Use of Artificial Intelligence for online research

In this episode, I spoke to Henrik Vold, a Norwegian journalist, instructor and a tech enthusiast. Henrik has had a long career in journalism during which he's simultaneously honed his technical skills. His current passion is helping journalists explore the world of artificial intelligence and incorporate it into their work. 

He discusses the importance of journalists branching out into technical roles and learning data analysis skills. Henrik also talks about his involvement in creating tailor-made courses for organizations and his efforts to convey AI knowledge to a broader audience.

Throughout the interview, Henrik emphasizes the need for caution when using AI, especially with sensitive data, and stresses the importance of keeping information secure and maintaining a computer security mindset. He discusses the potential applications of AI, its limitations, and how it can serve as a support tool rather than a replacement for human creativity and artistry.

Henrik also shares insights into using AI for various purposes, such as transcribing interviews and generating meeting notes. He talks about iterative prompts to improve the AI's responses and mentions potential concerns about AI-generated content and copyright issues. Lastly, he discusses the evolving role of AI, the importance of news organizations in interpreting information, and how AI may influence the way people consume information in the future.

Connect with Us:

• LinkedIn: https://www.linkedin.com/showcase/integrity-insights/?viewAsMember=true
• Berlin Risk Linkedin: https://www.linkedin.com/company/berlinrisk/?viewAsMember=true
• Website: https://berlinrisk.com/

Transcript

Yeah, sure. Um, well, I'm a journalist trainer, uh, at the Norwegian Institute of, uh, of journalism right now. Um, I'm a journalist by, uh, education and by, uh, by careers. I've been, uh, a journalist for 25 years. I've been an editor for two years, been a project manager for. God knows how long. Uh, and I've had a lot of technical roles, uh, in, um, in, uh, the public broadcaster in, uh, in Norway.

Um, so basically I've been working with a lot of, uh, digital projects. Uh, but I've also been doing, uh, open source intelligence in, um, uh, in, um,  by sort of the unit that was, uh, created after him. After the outbreak of the, uh, Ukrainian war. Um, so basically, uh, that's where I'm, I'm, uh, I'm trying to learn as much as I can trying to, um, uh, convey it to as many people as I can.

Yeah, that's, uh, that's where I'm.

Yes. Yes, that's correct. Um, it's hard work. It's hard work to try and do a hundred percent job. And, uh, so yeah, I've.

I think that's, uh, just a sign of the times. Uh, journalists need to branch out. They need to, to keep learning, and we need more technical, uh, journalists. We need more data journalists. We need more people who can, you know, look at a spreadsheet and uh, and, uh, make sense of it. And, uh, and, um, In such a fashion that you can actually tell new stories?

Yeah, no. Um, normally I would be, uh, teaching open source intelligence, uh, and uh, and ai. Usage, safe usage of AI for journalists and, and editors is, that's my main focus at the moment. But, uh, I do do a lot of, uh, a lot of OSI training and a lot of, um, consulting for, uh, smaller newsrooms.

Oh no, it's, uh, both, both really. Um, we have, uh, newsrooms with, uh, a lot of different, uh, different needs and different, uh, different, uh, sets of, uh, skills that they want to, uh, develop. So I. It's, it's all a matter of what the, uh, what the order is. Uh, I do do generic, uh,  training, um, and I do do, uh, master classes, uh, and uh, also do really tailor made, uh, courses for organizations.

Yeah, sure. Um, mainly, uh, right now what I'm doing is, uh, is. Uh, so focusing on what's, what's safe, how, how to get started doing, using AI for, um, for, uh, journalistic purposes, uh, without actually, uh, uh, either making, um, The newsroom, uh, unsafe. Um, I mean in terms of, uh, privacy and, uh, unpublished material. Um, so I'm, I'm, I'm, I'm teaching, uh, people exactly what AI is at the moment and what it does and what it doesn't do.

And obviously what you shouldn't, uh, definitely shouldn't do. Uh, when it comes to, uh, when it comes to using any of the, uh, the, um, generative models, That's my main focus at the moment. But I'm also doing, um, we're developing a, a, a course, a two day course for, um, for using AI in journalistic research. And I mean, it's, it's an exercise in humility every time, uh, I stand up and, and try to make some sense of it to people because it happens.

