AI and the Erosion of Judgement

Speaker A: Wonder about the future and how you'll be working and learning. Welcome to you in 2042, the future of Work, with your host, Danielle Wallace.

Speaker B: Hello, and welcome to you in 2042, the future of Work. Joining me today is Jelena Marzanovic.

Speaker C: Hi, Danielle. Thanks for having me. So happy to be on this podcast and talk about these extremely important topics. As you know, I work at the intersection of, uh, corporate education, training, and AI coaching and adoption, and I've been in learning and development for more than 15 years. And so it's extremely exciting to be able today to blend what we know about learning with AI and how to implement it in the workplace. And this is the field where I come from, people, human side of implementing AI, helping people design workflows around it, implement it, and learn it in the way that really sticks for their teams.

Speaker B: So, Jelena, with that as a background, I'm wondering if you're seeing some of the things that I'm seeing. What is becoming so apparent as I connect the dots on things is this idea of cognitive decay. This was brought as a loose thread from conference. I was in Europe, uh, recently and then came to bear with other podcasts. It came to bear with my own examples. It came to bear with many of the executives I spoke with around the world. This cognitive decay feels like it's an emerging unknown issue as we are getting more reliant on AI. One of the speakers brought up the point that it's easy by human nature to want to do less effort. Like, hey, if I can not think, why not not think? And then the same fold, our increasing reliance then on generative AI to offload. That thinking is coming to bear. So on one hand, I had speakers saying, watch out for this. You're taking the easy way out. And then I was at a different conference the next day in a different continent, and I had the fascinating conversation with some really, really smart executive leaders who were at the forefront of adopting AI in their own work and their own workforce. And the common thinking then that he had me is like, oh, Danielle, Danielle, this is how I'm using and getting great results. I would use my Claude, whatever their LLM of choice, to be act like a McKinsey consultant. These are the parameters, you know, what should I do? And it struck me right then, Jelena, uh, that it was the exact same thing that the experts were speaking with, uh, at the conference in Europe was exactly what this guy was doing. He's offloading his thinking and inadvertently taking orders from artificial intelligence. So it's this weird Weird shift. And these are brilliant people. Everybody is smart and brilliant and capable. And it made me aware. There's perhaps this emerging thread that's coming that we're not yet aware of, but is already here. What are you seeing in your parts of the world?

Speaker C: Yeah, that's a really, really valid, good point. And actually backed by research, I'm pretty sure everybody who listens to this and yourself, you've seen that study from MIT that says, hey, people actually like cognitive engagement decays, uh, when using AI, generative AI specific, which makes little sense. Like we as humans, we are wired to look for shortcuts and we want to have shortcuts. This is like neuroplasticity behind it. We want to do things in the least effort ways. And then as background for that is people that you've described like decision makers, leaders, people who have to deal with constant pressure of decision making and spending a lot of time on mental tasks. I'm pretty sure this is like the easiest way to get that shortcut because AI output sounds really really good. And ah, that is the variable that makes it so easy to fall into that temptation of okay, I'm just going to do what AI said is because it does sound good. The reason it's that good, because it's well made, there's confidence in the output. But also because as you said, we literally delegated our judgment and our thinking, we offloaded our thinking. So not only are we seeing something that sounds plausible, sounds really really good, but we are not even engaging with it to review it and to uh, like find flaws in it and to debate with it. And the problem with that and what I'm seeing is exactly the same, especially with higher like Senior functions like CEO, CEOs, executives, et cetera, because they crave, normally crave that efficiency gain that like oh, if only somebody would tell me what good is and confirm because we are constantly with the anxiety. Well, are my decisions good? Are they appropriate? There's a lot of pressure there. So psychologically it makes sense. Like oh wow, this looks like a magic pill. AI is a magic pill. Which, yeah, it's totally understandable why people will think that. And I've spoken a lot about when you mentioned reliance, I've spoken a lot about the over something I call over reliance curve, which is a pattern I've seen with all the clients I've worked with. Most people that use AI that I've talked to, we start by not using it. We start with simple use cases. Then we realize, oh wow, this thing is powerful. I Can save time. It sounds good. It seems extremely magical. We fall in love with what it can do for us and then we reach a point where the quality of the output just starts to be off, like it doesn't make sense anymore. It's not good. We didn't check it properly. The more critical we are in general, the more we notice that there's flaws. And so this is a, uh, crossroads. Some people from there they either go the road of like, this is not working or they abandoned using it. Other people, they don't necessarily notice and they just continue with the flawed use. But the third group of people, which I really like and find fascinating, they actually start questioning the output, adapting and then learning from it. One example of that is email writing. So I have clients who have actually switched from manual writing emails to Jenny I helping them write emails back to manual writing emails m in a more efficient way, in a better quality way because they've learned, they observed what AI does to their writing, they caught the patterns, and they now find it even easier and faster to apply what they've learned from ChatGPT or whatever geni they're using into the email writing. So it really depends on how you interact cognitively with the tool. There's different outcomes that you can get

