Advancements in artificial intelligence (AI) are impacting nearly every industry, including learning and development (L&D). With all the hype around AI, it makes sense that L&D professionals have some questions: How will AI impact the learning leader’s role? How can it be used to make their jobs easier? How will it shape the future of corporate training as a whole?
Training Industry’s CEO, Ken Taylor, and Dr. Tom Whelan, director of research, recently sat down to discuss how AI is bringing both challenges and opportunities to the L&D field. Their conversation offers greater clarity for learning leaders around the future of L&D in the age of AI.
In this episode, we outline five things you, as a learning leader, need to know about AI and its potential to impact corporate training as we know it.
Listen Now:
Additional Resources:
- Webinar: AI-powered Learning: The (Still Human) Future of Corporate Training
- E-Book: The AI Advantage in L&D: A Strategic Guide
- AI Article Collection
The transcript for this episode follows:
Speaker 1:
Welcome to the Business of Learning, the learning leader’s podcast from Training Industry.
Sarah Gallo:
Hi, welcome back to The Business of Learning. I’m Sarah Gallo, senior editor at Training Industry, here with my co-host Michelle Eggleston Schwartz, editor-in-chief.
Michelle Eggleston Schwartz:
Welcome. Today’s episode is brought to you by Training Industry Research. Here’s a brief message from our sponsor.
Ad:
Recent advancements and artificial intelligence have taken nearly every industry by storm. The corporate training industry included. AI tools have immense potential to revolutionize how organizations train employees to improve performance. But as with any evolving technology, it’s important to leverage AI strategically in both your learning and business processes. Training Industry’s recently released e-book, “The AI Advantage in L&D: A Strategic Guide” offers expert insights that can help you leverage AI for more impactful training, and as a result, better business outcomes. TrainingIndustry.com/ShopResearch or click the link in the shownotes for this episode to use the code PODCAST10 for 10% off the cost of this e-book, and to unlock the AI advantage in your organization.
Sarah Gallo:
Today, we’re excited to bring you a special bonus episode of The Business of Learning on a topic that everyone is talking about: artificial intelligence. AI is impacting nearly every industry from business and professional services to higher education, to customer service and more. And the corporate training industry is no different, which is why recently, Training Industry’s CEO, Ken Taylor, and Dr. Tom Whelan, director of research, sat down to discuss how AI is bringing both challenges and opportunities to learning and development.
So today, Michelle and I will outline five things that you need to know as a learning leader about AI and its potential to impact corporate training as we know it. Michelle, let’s get started.
Michelle Eggleston Schwartz:
Yes, I couldn’t agree more. Artificial intelligence is really flooding our newsfeed lately, with so many articles, social media posts, research reports and more shared daily around this topic. So today, we’re hoping we can offer some insights here that’s specific to your role as a learning leader. To get started, the first thing you need to know is that despite the recent hype, AI isn’t new … but it is advancing. Here’s what Tom had to say about the recent buzz and about the rise of what’s commonly referred to as “super AI.”
Dr. Tom Whelan:
So, I think it probably goes without saying, [that] AI is not very new. I mean, it’s been in movies for decades, but the use of AI technology, especially in business, is not something that is just now knocking at our door. It’s been around for a while, but to kind of categorize it, I think what most of us probably have had experience with over the course of several years at this point is what we would now call “simple” or “narrow” AI. So the things that drive search engine algorithms, it’s helplines simple kind of chat bots that don’t understand the question that you phrased to them, and recommendation engines that have algorithms behind them can handle complex information but aren’t necessarily, there’s not an exchange going on. You’re throwing something into an algorithm and it’s spitting something back out to you. For a long time, I think a lot of us have experience with these, and there’s a lot of organizations that have found ways to build these into their business processes and somewhat into their training.
