Episode Transcript
[00:00:03] Good morning everyone. Today I want to talk about a phrase that I think you're going to see a lot in the technology industry, especially in hr, AI first or AI Native. And what of course is happening is that every software company in the HR tech space and virtually every other part of business is using AI, implementing AI, adapting to AI in a whole variety of ways. And the there's starting to become a bifurcation between the AI first vendors in the AI add on vendors. And I actually talked about this over a year ago at the HR Tech conference that initially the way generative AI was implemented by software companies is it was added onto core platforms to enhance features. So if you had a feature that was a search feature or a documentation feature where an analytics feature, you could use AI to enhance that feature. If you were vizier and you had a big people analytics platform, you could add an AI tool on top of it. If you were lightcast and you had a massive data set of labor market data, you can add an AI on top of it. By the way, that's what Galileo does. Or if you were workday, you could add AI modules into workday. But the core system of workday is still a transactional system, it's not an AI system. And if you're a learning platform company like an LMS and you know who all those companies are, are, those are 25 year old architectures, you could stick AI on top of it with a development tool. So it's all becoming very confusing because we have been told that AI is going to improve productivity and we want to use more AI features, but we're not sure what's real and what's not. And then you add to that the definition of an AI agent which is an AI that can do things, not just answer questions or analyze information or generate information. And is that an add on or is that native or AI first? So, so let me talk about it in general and then let me talk about it relative to learning and development in particular. So generally speaking, just to remind you, an AI system is a system that learns from the data, not necessarily the user. The user can inform its training, but its core functional difference between a non AI system is it's looking through the data using very advanced calculus to figure out what the data means. And then of course we take that model which is pre trained and then we ask it questions and it not only answers them directly, but it can reason through the questions and develop its own pattern of analysis to answer the questions. The reasoning model is what's in the mercury version of Galileo. It's in what's O. It's what is in O3, which was just announced by OpenAI.
[00:02:55] And you don't really use the AI system for transactions. So if you had 1,000 people an hour purchasing products on Amazon and you needed to put those systems into a transactional system so that you could charge them the right price and set up the shipping and send the appropriate fees to the providers, you wouldn't really use AI for that. That's a very transaction intensive process and you need two phase commit. You need to know exactly what the transaction is. It has to be correct. And AI is not designed for that. That's transaction processing. It used to be called oltp, Online Transaction Processing. I actually learned about it in the 1990s 80s at IBM when they had CICS, which was the grandfather of all those systems. And they had at the time developed a lot of algorithms to make sure that when you did a transaction and something failed, you could either back it out or know that it failed. You don't really know that in an AI system. It's not really designed to give you that information. So regardless of how much you like AI, at least today these systems are used for other things than the core business applications we have. So if you're a payroll company, you kind of don't want the AI guessing what somebody's pay should be. You might want the AI analyzing the pay against their performance. But the actual payment, that goes to the tax authorities and the payroll system and the insurance policies, that's gotta be correct, and that's gotta be validated to be correct. Ditto in the learning market, when you take a course and you get credit for it, or you pay for it, or you get credentialed for it, that has to be correct also. And that's why learning management systems exist primarily, is to track that stuff. So in the learning industry, where I'm going to be spending quite a bit of time for the next couple of weeks and through the end of May, we have these systems that essentially track what we've learned, not what we've captured in our minds, but what training we have consumed. And the way these systems were designed, believe it or not, actually goes back to CD ROMs. In the 1980s, there was a group of airline companies, airplane manufacturers, that decided that for all of the training they were building to learn how to manufacture and safely maintain airlines, they needed to be able to track it. And that training was stored on CD ROMs. Most of you have never seen a CD ROM, but it's basically a disk that stores data. And so they created a standard called the Airline Industry CBT Committee. AICCP or T I forget, defined how a piece of content, a course, would send information to this database. And so thanks to the AICC standard, you could track who had taken what course and who was certified in what. It was pretty simple. And we actually implemented it at one of the companies I worked in, but it wasn't very expansive. So they expanded it into Scorm S C O R M. And Scorm can do things like define what course you've started, what chapter you're on, what score you had, did you complete it, did you get a numeric score or some other form of score, Basic stuff that was for tracking your path through content. This was a long time ago, long before the Internet and advertising and other types of cookies and things. But it worked really well. And so all of the training technologies for many years used Scorm. And what SCORM does is it separates the content, the course or the modules of the course from the platform using a standard. So the platform doesn't really know that much about the content. So even if you want to give the user a very personalized experience, the course doesn't. The platform doesn't know what's in the course, it just knows where the chapters are. So it's very hard to create a personalized experience for learning. Now, learning's not the only