Hi and welcome again for another episode of the Data Strategy Gurus podcast. Today again, we have Angelica Clas with us. I mean, Angelica, it's been a long time. Last time it was more exotic—we were in Las Vegas still talking about data—but today we will, of course, have it a bit more about AI and demystifying AI literacy. Can you maybe elaborate a bit more around AI literacy and how you see that?
Well, when it all started about five years ago—or six years ago, maybe even—we had the amazing marketing data literacy. And from there on, now we started working on new programs called data literacy. And then suddenly, you know, when OpenAI launched its large language model or its system, um, in—I think it was on the 30th of November 2022—and since then, AI has been rising to the max. And everybody is thinking, "It's new, it's new," but yeah, unfortunately, it's not new. It's already very old. And then suddenly, the term AI literacy came up.
And I know in the world, there will be different views on it, but from my perspective, I don't think AI is a very good term to use.
And you mean AI is not a good term to use in the sense of artificial intelligence? In the sense of AI, I know that everywhere you look at a product, the SaaS product, we have AI on top of that, where I think—well, I didn't mean AI as a technology or a methodology on its own, but I mean the term AI literacy suddenly came up, and it made me wonder. So if you hear AI literacy, what do you think about?
I mean, we had data literacy there. I could feel already—it's data fluency. It's understanding the data, looking at the data, working with the data. Do you think they mean with AI literacy it's more as well of understanding how AI works or working with AI? But for me, it's very broad because AI is pretty complex. You have neural networks, you have the large language models, you have a lot of infrastructure. It's a pretty complex combination of what really is artificial intelligence as such, and it's a subdomain, in fact, of AI.
But I feel like everybody is throwing AI—if a product uses a large language model, which, in my feeling, is a part of the solution, but it's not the holy grail as such. There is a big promise in what AI can do as such. So what do you think that—well, most people or the marketing people, let's call them like that—mean when they talk about AI literacy?
Well, to be honest, some people say it's the holistic view of what AI can do for your company. We see many different definitions of AI literacy around the world when you look at CA or DataCamp or even other authors of books and so on. I don't believe there are many books around the topic AI literacy just yet, but I think it's like a marketing term that is used to address, "Oh, hey, I know something about AI."
Well, to be honest, I know the concepts of AI. I know what I can do with AI. I can think of pretty good ideas what to do, but my colleagues at the office of the business data challenges—they are much better in actually making the things work right. And I—well, yeah, AI literacy makes me sometimes a little—hmm—or it makes me a little anxious—or not anxious is not the correct word—but a little bit upset because when you then learn about those books or those websites, and then the next sentence is, "Okay, but when you want to become AI literate, you need to be data..."
Literacy. And then I started thinking, what? So from my opinion, um, I think everything is fueled by data, and that's the basis of everything. So I think data literacy, with a subset of being, um, uh, more convenient in working with AI solutions like the large language models like Geni or other solutions like, uh, um, using AI methodologies to develop new products and so on. Um, I think that's a better term.
And my colleagues, you know, they last week—I think it's last week or the week before that—they, uh, they were talking about AI in an article, I believe, uh, in our WhatsApp group. And, and my colleague HTI, uh, Thisel, uh, and I think she is really brilliant when it comes to AI solutions and thinking about, uh, how to develop them. Um, she said maybe it's not literacy, but maybe it's an option that should be part of the description of it. It should be AI adoption and in the broadest form because statistics is also, you know, something like it comes from working where they are. We get results. We need to be able to read statistical information, um, but that's also a part of data literacy.
So yeah, adoption. But adoption is that you already know what's what to do with it, what it can do, and how to do it. Uh, I think that that's already where you are in that phase. A literacy points more to where awareness of really understanding, uh, what can we do with it? How can we build it? Uh, what are the implications? What are the risks? Um, all of—all the ethical discussions, all the governance around it—these are important parts that you need to be aware of if just jumping on AI.
I mean, if you take the Microsoft Azure platform, you have a lot of cognitive services. The same for AWS where you can build these AI solutions. But if you're not looking at privacy, security, uh, putting company information into the cloud, that's a big risk. And I think there is an important educational part. It's not because people are stupid, but educational of helping them understand of what is possible with the solutions. And yeah, finding the right term around that.
I think adoption is a nice direction. I had recently a discussion on data literacy where they say it feels like, yeah, people can't read, and you're teaching them to read. And if you say, "Well, we're going to help you become data literate," they feel a bit offended because it feels like they don't know nothing, uh, but in fact, they're smart people as well. They just are in a new type of domain—data management—where they need to look at the data in a certain way, like you said, how we do that with business intelligence solutions as well. You get a report, uh, you learn how to interpret it, uh, the same with, um, with statistics, that you understand what these numbers mean. And this is the bigger picture. I think the same for AI literacy as well.
Yeah, yeah. But I, I don't like to use the term anyway, um, but sometimes from a marketing perspective because you want to sell your trainings, your products, uh, and so on, or your services. Um, I think, uh, from that point of view, it's okay to use the term data and AI literacy or AI literacy, but, uh, I think the definition should be, uh, something else.
Because now we had the discussion—I had the discussion with, um, Donald Farmer as well. You know, he's one of my heroes, and he, he also said, "Do you think I'm dumb when you say you need to be literate?" And I think, yeah, you have a point, uh, but as, as you can't be driven by data because you have to do something with it, um, uh, so data-driven is not a good word. AI-driven is maybe not a good word, or AI literacy is not a good word. Even data literacy is...
