- You are listening to the HR Mixtape, your podcast with the perfect mix of practical advice, thought provoking interviews and stories that just hit different so that work doesn't have to feel well, like work. Now, your host, Shari Simpson. - Joining me today is Michael Vaughn, an adoption and education specialist at Open LMS, where he helps LMS administrators and instructional designers effectively leverage learning tools and platforms.
Michael is an accomplished speaker and presenter having delivered training on a wide array of topics, including AI and ChatGPT, designing better presentations, core principles for accessible design, maker education, and improving DIY instructional video. (upbeat music) Michael, thank you so much for jumping on the podcast with me today. - Yeah, thanks so much for having me. I know we've been trying to connect for a bit here.
- Yeah, this summer cold is killing me, so if you're listening and my voice sounds a little off, we decided the show must go on today, cough or no cough. So I will do my best to mute so that the audience doesn't hear me coughing in the background. So I appreciate you being flexible. - Yeah, absolutely.
- So I would love to start with us kind of talking about the role of artificial intelligence, particularly ChatGPT and how it's possibly reshaping learning and development initiatives in the HR landscape. - Yeah, I would love to see a bit more research show up on how people are actually using some of these things. Right now we're kind of in that article phase where people are talking about some of the cool things that they are doing but not at a bigger scale.
But there's a ton of potential because all of a sudden you have like this assistant, right? I would almost qualify as like an AI intern if you wanna look at it that way who can help you out with things like generating some initial writing that you then edit or provide feedback on and refine over time based off your interactions with the AI.
But a lot of those time consuming tasks that eat up your day and eat up your week are suddenly things that you can start to speed up rapidly just by incorporating an AI powered tool like ChatGPT. One example I would give is our company hosts the Moodle Learning Management system. One thing that can take a little bit of time is to build out individual questions in the quiz activity because you are filling out a form for every single question that you make.
That's essentially how websites work when you're building content. If you're not coding you're filling out a form and you are telling the website what you want it to do. And there's a special type of question formatting called gift format and there's another one called Aiken. And basically you just type the text in a really specific type of way in a document and then you can just upload it and it bulk creates all those questions.
But learning to make that, learning to type in that syntax is really difficult depending on which one it is. One of them is kind of simple but one is a little more challenging. And then you're asking your instructional desires and developers and content creators and SMEs to take on even more of a burden in terms of learning to use new tools. What I can do instead is I can explain to ChatGPT, here's how the syntax works for these questions.
Here's the questions and answers that I just copied and pasted from this Word document reformatted in the proper syntax. And it does a great job and it saves a ton of time. So I'm just looking at all these little opportunities along the way to take the things that are, I hate to use the word burdensome but really do eat up a lot of time and dramatically speed them up.
And it's like now all of those passion projects, all those wishlists that we usually just push to the side 'cause they would take too much time, we've got that time now. And I think it just opens up this really exciting new doors where we are getting our time back and being able to expand into new areas that we've always wanted to but maybe haven't been able to. - Do you see HR professionals now learning a new skillset?
I'm joking as I say this because I had a colleague post this on their LinkedIn and I had a good chuckle about it. But being prompt engineers. And if you don't know what I'm talking about it's basically how are you asking the AI to output the data to you? Do you think we're gonna see an upskilling in that? - I do, especially because it's a key part of how you interact with a lot of these AI agents. And not only that, every service requires a slightly different format for their prompts.
So if I'm interacting with ChatGPT, I can talk to it like I'm talking to a person, but I also need to be really specific about do I want it to pretend to be a type of person? Do I want it to pretend like it has a certain role at a company? If I'm using an image generator like Midjourney, the prompts are more so staccato, single words and short phrases to explain what you want the image to look like.
And not only that, there's an additional syntax where you have to add these special dashes and coated words to let know how big should the image be and how fast should it be generated, should it be weird, should it be a rough draft, stuff like that. And those can shift and change depending on which network you're using. So yeah, I do think that prompt engineering is gonna be a new skill that folks will have to pick up.
But more importantly I think HR professionals even well outside the learning and development scope are also going to have to start thinking about what is the role of AI or generative generative AI tools within your company, business, school, organization, and what are the guidelines for appropriate responsible use? And not only that, what is inappropriate use and what are the consequences for inappropriate use? And right now, I don't think a lot of people are really prepared for that.
