Investigative journalists and AI: true friends?🕵🏻
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!
Before leaving you to read the interview, I will share some extremely practical production notes that helped me a lot in the creation of this newsletter.
The interview took place online, on Zoom, a tool that provided the recording of the call in audio and video format.
I used an AI transcription tool to obtain the raw transcript of the interview.
The document had a number of inaccuracies due to the natural conversation between two people: i.e. interjections, repetitions and dialogue far removed from a simple repartee.
I asked ChatGPT Plus to fix the interview by removing repeated words such as 'um', 'uh' and to format the document for a more comprehensible reading of the text.
This process from transcription to first formatting took:
• 1 minute for transcription
• 10 minutes of brainstorming on ChatGPT Plus
In 11 minutes I had a very good draft of the interview in front of me and I was relieved because, of all the activities I have always detested most in my work, transcription definitely wins first prize.
I spent another 25 minutes sorting out the interview to make it readable and another 10 to layout it on Substack. I also asked ChatGPT to suggest a SEO friendly summary to help me share the content. This saved me at least 15 minutes.
Interesting, isn't it? How do you use AI in your daily work?
I will now leave you to read this month's guest of the month, Jaemark Tordecilla, Nieman '24 Fellow at Harvard University, who has created a chatbot that could be very helful in investigative journalism.
Jaemark, thank you for accepting MyType's invitation. Before we get into the interview, tell us something about yourself.
Yes, sure! Right now I am a Nieman fellow at Harvard University. I'm doing my fellowship here. I didn't really intend to study AI directly, but I was just drawn to it because there have been many, you know, people purporting to be AI experts in the journalism field but all they talk about is the high-level stuff, and we're not seeing any solutions for actual problems in the newsroom. So, I figured there needed to be more people who are trying out solutions that journalists could solve.
Before telling your experience and your project, can you briefly describe the problem that you're trying to solve with your project?
Coming into this, there was this one particular pain point that I wanted to try to solve, which was parsing large volumes of documents and helping journalists try to get through them. In the Philippine context, there are these documents called audit reports, which are government audit reports of each agency and local government. Each reporter is tasked with going through this mountain of documents to figure out if there's a story there. And so they spend hours even just trying to figure out whether they have to dive deeper into the report. The problem is that, usually, people who cover the beat for this agency also cover other beats. So it's a big investment of time to find stories in these documents. But then there's also a payoff because sometimes there are significant anomalies that could be uncovered if you really find the right report and the right story from it. So, I wanted to do a test case for this kind of document about whether a GPT tool could help uncover that information from this type of document. I was in an introduction to AI class, and the professor told us that everything she was teaching us would be obsolete because ChatGPT had released a new feature called Custom GPTs that would allow you to build applications without any coding. So I tried it out, and then I've been thinking about the problem of the documents for a while trying to figure out what implementation I could use. I was thinking about machine learning and other tools. Then I thought about whether I could use Custom GPTs for it, and all in all, it took me maybe a couple of days, and I was able to develop a solution.
How many days you try and maybe what you try for, for became to the final solution?
It took me about 16 hours of work but only because it was my first time approaching this type of bot. In general, once you understand how it works in two to four hours you can create custom models on GPT.
What were the main technical problems you encountered during the development of your project?
First of all, you have to subscribe to ChatGPT Plus to be able to use custom bots to process documents. There are also limitations even if you use the paid version; you're limited to the number of queries you can make in a day, something like 40 questions every two hours, I think, or maybe less than that. I also noticed that ChatGPT has problems dealing with very large documents. So if you give it 100 pages, it's going to perform worse than if you gave it a 25-page document. I don't know the limit, but that's the observation I've had. I've shown it to other Filipino journalists, and they were asking me if we can use the bot to process multiple documents, like an audit report from this year and last year to get observations. I've tried it, and ChatGPT doesn't handle multiple documents well. So those are technical limitations. I mean, it’s possible to upload documents, but it's just the answer degrades because it's a problem elsewhere. I think once ChatGPT finds the answer in one of the documents, it won't process the other document anymore. But that's what I found so far.
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!
What are the main challenges or lessons you learned from using this technology in your project?
Of course. The first is we need to play around with this technology to find how it's going to work for us.
And we have so many challenges, and we need all the help we can get. So think of AI as a tool, maybe like Excel or your word processor, to make your jobs easier. We need to figure out ways to help AI make our jobs easier. The second one is that there are tasks that are good for AI and that there are tasks that are not. And so what I found in this experience is that it's not going to replace our critical thinking. Generative AI it's good for summaries. And so it's going to be more important for journalists to develop that critical thinking facility because that's the value that they bring into the process. The writing maybe can be replaced a little bit by and AI Grammar Checking. And that might go away, but what won't go away is the nose for reporting of a good reporter, the nose for journalism. And so that's the important part.
What advice would you give to a journalist who wants to start studying this field as well?
I think that it's an exciting time to get into these technologies right now because there are no experts, except maybe the people who are developing chatbots directly or doing hardcore AI research. But in the journalism field, there are no experts. So there's an opportunity for people getting into it now to really learn how it works and to find use cases that fits in their own newsrooms. And to figure out, like I said earlier, how to use these technologies to make the lives of journalists easier. In the previous generation, we were behind a lot of the bad actors when it came to publishing on social media and using emerging platforms. And so we had a lot of problems with misinformation and losing our audiences to influencers. And so we can't afford to do that with this new generation of technology. We have to be good at this. Because if all the hype is true and this is going to transform not just our industry but every other industry, we need to be ahead of the curve and figure out how it's going to affect us, how it's going to help us, and what are we going to guard against? And it's particularly important because there's a lot of fear about these technologies and if you don't understand it, then how are you going to cover it as a journalist effectively? And so there's a lot of opportunity to develop expertise in every newsroom out there.
Big announcement!
We are very happy to present you with the new MyType website, which tells you all about the specifics of our innovative CMS!
We really like it but we are looking for feedback, let us know what you think!
There would be no 'digital joynalism' without sharing. If you like this newsletter you could share it with colleagues, friends and family :)
For any ideas, suggestions and thoughts on the subject, don't hesitate to email me at sforni@atex.com.
Have a good weekend,
Sara