Everything just changes from one day to the next. So it's, it's all a matter of trying to, trying to. Yeah, keep up with the train that's running.

I would definitely, definitely, uh, tell 'em to be curious. Uh, tell them to be curious and to experiment and to, for god's sake, keep any and all sensitive data away from, uh, from these models. But try, I. Play with, uh, with the, uh, with the AI and, um, you know, keep, um, see if you could have some, some, uh, low level projects that you, uh, that you can, uh, envision AI helping you to solve, uh, the generative models.

That is, um, I mean, just, uh, create a project and, uh, let, uh, just let your, the playful side come out. But, uh, yeah, I think you need to edit that out. That's not a very good answer. Um,

Yeah. And yeah, uh, if we're looking at, uh, if we're looking at chat, g p t, they're, uh, an open ai. They're not very forthcoming with what they train, uh, their model on. Uh, so knowing what goes into the next training set is, is obviously an issue. Um, you should always compartmentalize your, your information so that, uh, any.

Any sensitive, uh, issues such as, uh, you know, uh, privacy issues or, or contacts, any information that can be, uh, used to identify, uh, sources, just keep it the hell away from, from, uh, from AI and, well, as you would, as a normal investigative journalist, you would, you would, uh, take really good care of that information and keep it, uh, separate from, you know, your, your, your open.

Sort of open, um, computer systems. So it's, it's just basically, uh, basically just keep, uh, the computer security, um, mindset e even when using ai, or especially when using ai.

Hmm.

Well, the, i I, the, the, uh, the overruling principle I, I think would be, um, well you can use an example for this. Um, you. If you were to, uh, give chat g p t access to unpublished material, say you, uh, have a block of text that you want to, that you want to somehow work with in chat G P T, but it's not been published and it's, and it contains some, uh, some, uh, information that's not, uh, known out there.

Um, well, theoretically, if you were to share that information with chat, G P T. Uh, the information could be made, uh, available for someone else, uh, at the same time looking, uh, at the same, looking for the same information, looking for the same, uh, same field of, uh, of information. And, uh, and that obviously should, that should be, uh, as a journalist, uh, a huge no-no.

Uh, in terms of, well, you know, the, the unpublished material you should never share with anyone. Um, and, uh, yeah, we do have some products in Norway that, uh, that works in this sense, uh, that helps journalists make, um, summaries of articles, uh, Makes them, uh, helps 'em, uh, create titles, um, slugs, that kind of thing.

Um, but the over overruling principle here is, um, is to involve AI as close to publication as, uh, as possible. Uh, just to minimize the risk of.

Yep. Yes, definitely. And, uh, it is, you know, it's, uh, it's, it's coming more and more. Um, OpenAI is, is famously closed in, uh, in, uh, How they train their, uh, their, uh, chat, g p t. But there are other, other, um, also Norwegian, uh, companies, uh, creating their own, uh, gpt. Uh, we have a, a Norwegian one called, uh, nor g p t, trained on a very small dataset with Norwegian, um, Norwegian, uh, uh, language, Norwegian values, uh, that kind of thing.

Um, And that's being developed by, uh, sort of the largest, uh, Scandinavian and Nordic, um, a news organization or news, um, uh, umbrella called Ships Dead among Others. Um, so yeah, obviously, and we, we have had, um, a few, a few projects in, uh, the Verba, the, uh, the Osint, uh, unit where I've been working where, um, We do have a, uh, a tank classifier and, um, a language recognizer for use, uh, in, um, content from the war.

Uh, so it's possible, it is definitely possible, the supremely possible and, uh, something every news, every large news organization should be, should be looking into doing. But I dunno if, uh, language is, is the way to go for that.

Oh, that's a good question. That's a very good question. I would say, uh, language wise, uh, keep it away. It, it'll never replace, uh, at least not in its current state. It will never replace a journalist. Um, it can help you summarize, uh, a huge, um, Huge amounts of text, PDFs, uh, that kind of thing. Um, and g PT four has, uh, just a suite of really, really nice tools.