I don't see a lot of guidelines out there. I don't see a lot of policies and procedures or even laws about how to govern these incredibly powerful tools. So that might be one of my bigger concerns extending beyond just learning to use prompt engineering as a skill. - Yeah, I totally agree. I think there's definitely an appetite to understand how to educate your employee population and HR professionals about proprietary data, right?
Not putting that into generative AI models and how you are asking it for questions and being very cognizant of your company secrets, those kinds of things so I completely agree. I think we're going to see more and more policies and guidelines come out like that in conversations around how do we have those conversations with our employees and leverage new technology and not be afraid to use that.
I think the caveat is that sometimes our reaction to it is, okay, let's not use it 'cause we don't know how to totally use it. And I had a conversation with Jennifer McClure a couple weeks ago and we were talking about, the DOD is figuring this out, how do we use generative AI? So if they can do it, we can do it, right? And so it's just how do you think about it?
As you think about some of the HR initiatives and you mentioned if you are kind of thinking about ChatGPT as an intern, how do you see using tools like this maybe impacting or improving things like employee development, retention and engagement? - I think one of the biggest challenges that you run into in any business, school, or organization, is very often when you want feedback from someone actively, you want it with a sense of immediacy, you want it now.
And the person that can provide it to you is not always available. And one thing that I find to be really helpful in terms of how I use services like ChatGPT is it gives me rough draft feedback right now, which makes me even better prepared when I get higher quality feedback from the human beings that I work with. So that's one area that I'm really seeing a lot of promise in. One area that I really appreciate about it is, again, sometimes getting started with writing is tough.
Sometimes getting feedback on your writing is tough. And I know alluding to what we were just saying, you have to be very mindful about what you put into these services so that you're not sharing proprietary confidential information. I basically treat ChatGPT like it's a third party vendor. I need to know basis. You don't need to know the vast majority of things I can keep a lot of stuff really generic.
But being able to rapidly generate ideas and think about, okay, now I have this pool of ideas, how am I gonna develop this to accomplish my goals? Or to say, hey, here's my writing, what opportunities for improvement are there? What have I missed? That's really, really helpful in terms of the drafting process.
So by the time I get access to time limited, higher quality feedback from let's say my supervisor, I'm getting her a stronger product, I'm getting her a stronger draft so that when she provides her feedback like we are much further along in the development process than we would have been if I had asked for multiple rounds of feedback or revision.
- I love that example, and you mentioned something earlier that I wanna pull in when you're doing that is the idea of asking the model to respond to you in a certain persona. So pretend I'm talking to the CEO of a Fortune 500 company in this vertical who has this background. I'm pitching an idea on X. What are the questions you would have?
I'm so amazed at some of the stuff that it comes back with that from a draft perspective again, I think all of these tools are really great and they should be looked at as a tool, as a resource, and they shouldn't be replacing your own voice, obviously. And I think like you said, like an intern, it's really kind of a cool idea to think about it that way. As you have worked with these tools and you think about the learning and development space.
As technology evolves we're always talking about, what kind of bias could be introduced in using different tools and different modalities of technology. Are you hearing or is there any ethical concerns or guidelines that HR professionals should be really mindful of when leveraging AI? - Yeah, absolutely.
The bias has been one of the biggest sticking points when it comes to generative AI because up to this point, most of the technologies that are really, really powerful, and when we talk about using them in our day-to-day work, they are usually algorithmically based, right? Everyone talks about, this company has got a better algorithm than that company.
And now with generative AI, don't get wrong, algorithms still play a role but really what matters is the data set that is being used to train that AI. And if that data set contains bias, that will be amplified in the results because if there's bias in the data that means that there's a lot of bias, right? Because it's not just data from one person, it's data from potentially hundreds, thousands, millions of people in that dataset.
OpenAI says that their ChatGPT their GPT-4 model which is the current model that is most powerful, it's passed the bar exam, it's become a sommelier like it's passed all these, it's almost become a doctor in the United States passing the USMLE and a very powerful model. And it's been trained on over a billion words. Now, we don't know what is in those billion words, but if there is data from human beings, there's going to be some level of bias in it.
And if that bias is not counted for on the output, then what's gonna happen is it's going to amplify stereotypes, bias, discriminatory statements and ideologies. One thing that we're seeing right now especially with image generators, is if you ask it to generate an image of a person of a certain gender or race, it tends to exaggerate historically stereotyped features of those people.
Without going into detail you kind of use your imagination there but that is one thing that's happening and that's because the data dataset was built from probably artwork or artists or photos that really exaggerated those features or stereotypes.