thing that this impacts. Because if you think about your job today, whether you're a blue collar worker, healthcare worker, white collar worker, executive, whatever, a lot of what you have to do during the day is communicate with people and look at information that's changing all the time. You have to talk to a customer and then you want to tell somebody about what the customer said and you want to share that information on and on and on. So there's a vast amount of knowledge, information capture, retrieval, analysis and learning going on all day in your job. Even if you have a pretty routine job, things break, things change. You need to figure out what happened. So we're always learning at work. And the learning and development industry that really started my career in hr, which I've always loved, is filled with academics and really thought provoking people that have been trying to figure out how to make this learning part of work, better learning in the flow of work, micro learning, adaptive learning, virtual reality learning, simulations, cohort based learning, lots and lots of ideas, 70, 20, 10, et cetera. And in the middle of all of this, these new ideas in this really badly needed domain of business is this stupid standard called scorm. It feels stupid now. It wasn't stupid at the time, but it's basically a blocker that gets in the way between a system trying to help a user and. And the content that they need. So there was another standard that was created called XAPI that has a more granular data set so you can get more information from the content. But regardless of even that, this idea that the content is over there and the system is over here, and we got to get these two things to talk to each other is the fundamental problem we have in the information retrieval and learning in business. It is the number one fundamental problem. Because now that you have AI, and I'll get back to this idea of an all AI platform a second. Now that we have AI, we can generate the content from the platform. The platform becomes the content. They're not separated anymore. That is what an all AI system could do, an AI native system. And there's a lot of them out there. SANA is the one we work with for Galileo, there's Eightfold. For talent intelligence, there's others. An AI native system can look at vast amounts of data unstructured, including graphical data, numeric data, text reports, documents, checklists, compliance data, audio, video, et cetera. And it can then assimilate it and regenerate it in real time, near real time, into new formats that allow you to answer questions or learn. That is a spectacularly different approach than this AICC SCORM XAPI separation model we've had. So architecturally, in many cases of HR and business, we have a completely different way of thinking about systems. So in the learning industry, which is a $350 billion industry, by the way, a lot of people think it's a sort of a sideline, but it's actually huge part of business. This is going to change everything. And this is the really research we're launching at our conference is how much this is going to change everything. So AI native systems are radically different in what they can do and how they behave, and you're going to use them in a very different way. So let me tell you about AI native and what it means to learning and development, which is a sort of a compliment to the podcast I did a couple weeks ago. So let's go back in time a little bit, talk about learning and development in general, going back to these airlines. The reason AICC was created, it's called aicc. I just remembered scom, et cetera, is because when you're manufacturing a Business critical life critical of piece of equipment like a plane. And you've learned over many years precisely how to weld something or what torque to use on a bolt or whatever it may be. You really want to know that everybody is doing it the right way, because if they're not, something could break and somebody could die. We have in all of our businesses things that are in some sense mandatory for safety reasons, for business reasons, for government compliance reasons and so forth. And we need to validate that. Our workers, our employees, our teams know what those things are and can do them correctly. And then we have a whole bunch of things that are related to that, but they're not as compliance centric in terms of how do we use our salesforce system, how do we log our time off, how do we talk to customers, what is our culture, what is our management behaviors. So there's this whole gradation of learning that goes on in companies on how to do your job. Some of it's very soft, some of it's very hard. Now we as humans all consume information a little bit differently. Some people like to read. Like in my case, I'm a reader. I would rather read than listen to a long narrative, frankly. Other people want to listen, other people want to talk, other people want to do hands on stuff. People have a lot of patience. Some people don't have a lot of patience. Some people are in jobs where they could take an hour off, some people are in jobs where they can only take five minutes off, et cetera. So what we've had in training and learning industry for many years in the business side, I'm not talking about academia, is content developers or instructional designers who take this domain of information and they break it into pieces and they put it into chapters and they create books and courses and simulations and tests so that we can all consume it in the ways that are best for us. Unfortunately, we use what I call the publishing model. The publishing model is we as the designers of the content, take the corpus of information, we study the problem and the users, and then we build a thing for them to learn from. This thing we build is usually called a course, but sometimes has other names. And then we launch it. We stick it in the lms, connect it rather to the LMS so that the LMS can track it. We launch it, and then the individuals consume it. And the consumption experience is of course varied by person. Some people like it, some people don't, some people take it, some people skip it because the person taking the course may not need it or maybe they need a small part of it. Or maybe they don't like the instructor's voice or they don't understand the language and you wrote it in, you know, PhD language and the person reading it only has a bachelor's degree, so