Um, the advanced data analysis can give you, um, a lot faster visualization of, uh, of data and of, um, Uh, really complicated data sets. It can, uh, strengthen your reporting That in that sense, uh, but this is for research purposes, not for publication purposes. I would never use it for, to, um, replace my written text at all.

It's just not good enough. It's not, uh, it's not, uh, yeah, it's, it's, it's a support. It's not a, it's not a replacement.

Oh, exactly.

That is another great question, Philip. Um, yes. The, the, the short answer is yes. Uh, with a caveat. I would always, uh, check summaries. I would always check information. I would always treat it as any other human source You wouldn't trust, uh, a single source of information, would you? You would always check. Um, it's the same thing with ai.

Uh, it, it doesn't give you, uh, , it won't solve all your problems, but it, it, it will be a, a great, uh, support for a lot of the menial tasks. Um, right now, the, the greatest things, uh, for us is transcription of, uh, of, uh, interviews, uh, and summaries of meetings. I, I do use that for, uh, nonsensitive meetings. I, uh, I transcribe, uh, transcribe the meetings, and then I, um, I upload them to G P T and then I, uh, I, I have meeting notes that I can converse with, uh, which I think is a, is a, is a great, uh, great tool, great strength, but, but again, it's uh, it's a support function.

It's not, uh, it's not part of the, the publication or the writing, if you understand.

Hmm.

Yeah. Yeah, of course. Uh, the, uh, uh, it's the same with, uh, with the Dali and the mid journey. The more you know, the more, uh, the more concise your, your, uh, prompts are, of course, the better the output, the, so, uh, the better the photographer asking for, uh, an image in mid journey, the better the output will be because they just

Have more knowledge of, of what they want. Same with journalists, uh, same with any anyone who's used to producing text and, uh, and, uh, used to asking sources for information. The more succinctly and the more precise, uh, the, um, the, uh, the question, the, the better, the more precise the, the answer. But again, With the prompts, it's, it's possible to, to improve, uh, improve the answers, uh, through this, uh, conversation that you will have with, uh, G P T four.

And it's a, it's a great thing that you're not, uh, actually being judged by the, by the interview, uh, object. You could just ask all the the dumb questions you want.

Yeah, I would, uh, again, I would, I would go ahead and, uh, just learn as much as, uh, as you can. And there are some, some great prompts out there that, uh, that will, well, I use an iterative prompt that, uh, that. Helps me develop the really, really precise, uh, prompts when I don't know what I'm looking an it, sorry, an iterative prompt, like a, um, a prompt that will improve, uh, a prompt that will improve your prompt.

It'll, um, you. You can add it as a custom instruction to your jet, G p T, but essentially it's, uh, instructing G P T to act as a prompt engineer for you. Um, and, uh, what you'll get back is, uh, is a revised prompt and, you know, 5, 6, 7 really good questions. That, uh, G P T four will use then, uh, to further, um, refine the prompt.

Uh, I did that with, uh, for my wife. She's a, she's a garlic farmer, or she wants to be a garlic farmer. Uh, long story, uh, pandemic, uh, related, but, um, she needed a business plan. So, uh, we started with this iterative prompt. Uh, she is the kind of. Uh, technological Luddite. That'll usually just, if, if something, um, isn't working, if, if the remote isn't working, it takes about three seconds until it hits the wall.

But for this, this, this, uh, this, this prompting here was amazing because it's, um, it did ask her really good questions. It felt like, uh, for her, it felt like she was talking to someone who knew what they were. I. What they were on about. So she answered the, the questions and, uh, it made her a business plan.

In seven, eight minutes, she had, uh, something concrete to, to work with, um, which is, you know, it's not anything original, but, uh, but uh, what she put in is what you get out of G P T really.

Um, I have used, uh, G P T to, to copy. The style of, uh, of comedians, uh, when asking for text. Uh, and I mean, it, it looks like they've been, uh, taking some, some form of, uh, of steps to, to stop, uh, copyrighted issues. Uh, I've had a few prompts, uh, return with, uh, I'm sorry, I can't do that for you. It's copyrighted.

Uh, this was recording, you know, translating songs and that kind of thing. Um, but. I think it's, um, it is an, a major issue. Um, it has been an issue in Norway as well that, uh, G p t was trained on, uh, books, for instance, written by Norwegian authors without their, without their knowledge. Um, but I mean, the, the, the fear of what's coming out in the other end, I think it's, um, I wouldn't say it's unfounded, but I think it's, um, I.