But what we are kind of grappling with now in the market is the advent of what’s called “generative,” or “super” AI, and this is the fun and expensive stuff. So these are things like large language models, the ability for things in training or customer service, whatever. It’s to have complex conversational text, and a lot of those stuff is built on the back of what’s called natural language processing. And this is, I guess they call it a flavor of artificial intelligence, but what it’s really focused on is the interaction between humans and computers, machines basically. And what it’s trying to do is close the gap between the extremely imperfect way that we write and speak language and the rigid way that machines kind of accept input. So it is trying to fine tune. We can casually phrase a question 50 different ways and machine’s ability to parse that apart and understand, or at least in quotes, understand what the query is that’s coming in.
Sarah Gallo:
Tom goes on to explain that while simple narrow AI has been around for a while now, we’ve seen a lot of advancement in super AI over the past five years. In addition, AI tools like ChatGPT and Bard have only recently become available to the masses. So this leads us to the second thing you need to know, which is how these AI tools actually work using natural language processing. And as you’ll hear from Tom, a whole lot of testing
Dr. Tom Whelan:
For most of the things that we’re interacting with nowadays, whether it’s just with ChatGPT, whether it’s with Google’s Bard or any number of other products, that are APIs that sit on top of one of those Ais. All of that stuff at the core is driven by what’s called a large language model or how these get created is with machine learning, which is a kind of scary phrase, but just means you’re trying to teach a computer to do some stuff. But it’s focused very, very specifically on language. And in particular it is focused on the probabilities within language. So the model itself for what it’s called is full of what’s called parameters. Really all that is, or what that represents is associations between words and contexts. That is what these models are trying to parse apart. So it’s looking at, okay, if you have this word, what tends to proceed it? What tends to follow it?
And that might seem like a simple appraisal for it to make, but it’s repeating that billions and billions of times over and over and over again. And what it’s transforming is kind of the language. I mean for, I don’t know how many people are aware, but the GPT in ChatGPT stands for generative pre-trained transformer. It’s thrilling as a dentist’s office visit in terms of a name, but it’s a transformer. That’s what it’s doing. But also key to this whole idea is that it’s probabilistic. So it doesn’t really have an intelligence like we might think it does, but it’s trained on mountains and mountains of text to try to predict what comes next when you throw a prompt at it. And to have all of these parameters at its disposal, for us to be able to ask it something silly and be able to give us an answer back, it takes a lot.
It needs at least billions of parameters to be fed into it, meaning you’re talking about gigabytes upon gigabytes of text. One easy way, or the best way I saw it phrased was anybody that’s familiar with Mad Libs, it’s fun, complete the word things, it’s that, but millions and millions and millions of times over and over again. That’s kind of what the process of trying to train one of these models is sort of like, and I mean, it can be expensive for a company to try to stand up an estimate. I found granted a couple years old at this point, but in 2020 to train a model was estimated to cost about $1.6 million. I’m sure that cost has come down in that time, but still for a long time these things weren’t cheap. So even for it to not work well, it was still expensive to have something not work all that well.
For the most part, they need a distributed software to be able to run on, especially if it’s something that you need to be able to handle queries a lot. And the biggest gotcha with these things and kind of what is I think core to some of the problems or some of the humorous mistakes that it sometimes makes is that that any single one of these models requires a whole lot of testing and model validation. So basically you can feed it everything that you want. You think you could have all the parameters, right? But you have to test it and test it and test it some more to make sure that the kind of output that it gives you at least looks right or it’s quacking like a duck and walking like a duck and so on and so forth.
Sarah Gallo:
The most noteworthy thing here is the way in which we access and interact with the data, which again is through natural language processing. That’s what enables us to ask generative AI tools, questions and queries, whether it’s translate this text into Spanish or summarize each level of Kirkpatrick’s training evaluation model.
Michelle Eggleston Schwartz:
Yes, the capabilities of these tools are just amazing. We’ve already covered a lot of ground here. So to summarize quickly, the first two things you need to know about AI as an L&D leader is that AI isn’t new, but super AI is quickly becoming more powerful and more accessible through the creation of generative AI tools like ChatGPT. Tom also touched on how these large language models actually work and explained that they need to be trained using billions of data points. The third thing you need to know is that there are many use cases for AI and L and d as Ken put it. AI has the potential to be used pretty much everywhere in the corporate training process. Here’s what else he had to say about AI use cases in L&D.
