Speaker 1: Hello and welcome to this video on MaxQA and its new AI functionality. Before we dive into MaxQA, I want to quickly switch to my webpage and make you aware of the blog I have been writing and I can add the link. I've been writing quite a bit about the use of AI and then ChatGPT came out and in the beginning it wasn't integrated yet into the software tools. So I was kind of playing around with what was already there. The machine learning was kind of built in already in some tools, but not the generative AI that we got this ChatGPT and the language models around that. So if you're interested in that, you can kind of read a bit how things evolved. So that is, yeah, I think the first one here I wrote in January and then in February and so on. And then I tested here two programs that then started to integrate generative AI and you think you see technology is moving so fast at the moment and I think there will be a number of other videos that come throughout this year. So it's kind of crazy how fast things are implemented right now. So in this case study, I actually compare a software that's called Canvas. It's more used in the business world and Atlas TI. And Atlas TI has implemented AI for coding and I will return to that in a second or not in a second, but at the end of the video. So let's focus on MaxQDA now and where they have implemented AI and for what kind of purpose. So when I was testing, and these were the earlier articles when it was not yet integrated, so I kind of did like manual copying and paste from things I, like for what could we use ChatGPT for qualitative data analysis. And looking at the machine learning tools that help coding, that wasn't always that great. So in addition to look at Envivo and Atlas TI, I looked at OpenAI that's also more used in the marketing context and that wasn't really satisfactory what they came up with in terms of coding. So my conclusion was coding with the machine learning capabilities, not so great because it gets stuck on the topic level and it doesn't understand context and what we actually do when we code in qualitative analysis. And then I came actually up with the idea, I tried to copy and paste coded segments I generated and plugged that into ChatGPT and let ChatGPT write summaries. And I thought, well, that would be great if we get that functionality included into software. And exactly that has been happening now in MaxQDA. And I've been testing it. I can only say congratulations, MaxQDA team, you have done a great job. And I think it's gonna be wonderful and helping people along in their analysis. Because often what you do anyway, I mean, if you work through data, you start summarizing what you have been coding. And that means you have to read all of it and then write your own text. And it's so much easier if somebody else writes a text for you. You can still adapt it and change it and modify it, but at least you have a base text. And I think those of you who work with ChatGPT already, well, that's how I use it all the time. So now let's look how MaxQDA is doing it. Let's start out with the summary tool, because that's where it's all kind of built in, in the software, helping people to kind of see how do I get from A to B to Z. So what actually I now do when I have coded some data. And that's, yeah, if you're familiar with MaxQDA, you know the summary grid. So what we have here, we have comments on a parenting blog. And that was about the topic, summarizing this article that was published in Psychologist. And then this person back in kind of summarized the gist of the article so many years ago without ChatGPT. And then her readers were commenting on that. So that's what this data is all about. And the other document are comments on an article in the New York Times magazine also on this topic. So now we can kind of compare what did people write on the blog and what did people write in the New York Times, if you just focus now on these two topics. And so what you can do, well, if you collapse a table, you can just write the summaries on the aggregated code, yeah, so not the aggregated code, but on all the content that is here in the subcode. So it's all then in aggregated view. And I already did that. And we can also do some new writing here. So this is the summary here, and that is that AI button. So when you first want to use it, you have to sign in. If you don't have a MacStudio account yet, you will ask to create one, you have to enter your license key, and then it will be checked whether you eligible to use that AI tool. And if you have a yearly license subscription, then you can use it. And what I've read on the website, if your university has a license, you might have to wait until your university is, whatever the university needs to do. But it's all on the website. So I've already done that, signed in, my license was eligible, and I could start using the tool. So I click here on that tool, you can also do it multiple times, or you can pick the language. This text is in English, you can get a standard summary, a short summary, or text in bullet points. And we will see that, yeah, sometimes the standard or the shorter summary is the same. So it all depends also on the length of your quotation. So what I have done here, so that's the negative experience here on the Belkin blog. So my parents experienced negative emotions and stress related to the sacrifices they make for children and so on. I don't think it's so necessary now to read all what has been coming out. But you see, this is kind of the gist of what is Sunrise. And if you do want a bit more detail, well, you can also get the bullet points. So the bullet points kind of summarize the main points that are in the coded data segment. So I like actually this combination. So you get the summary, and then you can remind yourself more, okay, you know, what's actually in more detail in the coded data segments. And of course, at this point, you have coded the data yourself, you familiar with the data. And if you read the summary, then you also know that, yeah, whether it makes sense or not, or whether the summary tool has, you know, the AI tool has kind of messed up something here. Yeah. And then, you know, I can, I've done that here also on the articles from the comments on the New York Time magazine. And now I have a summary very quickly done, and I compare the two groups. So let's move on to the, yeah, the subcodes. And if there is