Speaker B: and all of that is incumbent upon the individual in those examples actually putting the effort to learn. Uh, I echo you in seeing those three, the three different paths one could take. And even with that third or maybe a fourth option is just the continual. The fourth option means that the fighting or the continual battle, the whirlwind of trying to fix the AIs. If it was a person that you could coach, maybe I fell into that, into that pattern and the output ends up not output is never as good as if I had actually maybe just stopped earlier on and just exercise the critical thinking, the judgment exact to unpack what's needed in that case. I find it interesting that those who not the option one of abandoning, but those who are in the option of less use actually of a large language model might actually be better exercising their critical thinking because they haven't yet abdicated that responsibility. And then within those four examples, Jelena, uh, those four paths, you also bring up the idea of those who are going down this road of polished fluff but might not realize it. Maybe we can share a clear example from one of our, uh, global group we were working with which was creating clearly through a large language model, no problem creating content that's in the danger for a training program. Not only was it Content, but it included the instructional methodologies, which also becomes part of the danger. Use that as one example, uh, many others where people creating processes which, again, is content. And then, you know, the connections between them, which is that context that's needed and this throughput in that is, unless we have that clear critical thinking now to judge that, how are we to know? How are we to know? It actually doesn't make sense. So for you, you could take a look at that and you'd be like, oh, my gosh, this instruction methodologies are off. Like, this doesn't make any sense. The sequencing's off, the context gone. This just doesn't make sense. What's interesting, though, is we're not speaking about the standard hallucinations. We're not speaking about the standard content level. Oh, the content's wrong. These are all examples from really smart people, really smart people who do know some. And it's this extra critical judgment layer to layer in and go, wait a sec, I need to really work my brain now. It isn't just content. Hey, the content's wrong because it's so polished. It's like, I really need to work my brain and think about, does this actually work in my work setting? Are these connections correct? Is this the best way of doing this? Or am I just succumbing to this polish fluff, which is at a very high level of polish and a high level of fluff, but it isn't meeting the end goal needs? Are you seeing things like this?

Speaker C: Yeah. And I, uh, do want to build upon that specific example that you mentioned, because if you go back and kind of deconstruct how we ended up with that example, what we did is we looked at it and at first glance, we didn't notice all the issues and gaps that you mentioned. Right. So you and I had to go through that file, think about it, go back to the beginning. There was something awful. We had to deconstruct it. So it's not obvious even to what I'm trying to say. There could be domain expertise issue here, which, which we talk about a lot when we talk about, uh, critical thinking in AI. Like, domain expertise is important in order to understand if the output is actually appropriate and quality. But even with domain expertise, it takes effort to go in and to analyze it. The reason why we did it or why somebody else of similar attitudes towards AI or experience could do it is because, first of all, there's something that's off. Uh, and second, because you don't take it at face value. So that's a habit. It's a habit that we need to form and it goes against what we talked about in the beginning when you mentioned of the shortcut that we prefer to take because it's easier Now. I wanted to add another layer to that specific example. I do believe that, uh, that issue that you mentioned is much older than AI, so it goes much, much back in history. Unfortunately, it may not always be clear what good looks like in instructional design in learning and development. So we've faced this issue even before AI was there. As we know, learning and development standards may not always be clear what good means, what quality is, may not always to people who are not from that domain. And so that kind of blends with the professional judgment and critical thinking and everything that goes with the now this AI situation. So it only highlights it further, like to shed additional light to the fact that there is a segmented area or segmented group of people who know what good is in learning and development, what effective learning is, what learning objectives, performance objectives need to accomplish, etc. Etc. And if you don't know that, then it's very hard to understand if what AI is giving you as output is good or not. One example from my work with people, and I'm going to speak to an individual so somebody who wanted to monetize a specific product, they have, they thought of a course and they asked Genai to create that course. And they were fascinated with the output because it gave me quiz questions. They said it gave me a good structure, the quiz questions were good. And of course, even without seeing it, I know that Genai does not have the expertise. It's still not there to make good instructional design. So we've seen it again, it's like if you don't know what good is, it's very hard to judge the quality.