Ken Taylor:
So we’re seeing it in the context of advanced technologies, in other words, technologies that allow you to quickly build courses. And I’ll talk you through specifically how it interacts with the course development cycle. It also plugs into most experiential technologies that you use being things like Zoom, whether it be your LMS that has classrooms in it or any technical experience, a lab, a coaching platform. It has the ability to touch all of those types of technologies when you’re experiencing learning delivery tech. So anytime that there’s an interaction over a digital interaction in the context of learning, AI can be used to better understand and process the analytics associated with those interactions. So you have the total experience technology, you also talk about delivery tech and then admin tech. In the context of the LMS, it’s going to…. I was at the learning tech conference, and there were hundreds of examples of how using machine learning, as Tom mentioned, there are certain processes that get repeated over and over and over that are becoming simplified by using AI to prompt you through the creation as opposed to making you the administrator have to do all of those individual transactions yourself. So just think of it like it’ll prompt you to say, what information do you need when you load the new course in the system? Or it’ll prompt you through scheduling exercise when you set it up, or it’ll prompt you to make sure that there’s an instructor associated with that event. It’ll do the things to make sure that you don’t set something up incorrectly because those patterns are repeated over and over and over and over. So it’s using that data that it’s collecting from those various places to improve, simplify routines, improve quality and improve experiences.
First off, we’ve seen models that will do the complete authoring. You give it a set of learning objectives and a topic and excitingly it will give you a first draft of a course, it will do all of the design, but there’s several and there’s several elements within content authoring, whether it be the creation of some of the imagery, whether it be the actual eLearning slides, whether it be creating instructor manuals. It can give you a first draft of the content for a course by just prompting it correctly up-front. And I use the first draft of a course very specifically in terms of the language, because it may or may not be completely correct, but it’ll give you enough that you can interact effectively and much more efficiently with a subject matter expert because you come to the table with at least a framework curating content.
So, it can now independently select the best of the content available to it and make recommendations to the learner as they progress through their learner journey. So it’s this notion of it watching the consumption pattern of you and those other folks who have also interact with a particular course or an assignment or an assessment and then curate the best, the most liked. It’ll take into consideration social context. It’ll take into consideration quality of the article. It will take into consideration recommendations by leadership. There’s a lot of ways you can influence what gets curated, but the cool thing is it will do that and it’ll allow you to just override it as part of building these learning journeys and creating these sort of rich multimedia learning experiences. So some really cool stuff there in combination or discussing combinations with personalization, the idea where it will prompt you.
There are programs that I’ve personally demoed that will actually prompt you through the creation of using competencies of a program that you can either author or curate preexisting courses from, say your course catalog, that allow you to assemble those materials around various subjects to make the best learning experience as quickly as possible. So those two tend to come together. So curating content in terms of extracting from the big content pool or personalizing journeys, helping you select those key elements that will make the learning best for the learner. Coaching [and] feedback is probably the coolest area that I’ve seen is we’re seeing a lot of tools establish a rubric for feedback and then basically watch the session or the conversation or the coaching exercise and then provide the coach, the teacher, the student, the person practicing with a set of feedback that will include not only how you performed, but what perhaps you should do to master [a certain skill].
And it’s really cool to see there’s even the ability for the heads of L&D or the heads of the program to go in and tweak some of those variables to make them unique to your situation. But basically they’re establishing models that can, and you drop in the video in whatever way it gets shared. And then basically it evaluates your performance on a variety of attributes, whether it be engagement, how reinforcing your messages, whether or not you use data correctly, whether or not you mastered those elements of the portfolio that you should have been sharing in a session. It’ll evaluate that session on multiple levels. And how powerful is that for you to be able to go somewhere, practice and then get immediate feedback with no individual intervention. This will be purely from the system. You can evaluate it, it’ll show you how it developed, the score it gave you.