It's the fear is maximized. Um, you could never replace, um, a great, uh, work of fiction. Or you could never copy a great work of fiction, um, from, uh, a great author and have it sound natural and, you know, you, you lack the divine spark that, that a human will will give to any sort of, uh, any sort of, um, artistic, um, material.

So I think you could, you could probably produce quite a lot of mediocre content. Um, not probably, of course you can, you can, you can produce, uh, a lot, a lot, a lot of mediocre and pretty poor content. So I think, um, people who want to, um, keep producing mediocre content should worry. I think they're the ones who, uh, who will be, uh, Who, who will have problems in the future.

Um, but uh, yeah, the, the true artists, um, musicians, the authors, uh, anyone who produces, uh, copyrighted material that is of high standards, they have nothing to worry about. I think.

Hmm.

Although it depends on usage. I think. Uh, for anyone just, uh, wanting to test it out, just do the free version. Um, the, uh, G P T four will give you a lot more, um, a lot more options. Uh, I use the advanced data analysis. Quite a lot. Um, uh, it will allow you to upload files. Um, so I did a Q G Q G I S uh, course, uh, a couple of weeks ago, uh, which is really an old, old technology, uh, using shape files to visualize maps.

Um, and it was quite complicated. Uh, but when I uploaded the same shape file in the same, uh, the same folders in, uh, in, uh, into G P T, it actually gave me exactly the same, uh, using clear language. I could ask for the same thing in, yeah, you know, again, seven or eight minutes. Um, and I think, uh, the G P T four is for someone who wants to go somewhere with, uh, with what they do with ai.

Um, you get access to the APIs, uh, to the open AI a p i, so you can create your own tools, um, for smaller newsrooms. I don't think that's all that interesting, um, but. It does open some doors when it comes to, uh, information security. Um, yeah. Uh, the, uh, if you, if you pay for OpenAI, they, uh, the, the, the, uh, end user, uh, license agreement will, uh, will, uh, state that they do not use, uh, your material for further training.

So, and if you, if you do trust OpenAI, that will be a safer solution for your newsroom. Um, yeah.

No, the, the big issue here, I think with AI is that, uh, we've, uh, we have handed the, uh, the, um, we've handed the, the, how, how do I explain this? Um, the.

Evolution of AI has, uh, really just been, uh, handed to, uh, tech companies and, uh, developers and, uh, the social sciences are so far behind on this that, um, I would never, I would never trust any of these tech companies. They're thinking profits, they're thinking development. They're thinking evolution, revolution.

And, uh, I think a lot of, uh, privacy has, uh, really been, uh, left behind a lot of, um, there's a huge, uh, problem we are facing, uh, in the coming years because everyone's been treated as, uh, not as human beings. And, uh, training of these, uh, these, uh, language models, they've been treated as, uh, data subjects. Uh, so there's so much here.

Uh, will be unpacked in the coming years. Uh, that'll, you know, be, it will be huge issues for us as society, so I wouldn't trust.

No searching information is, uh, Google is still, uh, not, uh, a perfect product. Um, um, are you thinking of AI will replace Google? Um

hmm.

Oh, it's, uh, yeah, it's famously, uh, last words to, to try and, uh, um, predict anything, anything. But, uh, , everything is going really, really fast. The big game changer is on its way. Uh, when you have. Personal assistant, uh, you know, Alexa or CD or, uh, or hey Google. They will be, uh, when they, uh, are connected more thoroughly to, um, generative ai.

That'll be a huge difference. I think a lot of people will, who are not today using . Using any of the traditional, uh, news media will, you know, just keep getting their information some other way. And I think AI will replace a lot of that. Um, hopefully, hopefully. I think, uh, news organizations are gonna be more important in the future because there will be a lot more generative, uh, pardon my French garbage out there.

Um, so hopefully from a journalistic, uh, point of view, uh, journalism will be more important. Um, interpreting the world around us will be more important. It'll be a massive role for journalists in the coming years. Um, and I do think a lot of people will get their information from some kind of digital assistant.