Maybe not a good word, but we use it more from a marketing perspective, right? And as long as we are able—and I know you do the same as I do—as long as we try to tell the stories right and help people to understand how to work with data, how to read graphs, how to build them, how to use them, on trustworthiness in your data pipelines, build good data products that can help at the end develop also beautiful AI solutions, I think we're on a good way. As long as we, as the professionals, tell the story right, it's more marketing, I think.
Yeah, that's the same feeling I got. I mean, if we're discussing amongst the professionals—people really in the domain of AI, data—I see everybody has the same concerns. They're addressing the concerns; we're talking about it. So already, the people really in the scene are working around these topics and moving forward. We know there is a bad part on all the big, amazing solutions as well, but it's the same in business. I mean, there is a dark as well, and that's the same with technology—that it's being used for the wrong reasons.
But yeah, this is a part of what we have to do instead of only building the solutions: making people aware and sharing our journeys as well, where we fail and how to do that. I mean, I'm testing out large language models right now to help me in development, and I thought, "This is going great, fantastic—two days, almost there." And then you see it starts drifting off, and you say, "Hey, what's happening?" and you mess up the complete code.
What I've learned is I have to go back to the practices I've learned as well back in the days: be precise, be consistent, don't try to develop everything at the same time, do it unit by unit, step by step, don't go too fast, just be patient, do the testing—just the main principles of software engineering you have to apply. So by using the large language models and seeing how immature they are still in helping you in doing code development, it helps me become a better programmer at the same time.
It's the same if you're writing marketing material—you learn how to formulate things in a better way thanks to the large language models. But like you said in a previous discussion, if you feed it bad information, this is what you get out of it: you get bad AI solutions, bad AI products, bad AI results as well. So I think, yeah, getting those results right is something we still have to learn.
Like, prompting isn't the new art, right? I believe that you can ask simple questions, but when your question is more concise and more accurate, you get amazing answers. But therefore, you have to learn how to craft a good prompt in this case. And if you don't—I remember we had a discussion—do you remember that you sent me that picture of the man? It took you some time before it was something that you liked at that moment or that you wanted at the moment, and that your prompts needed to be concise and more precise and then corrected.
But even when you work with tools like ChatGPT and you ask for a picture with D, sometimes it just doesn't do what you want it to do. And even though your prompt—when you read back to your prompt—you think, "But I was concise, I was precise, I did exactly what I needed to do," but still, the image is not what I want. So yeah, it's still fed with things.
But I also believe, you know, there is a lot of information fed into OpenAI or in ChatGPT.
Here’s the formatted transcript with punctuation, capitalization, and paragraph breaks:
---
Or in the other large language models, and if the information that we add—the questions that we add, or the pictures that we ask—are not precise, not concise, and they're actually rubbish, then you get the same phenomenon like with data management: it's garbage in and garbage out at the end. Yeah, it's the same in business. I think if you can get bad products, you get bad results in the same way.
But we're not so aware. We think software is the golden egg, AI is the golden egg, and it solves all the problems for us. But it's still we human beings at the steering wheel to make them work in the way that we want it. And I always love these evaluations of the large language models where, for example, they're testing GPT-40 mini, and then they just ask the same question. But the GPT-40 has more reasoning already built in than the many things, and there you see exactly what you say: if you are concise, complete, and provide a lot of context, then you get the right results.
But it's the same if we have a human discussion, and you ask your assistant to do something. She already knows—or he already knows—your schedule, already knows the business context. So there is a lot of implicit knowledge available, and based on that, you make the right decision on acting. And this is still lacking in a lot of AI solutions. With progressing, I see a lot of types of approaches on getting there. RAG is one of these solutions, but there are other solutions as well to give that context to these AI solutions and get better results in such a way.
So for me, exactly, becoming literate is great in every technology because there is still a lot of lack and disconnect between talking between the techies and the business people and understanding each other—what you want to achieve and why sometimes things take time or are pretty difficult to achieve. So, think exactly: communication is still pretty important.
We're coming up to the end of our podcast again, Angelica. Time flies when I'm talking to you. So, do you have some takeaways from our discussion on becoming AI-literate?
In Dutch, we say, "Start with AI, but don't get fixated on AI." I think from that point of view, start also at the beginning. You know, people—organizations—suddenly want to start with AI while they still do not have their BI in place. So, find that balance again.
But also, do not forget that AI is also fueled by data. So you have to think of data management—that's the third takeaway. It has everything to do with AI, everything to do with data governance, data management, data quality, and so on. It's a chain, and when something at the beginning is broken or not entered in a correct way, at the end, we will have issues. So, garbage in, garbage out.
And the fourth one, I think, is that—and that was the biggest lesson I've ever learned at the beginning in 2015 when we started working on AI solutions—AI will do exactly what you ask it to do or what you train it to do. And it's actually, and what we do on all those OpenAI tools or whatever that are around the world nowadays, it learns from us by our doing, our behavior. And I think those are the points that we need to take into account in the things that we do when it comes to AI.
Yeah, I think these are marvelous takeaways and ideas which we have to nurture and keep with us to make the best of AI. Angelica, it was once again great to have you on the show, and let's go for the AI.
---
(Note: The phrase "20154" appears to be a typo and was preserved as-is. If it was meant to be "2015," that correction would align with the context.)
Here’s the formatted transcript with punctuation, capitalization, and paragraph breaks:
Solutions, thank you. Thanks again for your time.
If you're welcome.