It’ll show you where in the session that you used engaging phrases, or you were when you weren’t providing good eye contact. It’ll give you all kinds of feedback about your session. And that’s just one example of them. I’ve seen them in communication skills. I’ve seen them on sales interactions. I’ve seen them on customer service interactions. They’ve long been a big part of customer service interactions. But it’s exciting. That’s an exciting area because as we all know as folks in learning and development, we just need them to practice once they’ve learned what they’ve learned and we know that they’re going to be that much better. So super exciting area there for sure…. Coding support: One of the things that I’ve found particularly interesting is the ability to use ChatGPT and others to actually create code to allow you to interconnect existing learning systems.
Whether it be, “How do I get my code from my LMS,” or “How can I get that code put into another system?” Well, they’ll use industry standard data connectors and they’ll give you the code, at least the first pass of the code that you’ll be able to use to connect those two systems. And it may take an iteration or two or three. But the coding support is one of those things that I think will really help every one of us make connections across our systems using things like Zaps, for example, learner support and assistance. That’s kind of been around in a variety of formats for quite a bit, but I think they’re getting better. So that could be learner support in the course. So it could be a set of questions and answers that AI has developed to support the actual learning in module, or it could be learner support and assistance with sort of the system that you’re using or the environment that they’re currently practicing in or the lab that they’re struggling with.
It can be set up to help coach the learner through the exercises. And again, these things all just make a more enriching experience for the learner. And that’s probably at the key of why this is an exciting space for us. It can do first draft of job aids. You can tell what it wants. I think it’ll create workflows for you based on if you give it access to your current process, it’ll build you a workflow model that you’ll be able to use quick and easy. It’ll even guess what the steps are in your process so that if you don’t really know for sure, you can ask it. It’ll guess what the steps are, the logical steps are for providing good feedback as an example. So it’ll guess what the job age should be that everybody should have when they’re providing good feedback. And then you can go back and check it with a subject matter expert or adjust it to, again, the things that fit best in your work environment.
Obviously, anything that’s a computer interaction like simulations, it gives you even more processing power to understand all of the interactions that each employee did with the game. And I think that’s bringing out some really exciting and interesting observations. So again, it’s finding the point at which the student may be deviated from the learning by understanding its application in the context of the simulation. So I think that’s a really exciting area and probably one of the early places to emerge when we started seeing the value of data in both correcting the quality of the simulation and also observing where perhaps the instruction didn’t meet the need to make sure that the application would be correct in the simulation. So pretty interesting and exciting stuff going on there. First draft of a translation or a localization, my transcript, give it to me in language A, give it to me in language B, is it going to be perfect? No, but it’s going to be some percentage of the way there. And again, these are all just how do we get through these processes quicker? How do we get there? Even in the coaching support tools, it will do caption transcripts for the whole event. And like I said, it’ll tie, you said this phrase and this phrase was technically incorrect, and it’ll highlight it in the transcript so you can go back and see, “Oops, where did I miss?” “Where did I get it right?” And that’s the fun part about. Obviously I have a bit of passion about this area, but it’s the exciting part about seeing these things work is knowing how real they’re going to be for us very, very shortly, automated insights, strategy refinement. This can be prompts that are generated based on AI to help challenge your direction, whether it be in overall l and d and core strategy, have I matched correctly?
There’s this really cool tool, I’ll talk about it a little bit later, that actually you type in a competency and it tells you the skills that you should go and establish in order to achieve mastery on that competency. I mean, that would’ve been a bunch of work. Now is it going to be perfect? No, but it could be 80%. So it’s these kinds of things that are going to make it super exciting for us to get to the level of analysis around our learning and development programs that will allow us to make them so much better, so much quicker.
The last one I saw, which I think was one of those things that I didn’t even think of that in the context of l and d, we’re often responsible to promote the programs that we’re rolling out. So what better than to have ChatGPT to take a first crack at internal marketing emails on how to get people to be interested and excited about your program because sometimes it’s our responsibility to do that campaign to make the change happen. Well, there you go. You got a marketing assistant all of a sudden that you didn’t have before, where you can ask them, “How do I best SEO the page so that it gets found on the internal internet?” Or, “How do I set up an internal social page for the course?” “What should I be talking about in the context of the course?” Or even, “How do I refine my email marketing?” It can be there for you for all of these things. And again, I’ve been lucky enough to actually see specific examples of all of these working real time and it’s exciting and it’s coming.