they don't understand half of it. You know what I mean? So we've got this, you know, publishing based paradigm problem where the thing we're trying to teach people to do is separated from them. If you used AI in an AI native format, that problem would probably go away because the user employee would be able to interact or ask a question or tell the system, hey, back up, I don't understand that. Or what about this? Please explain this more to me, skip this chapter, I've already heard this. And the system would just adapt. It's a radically different experience. And this is why ChatGPT, by the way, is so popular. Because it does what you want, it doesn't do what it wants, and you can stop it and reformat the content whenever you feel like it. So an AI native system for learning and knowledge is completely different from the old ones. So if you take your LMS with the traditional architecture and you add an AI piece to it, you really haven't created this dynamic content solution, you've just added a feature of the transactional system on top of it. So I was at a conference two weeks ago with a vendor, I won't mention the name, who has branded themselves AI native. And you know, there was a lot of things going on there. They were showing how you could generate characters and talking heads and stuff. It's pretty cool what you can do. And one of the sessions I was in, we talked about AI based instructional design and there was a course from atd. I was looking through the course description and they're describing how you can use AI tools to develop courses. And I'm reading this and I'm thinking there's something wrong here. Why would you use an AI tool to develop SCORM courses? It makes no sense. Now if you're a big complicated company and you have a bunch of SCORM based platforms, yeah, you got to build SCORM courses so they plug in and can be tracked. But actually what you want to use AI for is to create content dynamically that doesn't need scorm. So we have to, as you think through your tools and vendors and architectures for all these HR tech things, you're going to be making a lot of decisions, whether you know it or not, about whether we're going AI native or we're adding AI on top. And this is true in recruiting, this is true in performance management. This is true all over the place. By the way, a surprising finding that has come out of a lot of conversations I've had. And I'm going to publish a podcast pretty soon that's going to really get you jazzed about this is how big AI can be in performance management. Because performance management is something that everybody does, every company does it. It's a pain in the rear for every organization I've ever talked to. Employees don't like it, managers find it to be very time consuming and difficult. And we want the process to be positive and developmental and nonjudgmental. And it's hard to do that sometimes. So if you think about the performance management process, what an AI native system might do, it would presumably look at a lot of data about a person's work, compare it to the similar amount of data from other people's work, help the person see where they're strong and weak relative to other people in the organization, and give them developmental feedback on what they could be doing better, and evaluate them if you feel like that's necessary. I've always felt like the evaluation part of performance management is the part that's probably the least valuable. The developmental part is the main purpose. But everybody thinks about it a little bit differently depending on how they pay people and the culture. So I could go through every single use case of HR and I could give you an AI native scenario or an AI add on scenario. And you're going to have to learn as an HR professional what this feels like. Now, you know, the simplest way to evaluate it is just to look at what this new system does and how creatively it was designed. But for me and for us, since we talk to so many vendors, I can usually tell within a few minutes of a demo whether this is an AI native system or not. And some of that comes down to the creativity and ingenuity of the vendor. So there's a company we've been working with lately that's got a really cool AI tool for training and they originally built it for technical training, IT software stuff, and they've evolved it to use it for soft skills and they showed us the tool and how it coaches the user and creates scenarios. It's very really impressive what they built. And they're, I won't mention their name right now, but you'll, you know, I've written about them and you'll learn more about them. They're kind of a technology powerhouse. Extending their technology into new domains. And you know, they were showing it to me and I said to the guy, have you considered the fact that if the system is that dynamic, it could be plugged into recruiting, it could be plugged into pre hire assessment, it could be plugged into learning, it could be plugged into all these other applications to enhance them using the data sets that they capture. And he sort of stared at me and he goes, hmm, never really thought about that. That's not a bad idea. Because where I think AI is so different from transactional data systems is that the training nature of it means that the data corpus that you create can be reused in many, many other ways. And so the system that does pre hire assessment, the system that does onboarding, the system that does coaching, the system that does career pathing, they all really kind of should use the same data. It's not happened yet, but it will. And right now the interoperability between AIs is not even standardized. And there's a new proposed standard From Google called A2A agent to agent that's supposed to work on that, but that'll happen. So anyway, that's kind of my little session for this morning talking about AI, native AI, first AI architectures. And I think, you know, it doesn't really matter what you job is in hr. It's really valuable for you to learn the basics of this so you're familiar with what's going on. Because this is really different in the L and D space. It's 100% different from the way we did it before in the publishing paradigm. And that's going to be true in recruiting and all these other functional areas of hr as well as in many, many other employee facing applications that we have in our companies. Okay everybody, have a good weekend. Talk to you next week.
[00:20:09] SA.