Speaker B: I so hear that. And then not going what good is? Takes two paths. So on the one path, for that example that we were sharing, the course writer does know some things. So it becomes increasingly easy to take it at first glance. Oh good, I got objectives. I know. Do you write objectives? Oh, I've got some interaction because I know I should be doing interaction, but without the expertise or the depth of experience to realize what the content throughputs needed and cohesion that's needed and the fact that the activities actually don't make any sense at all because when we use a large language model, it'll input what's known on the Internet but then take it to the other. But at least as somebody who has some context um, take us through to the other example for um, your example of the course creator or I see this time and time and time again with the many organizations that we work with where a subject matter expert will be, hey, we need a course. We need it. They'll talk to their learning development team. We need a course. Can you have it done by next week? As if it's like here you go, Dutton dusted. You know, final like I type in a prompt, voila, I get a course. Because that's what they can do. They can type in a prompt, voila, they get a course, they get some PowerPoints with quiz questions and like wow, look at me, I'm a course grader now. Because in their mind that's what a course is. You've got some PowerPoints, some e learning whatever with some quiz questions. And not only that, it's exacerbated and re emphasized over and over and over and over again on the Internet. That's what a course is. It's quiz questions. So we're ending up in this really odd world where it's very easy to create courses.

Speaker C: Mhm.

Speaker B: I don't think those courses should be created in the first place personally. But this over ease of creating courses ends up in this odd proliferation of an overabundance of courses where my fear is they're all are not not only non structurally sound, but not meeting a business need. So what's actually the point? I'd rather not take any course and take yet another course with a quiz at the end of it and on some other random adjacency if there's a topic that actually has no bearing on my work. So I'm worried that it's diluting what is an effective course and putting that into a smaller corner. Because it is so easy in an instant to create a course. It's this interesting time and connecting back to what you said without the discernment to realize, uh, wait, what is this for? The goal was never training. The goal is never training. The goal is actually driving performance.

Speaker C: Again, going back to that is the issue that we've always seen. Now it's just amplified and yeah, it's exponentially more problematic because there's easy access to course creation. So as you said, anybody, anybody can create a course and not know if it's good or not. The quality decreases because there are so many. And the solution to that is probably unrealistic, but solution would be in an ideal world, learning and development or training would be a strategic point or would be involved in strategy creation. Not as a, um, punitive or as a solution to a specific issue. If organizations looked at training as something that has strategic value and something that needs to be measured as they treat other instruments and other interventions that exist in an organizational level, that thing would probably not be so problematic because quality would need to be more apparent. And so because it's not at the moment, there's no need for quality to be so apparent as we know what quality stands for, right, that people will be satisfied. We have a course, it looks good, it has a quiz, and we can know if people took it or not. And again, we just continue the same issue that we've had for two, three decades already in learning and development.