not that much data, you see this rather short sentences here that I've coded. And this is only a small sample project. So if you have 10, 15 interviews, you probably have more data to code. But if it's too short, you cannot get a summary. But then of course, while we have the four sentences, I can quickly write the summary as well. Here I have also only three segments, but yeah, it was able to generate the summary. So let's now try, yeah, I have the summary here. Let's do the bullet points. I have to wait a few seconds, but I think it's doable. It's that sometimes, this is actually rather long now, when I created all the examples before it was going quicker. So it's, you know, it's so, it always depends. So it gives me more detail what's behind the full summary. And I think that combination is nice, that I get a summary, but also the bullet points. Yeah. And then you could, yeah, go through that and, and do that for the, for the subcodes where it's suitable, or you only decide to do that on the aggregated level. So that's also a possibility, of course. So it all depends on your project and your coding. So let's look at the, that subcode here takes lots of energy. I have three quotations. And if I double click here on my code memo, then you see, I've already done it here. Then I do get a summary. And we had seen it also before, if the quotations are not just, or the code data segments are not just very short sentences, then you do will get a summary. And again, if you want a bit more detail, you can add the bullets in addition. I'm not sure whether I have done it here, but we can, we can try it. I tested it out before, whether, and I guess that's the case, because if it, you have very, you know, it's not lots of data, then the standard and the short summary is actually the same. Yeah, I can't pull that up here, but yeah, maybe I can copy and paste. Standard summary here. Yeah, that requires a lot of energy. So, that's kind of the same. If, well, we can try this on, yeah, let's, let's look, take one that, wrong click, take, so, that has 15, 15 data segments, and we can also check how long they are. So here's the retrieved data behind it. So we have 15, yeah, shorter and longer code data segments. Yeah, so let's create the shorter summary first. And then the standard summary. Here we have the short summary and here the standard summary, then we get some more information. So what we see now here is something I tried out earlier. And so I've created here some summaries, and that was, it's 1970, well, 37. I created a long summary and another one at earlier at 16 past seven o'clock. And if you compare the text of the summary, it is exactly the same. That's basically all what, yeah, what you need to know or how it works. Yeah, it works at these two places at the moment, summary grid and also for your code memos. I have just bumped into my limit, so I cannot play around until another five hours. So it has to reload and I think it might still be the beta phase and maybe at some point you can also buy extra credits because it does cost some money to use the tool. Yeah, so let me return to what I said before and what was in Andre's video that's in German. So if you don't understand German, you cannot follow it. But what he was saying at the end that he was teaching a workshop and people were asking for that when he didn't know yet that this would be coming out. People were asking or wishing for that the software could take over the coding. I can only warmly, warmly recommend, read my article, read my tests and I don't think it has anything to do with the implementation. I think at this point in time, also if MaxQDA would implement it, it wouldn't work, the coding. I think this is wonderful. That's a wonderful support that we get with AI summarizing coded data segment. But for coding, it doesn't work and I tell you why. Because it kind of looks at one paragraph at a time and then it codes that. And then it looks at the next paragraph at a time and it codes that. And oftentimes it applies two, three, four or more codes than that. And then very quickly you have a few hundred codes. I coded 50 interviews and I got over 4,000 codes. And then you spend hours and hours of cleaning that up. And also the tool tried to categorize something but the whole point, it cannot aggregate, it cannot understand. And if you have interview data, it cannot understand if something was said here and something was said down here, that it's very similar and maybe you just don't invent a new code, you just apply the same code again. Sometimes it does apply the same code, but often not. That's why you end up with so many codes that then you have to clean up. And if you haven't done the coding yourself, well, the cleaning up means you have to read the data. You can't just play with the code labels, you have to know what's behind it. And that's why it doesn't take any work away from you. So don't keep your hopes up that, at least at this point in time, I'm not sure where we are in a year from now. But right now JTTP or the language models behind it is not able to aggregate on the content level. It might be not aggregating on the code word, the code label level that it kind of categorizes, these are all emotions. But it cannot interpret the context. The context within a small area, but not the context within an interview or across an interview. And in that article where I also looked at Canvas AI, that is a software that only deals with open-ended question form survey. And their coding is not done by generative AI. They have developed their own proprietary system for 10 years already based on machine learning, deep learning, whatever is behind it. But it's not generative AI. And that's why their classification works quite well, but also not on interview data. It's also based, it's working on very structured qualitative data. So keep that in mind and enjoy the functionality you have here. And I hope Atlas listens and they will also integrate the summary function also in Atlas AI. There's much about this new AI tool. And go play with it, enjoy and reap the benefits. Have you enjoyed the video? Are you hungry for more on all things qualitative? Then hit that subscribe button to stay updated on our latest content. And if you're eager to take your knowledge to the next level, get in touch. You will find the contact details in the comment field below. And don't forget, hit that subscribe button.