So that is certainly something I'd be aware of because if you're using let's say ChatGPT to generate messaging for clients or refined messaging for clients, it could very easily throw in something that might not catch for some folks that would be deeply offensive for others. I think one example I would give is, let's say it's referencing like a totem pole or something like that, a bit of appropriation, something that could definitely come off as offensive to a lot of groups.
And so it might be something that some folks just, it doesn't catch with them, but ChatGPT might generate something like that because it exists in the data set that it's pulling from. So you could very easily end up in a situation where you are creating information or content that ends up being pretty rude or offensive or insensitive because it exists in the data set.
I will say, beyond bias, one of the big concerns that I have that has not been addressed really or resolved in any way is if there are a billion words in OpenAI's dataset they don't tell us what's in it, how do we know that that information was not protected under IP or copyright laws? And if they use data that they didn't have permission to have to train their model which I then use to generate content that I use for work, are there legal liabilities for that? Can my company be sued?
Can I be sued? Is it OpenAI that bears the liability for that? I don't know how it works at that point. So it's sort of like if someone steals a photo from you and I buy that photo from them thinking that I'm buying royalty free rights to use this photo, and then I use it for company documentation and it's published and it's sent out to all of our clients, who is liable in that situation, it feels like a similar thing.
Yeah, the ethics can explode really, really quickly in terms of talking about this. I'm doing a session just on this in about two weeks in London. And it's because it's not just the bias, it's like everything related to these models and how we interact with them and how we use them, that can create a whole host of new ethical situations that we just have not come remotely close to tackling yet. - How as HR professionals do we stay on top of those ethical dilemmas?
We're not gonna be able to solve them all, right? We're not lawyers by trade, so there's things that we don't know about. I don't know the in depth, in and outs of copyright law, but I'm sure there are things that I should be looking at or things that I should just kind of be Googling every once in a while. What are some suggestions there so that HR people can stay, can use these tools really to be helpful but also not be so afraid that they are putting their companies at risk?
- I think the big thing right now that companies should really be focusing on with AI is developing a cohesive policy for how employees can and should not use these technologies. And to help with that I've actually been working on a project that is going to be published at some point in the next two weeks. So by the time this airs, it will be out there.
And it's an openly licensed generative AI use policy for businesses, schools, and organizations to use as a framework or as a foundation for their own. Even been fortunate enough to have some folks from LTG HR look over the policy and give me some feedback from an HR lens. So this is something that I developed based off of research that I did. I did use ChatGPT to generate some of the text and refine some of the text.
And again, some HR professionals have had a chance to take a look at and provide some feedback for that. The idea there is to clearly outline to employees what these technologies are and what they mean, what are some responsible, appropriate uses of these technologies? What are the irresponsible uses? And what are the consequences for that?
And I think a key part of this is going to be training because there are some folks who are aware of AI and think that it can just change the world but they don't really use it. And may not have a high level of technical confidence so this all kind of feels like magic.
And then you have other folks who use it on a regular basis and probably use it responsibly but could certainly use a bit more guidance in terms of what can and can't go in there, because these are incredibly powerful tools that can do an amazing amount of things for us. But if you lose access to that tool, you are losing a competitive advantage for your business.
I look at Samsung and Samsung had to temporarily ban the use of generative AI tools like ChatGPT because some of their employees were putting proprietary code into ChatGPT to help with troubleshooting. And once you put something into a service like ChatGPT, it becomes a core part of their data set.
And the way I explain it to people is it's like you have a noodle that you're holding in your hand and you throw it into an Olympic sized swimming pool full of spaghetti and you hope that you can find it again, like that specific noodle again. It's just not gonna happen, you're never gonna get it back. So the the trick is don't put it in the pool in the first place and then you are in a good spot.
But in the meantime, I look at Samsung as a really great case study because they lost access to a really powerful set of tools because of misuse, because there weren't guidelines and policies and procedures in place to guide employees away from a really problematic situation. So what did they have to do? They had to pull the fire alarm, get everyone outta the building and like call full stop on this and now reassess what things are gonna look like moving forward.
I think having some policies and procedures in place first and foremost are going to be the most helpful starting point because, not to belabor the point but the metaphor that I've been using lately is, the most dangerous thing you do on a regular basis is drive a car.