Speaker B: And I'm seeing organizations do that. They are the solution. So I wrote about them in the talent development at work piece I did in partnering with the business, um, and then some future podcast guests also echo that. So the solution is fulsome. The solution is being that partner. It is being that strategic learning strategy, just as you said, Yelena. And then the whole department then becomes this key enabler to what the organization. So it's not creating training. The organization is supported by learning development. That's enabling the goals to be met. And from there, what these organizations are doing is, uh, two fronts. So besides being that strategic partner, the two front is exercising their critical thinking. So within learning development, fully exercising it. So not ending up in this world this weird polished fluff that we're speaking about. Even on that not, not even the overtly bad courses, but on the somewhat bad courses. So they are already doing lots to deliberately keep, uh, challenging their thinking and not to just complete cognitive offloading. And the second thing is furthering that is helping their organizations do that. So helping the whole organization realize, okay, that's a skill that's needed. We actually need to avoid this cognitive decay. What can we do? Oh, well, we're a learning development function. We actually, there's actually lots we can do to support that. So they're actively doing that. And then the thirdly when there is, because there is obviously a lot we can do with content automation, training automation, speeding up work, workflows and efficiencies. So yes, these organizations are embracing that, although they're really ensuring it's these organizations that I'm speaking with, they're ensuring it's done by their learning and development team, so they have visibility onto what subject matter experts are creating on their own to help avoid that dilution risk that I spoke of. So there's some, I'm seeing organizations deliberately put themselves into place and it's all stemming from them being that trusted partner to the business and avoiding in the end that cognitive decay.

Speaker C: Yeah, and these are the good example, right? The good bright examples from the industry. And I share that with you, that's why I do AI training. I do believe the solution is we need to upskill people, but strategically and in the right skills, not just teaching what AI is, what is not how to prompt. Okay, but there's the actual, the whole competence map of uh, being able to use generative AI in a way that is actually effective in reality. What I see from my end is people usually approach me with the efficiency metrics in mind. We see that we are using AI, we are leaving value on the table. Can we find ways to do it better? But then in the process it's very interesting and I had to have this interesting example of a smaller company, extremely brilliant, intelligent people, PhDs, the CEO, head of science, uh, somebody who works in psychology field even they are using AI and they're using it at a very advanced level. They came because they wanted workflows, but then ended up seeing that the cognitive work that they are supposed to put up front did not happen. Hence ah, the judgment is delegated without defining what good judgment is. And there's no standard of quality. So AI, uh, works with vague non existing standards. Of course the output is not good. Why did that happen? Well, it's because again it is easier to assume that whatever knowledge base Genai has accessible is the source of truth. And so it happens. So everyone, every user can succumb to that. And then when they realize, oh, actually I have to work cognitively to make this thing work. Well this is a realization. And I think they even said, oh wow, this helps me uncover my own assumptions about the way I work with AI. Uh, is a learning process. It's an iterative process. It's almost like a two way relationship. And what I find a little bit frustrating is that we are inclined to invest in leadership development in teaching people how to delegate. But we also need to teach people how to delegate to AI because it's a set of skills, right? And it's like to what standards? And as you and I know, good delegation involves defining standards to what quality. If that's missing, then you won't have good judgment from AI.

Speaker B: That's a fabulous example. Jelena. You do such great work helping organizations all over the world in this regard. Where can people get that benefit from you for their own organization? How can they reach out?

Speaker C: Yeah, good question. So people can reach out through my email that I'm happy to share with you and they can reach out through my LinkedIn where I post, uh, around similar situations and problems that we discussed today. So I'm always happy to answer any messages and share tips and um, yeah, anything that's of value for people.

Speaker B: Amazing. Those all be in the show notes. So overall, thank you so much for this great insights. We've really spoken to this enduring need for critical thinking. What these pitfalls that we are inadvertently stumbling, falling into these deep cliffs, some ways of getting out and these footsteps to emerge to either avoid that big pitfall cliff through some work with yourself or to help our organizations be more strategically inclined to further this critical thinking that we need for the future. Thank you so much for all your great insights, Yelena.

Speaker C: Thank you, Danielle. Uh, it's been a pleasure. I love that we are talking about this. So thank you so much for creating a space where we can have these conversations. And I really, really hope many more conversations will happen around this topic.

Speaker B: Thank you for being a part of the future. Subscribe now to Stay Current.