people when reading your work and understanding and grasping this information, for me that’s what matters most.

Branislava Lovre: And final question, what can we expect in or the years to come?

Mathias-Felipe de-Lima-Santos: I think we can expect for the next two years a lot of investment on generative AI, not only from news organizations but other industries as well. We’re going to see these models becoming more powerful and bigger and doing things that we don’t expect.

We saw now, Sophia, and all the other experiments of OpenAI, making their models very strong, and then the competition trying to reach out, you know, Google with Gemini, and another level as well. So I believe that the next few years will be really important. We understand the limits of AI and how we can use that.

At the same time, I think there’s going to be a lot of people creating more awareness of the risks of AI and somehow pushing people to have AI literacy, trying to understand what AI is. It’s not only about digital literacy and how you use that knowledge, but specifically about AI – what is AI, how is this data produced?

Because now AI is a buzzword, everyone is talking about AI, generative AI, ChatGPT, but what is that? People are not aware of what it actually is. They know a word and they know that everyone is talking about it.

So that’s one thing I’m expecting. I also expect, as I mentioned before, that disinformation is going to escalate to another level. The work of fact-checkers is going to be harder. And they would not be able to do most of their work without technology. I think it’s going to be a very interesting scenario that they will need technology to mitigate the impact of technology.

I think not only fact-checking is going to be the future – we’re going to see many other ways to mitigate the impact of disinformation. We’ve been talking a lot about prebunking in the last few months, maybe the last year. I think fact-checkers will do way more prebunking, producing a lot of content that they didn’t use to do.

We see a lot of organizations also migrating to do media literacy projects. I think that’s going to be definitely something really important. And from these media literacy projects, I think we might have something similar to what happened with data journalism – the data units that were producing data stories naturally started experimenting with AI. I think the same thing is going to happen with media literacy – they’re naturally going to start experimenting with AI literacy and trying to cover these topics as well.

So we are going to see a lot of changes in the next year or two. And also how we consume information is changing, as I was mentioning. When people are not using Google anymore to get information and they are using tools with data that was trained on older information and is not updated with the frequency that Google is updated – what are the results? What is the impact on information consumption for these people? How updated will they be?

And as we are seeing, all these different platforms, tools, social media platforms, they’re trying to also bring generative AI into their scope. So maybe we’re going to see a lot of conversations on Twitter, Facebook that are pretty similar because these tools start suggesting responses and then you see everyone writing the same thing. Maybe we’re going to have a lack of vocabulary, I don’t know. But it’s something that might happen.

So, I think we should be aware, open also to understand the risks and the opportunities of these tools. But yeah, not rely on them 100%.

Branislava Lovre: Thank you for your time. It was a pleasure talking with you.

Mathias-Felipe de-Lima-Santos: Thank you for the invitation. I hope it was a really good conversation.

Branislava Lovre: You’ve watched another episode of AImpactful. Thank you and see you next week.