And to drive a car, you had to go through rigorous training, you had to be licensed, you have to register your car, you have to renew that registration and get your vehicle inspected on a regular basis to make sure that it's running well. You have to follow our traffic laws and if you don't, there are consequences associated with that.
Sometimes that consequence is an accident, sometimes that consequence is a ticket and we have a social contract, so if you and I pull up to a four-way stop and you get there just before me, social contract says you get to go first, right? So if I go first instead, I'm violating that social contract, I'm creating some conflict. And I look at AI in the same way. It's an incredibly powerful tool, it's not just ChatGPT, there are multiple forms of AI out there.
Folks reading through the news have probably been horrified if they learned what deep fakes are but it's a technology that lets you take someone's face and overlay it on top of someone else's face and body in a video or a photo. And you could imagine that would be misused pretty horribly to harm others. So that's a very powerful technology that could harm an individual person. But we are also less than a year out at this point from a presidential election cycle.
And there have already been deep faked videos used in primary situations to denigrate one candidate for the benefit of another. And we don't have any laws in place to say that's illegal. It's deeply concerning 'cause these are really powerful technologies and we don't have laws in place. So I think the starting point is until laws are in place like you've gotta get some sort of policies and procedures and training if you want your employees to be able to use these tools and use them responsibly.
- And I think that ties into also your information security teams really educating about some of these things like deep fakes and phishing emails and all the other things that we get on a regular basis. Making sure your employee populations know how AI is affecting those types of thoughts. So I really appreciate you sharing that.
You know, Michael, as we end our conversation and you look towards the future, how do you envision the future of AI kind of enhancing LND and HR, or is there any emerging trends that you're really excited about? - My favorite emerging trends are not so much the generative AI at this point, like ChatGPT or image generators and things like that.
It's actually seeing how some of these underlying tools can be integrated into existing tools and services to make them more approachable and more functional. And one of my favorite examples right now is a video editor called Descript. And what I love about this is when you add your video in it uses voice to speech or I'm sorry, voice to text processing to turn the audio into typed text on the screen.
And historically, if I was trying to explain to an SME or someone like that how to edit their own video, that's a complicated skill. It takes time and effort to learn to edit video and it's not always straightforward and every video editor is slightly different.
What I love about Descript is a traditional video editor if I get partway through a video clip and I want to say, okay, we're gonna stop the clip right here, I have to do something called slicing which basically breaks the clip right at that point. And so I'm scrolling along a timeline where I'm looking at my entire video file and I have to find the exact precise moment and then slice it right there.
And Descript, all I have to do is delete the words that I don't want and it clips the video automatically for me. And that is way more approachable to folks. Like that is a much easier interface to use. They even have some really clever stuff where it will just take out the ums and ahs and just make it look like you took a moment to capture your thoughts before you started speaking again.
There's an eye contact feature in there where if you're reading off of a teleprompter, it tends to be pretty obvious you're reading off of a teleprompter, so it'll adjust your eye contact so it looks like you're looking at the camera instead. And just these little features in there that make life a bit easier I think are pretty amazing. The last example I would give along these lines is the idea of being able to create an AI powered version of your voice.
Now that could be misused very, very easily. In fact, there are already some phishing scams that are built around taking videos of people's social media posts, generating an AI voice profile and then calling up their older family members to say, I'm in jail, please send money. Like go to Walgreens and buy these gift cards and send us the codes.
But if I think about situations like you've had a cough for a couple of weeks and your voice felt pretty rough and it's been uncomfortable to talk, if you had an AI voice profile and you had to do like some voice dictation for a learning and development program or a course that you're developing or some content or some narration for a lesson, you could just offload that to AI and it's not gonna, umh or uhs, you're not gonna have to edit the audio.
It's gonna read the script as is in your voice or in approximation of your voice and offload that work for you and also provide you a little bit of a benefit so you can rest your voice during that time, right? So little things like this that are coming along are really, really exciting to me. Even more so than the generative AI stuff.
It's just how it's starting to integrate into other services and tools that we use on a regular basis to create content for not just learning and development but work in general. - Yeah, how cool, that's so exciting. And yes, I could probably use that today, a generative AI model for my voice. So Michael, this was a great discussion. I'm excited to see the future of these tools.
I think you're right there's definitely ethics and legal concerns, but also there's some really exciting stuff on the horizon. So thanks again for taking a few minutes of your day. - Yeah, thanks so much for having me. - I hope you enjoyed today's episode. You can find show notes and links at the hrmixtape.com. Come back often and please subscribe, rate and review.