Michelle Eggleston Schwartz:
I think it’s also important to note that many corporate training providers are already rolling out solutions that use AI in many of the ways that Ken just shared. Here’s what Tom had to say about how the corporate training market has adapted to recent advancements in ai.
Dr. Tom Whelan:
The market has already innovated anytime it feels like. We have wondered, is somebody using AI to do this? More often than not, there’s already four or five companies doing it. No sooner can you think of the idea that you can find several people in the market already trying to flex whatever that muscle is. So it’s an exciting time.
Sarah Gallo:
There’s definitely a lot of excitement and rightly so around AI’s potential use cases in l and d. But as with any developing technology, there’s also some challenges, or as Ken and Tom put it, “uncomfortable truths” we need to consider. So the fourth thing you need to know is that there are risks and biases to be aware of when using AI for any purpose related to corporate training. And here’s why.
Dr. Tom Whelan:
All it takes is a quick Google search to find grave and disturbing illustrations where it has poor performance. They call it “hallucinations” is the fun word for just flat out … It is making mistakes and giving wrong information, but it’s also been shown, it pulls through bias. There’s a potential for it to be misused, especially if it’s trusted or looked at as a subject matter expert in context where it shouldn’t be. Ken is fond of poking me about sometimes there’s a few things I get uppity about learning and ChatGPT will spout these things and then I will argue back with it and then we’ll go, oh, I’m sorry. But then you ask it again the next time and it’s going to hallucinate the same thing. It’s going to give you the same wrong answer. So when we’re using these AI tools, I think there is a risk of leaning on them as if they are a subject matter expert rather than kind of reserving the well, is this correct or not judgment for an actual person?
And to put it in less kind terms, it’s a flagrantly dumb, emotionally vacant, but incredibly resourceful assistant, not an employee stand. So it will be very confidently stupid in the responses that it gives you. Sometimes it has no sense of impact for what it’s doing. These tools lack a local comprehension of cause and effect, so they aren’t able to wrestle with what are the consequences of this information being wrong? What level of sure or not is the tool? The idea that these language models represent some sort of sentient intelligence is I think the fence that a lot of people jump over that we’re not at a point yet with these tools where we can actually jump over that fence. And the issue kind of when it boils down to it is all these things will learn. They do an awesome job of it, but what it learns is often the core issue because it doesn’t always learn or it doesn’t always pick up the patterns that are valid. And again, this ties back to it not being a subject matter expert or replacement for human judgment or reasoning. So these things can help us a lot, but they’re a tool, not necessarily a stand-in. They’re not going to call out sick because their car got a flat tire or something like that for work that day, but they’re going to make the same mistakes over and over again.
Michelle Eggleston Schwartz:
It’s clear that AI is still evolving, and it’s important to proceed with caution when using it to support your training efforts. With that in mind, the last thing we want you to know is that the future of L&D is still human. While AI can certainly help automate certain parts of your role as a learning professional, like creating a rough outline of a course or writing training video scripts, but it still needs a human to fact check its outputs, to avoid misinformation and to help personalize content to the needs of your learners and the business. So in other words, corporate training professionals won’t be replaced by AI any time soon, but they can use AI to make their jobs easier by automating certain processes so that they can focus their efforts on more business critical tasks and goals.
Sarah Gallo:
On that inspiring note, I hope you found this bonus episode useful as navigate these exciting but rapidly changing times. We’ll continue to release articles and additional resources related to AI’s role in L&D on TrainingIndustry.com. So stay tuned. You can also find more related resources in the shownotes for this episode at TrainingIndustry.com/TrainingIndustryPodcast.
Michelle Eggleston Schwartz:
If you enjoyed today’s episode, please let us know by leaving a review on your favorite podcasting app. We love hearing your feedback.
Sarah Gallo:
Until next time.
Speaker 1:
If you have feedback about this episode or would like to suggest a topic for a future program, email at editor@trainingindustry.com or use the Contact Us page at TrainingIndustry.com. Thanks for listening to the Training Industry podcast.
