How I AI with Matt Thornton

In the inaugural episode of How I AI, host David sits down with Matt Thornton, Vice President of Customer Experience at SNHU, to explore how AI has woven itself into both his personal and professional life.

Transcript

SPEAKER_00 0:09

Welcome to our first episode of How Do Buy AI. Here at SNOO, we want to talk with folks around the institution about how they use AI on a day-to-day basis. Our first guest for the day is Matt Thornton, who is the vice president of customer experience, to talk a bit about how he uses AI tools. Matt, thank you so much for taking a few minutes out of your day to chat with us. Thanks so much for having me. I'm looking forward to the conversation. Me too. So let's get started. When did you first begin experimenting with AI tools?

SPEAKER_01 0:45

Yeah, I think I was probably an early adopter, and it was really when they first released Chat GPT. I was one of those early curious folks who really didn't do a whole lot with it formally in those early days, but I just started by asking it really simple questions to try to understand how it responded within the context of the job I was trying to do. In the first early days, you just tinker around with it like you would with Google, just trying to trick it and get some answers out of it. And then I started using it for some really basic decision making, right? So this product versus that product, or it's maple season in New Hampshire. And when we're out there tapping trees, I actually held up Chat GPT to a couple of trees. I was like, is this a maple? Is this a maple? It actually did pretty well for me. We're using it to compare vendors. And then eventually they'll start making slightly more complex decisions. We bought a camper last year, and it was like, all right, what kind of camper should I get? And if I buy this one, how much might I be able to rent it out? Can I make a return on this? But getting a little deeper with it. And then I have a six-year-old, and we actually started playing around with this with my daughter. Her name is Luna. And she started talking to Chat GPT a little bit to ask big questions. We always say, Hey, we got a big question for you over here from Luna. And it turned like these cute everyday moments into really deep learning adventures with her. It was neat. She would always say to me, Hey, I got a question. I got a question. Then we'd pull it up. And it just really challenged the way that I think about how I get information and how I learn new things. And some of this was just professional curiosity, right? In my working customer experience, working with product strategy at SNHU, I knew it was going to be important to understand these emerging technologies firsthand rather than just hearing about them. I wanted to quickly get into these tools and figure out how I could start to accelerate my work and how it could benefit our students as well, thinking about the impact on them and how they might be able to use these tools and interact with the university in a different way, thinking about how AI could help students with self-service transactions that they would always have to pick up the phone to do before. So, what ways could we speed up delivery of student services? But really, it was just me thinking about how to incorporate this into my day-to-day life, how to incorporate this into my work life and thinking about how to solve this, particularly for our students. One of the things we do in customer experience, we talk a lot about how we get the right content in front of the right learner at the right time to the right channel. And it's an algorithm that we're constantly tweaking and trying to understand. And it hit me like a ton of bricks that AI might be able to help us do that faster and help us get the right stuff in front of the right learners to be able to help them be most successful. So yeah, I think it was those early days that I realized AI is not only going to help me think beyond just finding information. And that's when I realized this was really going to change the way that we work.

SPEAKER_00 3:52

At the same time, like there was always like this like crazy output that would happen, like massive wild hallucinations. But at the same time, it was wow, this has tremendous potential. And like even now, like what we're like, maybe two and a half years out. The difference in the quality is astonishing in that short amount of time. Even like at the time, like the model updates then were like astoundingly different. Now it's I don't think plateaued, but the how good they are at specific tasks are much more isolated. So it's like have like new innovations and it's for code or it's for computer use or something like that. Whereas before it was just like um like we didn't know what to expect. And so that that's all to say. I think that there is this like uh inextricably linked piece of uh using AI effectively that comes from experimenting and comes from like these like things like how often am I gonna have uh ChatGPT make a kid's story uh in like the office? Like never. But there's still there's still something to be learned from that, and that that is that it can tell a story. So now what can it do with data? So if the right data is entered, can it tell a story about that data? Is it enough to bring it up in a way that is um maybe more engaging for different audiences or helps to make connections that you never thought about? I think that you know, just using it for for identifying trees, for like you said, for planning out how you're gonna make some money back on your camper, maybe all of these things go into that overall fluency of if it responds well here and it does this job here, then can it do that job over there? And how well does it do with it? And so I'm like completely on board with you. I think that's a great way to upscale and to figure out, and there's no like right or wrong time to start. It's just like I still do it. I'll have ChatGPT or Claude or whatever one and play around with this week open, and I'll just throw stuff at it and see how it handles it. Um, and sometimes I get really great results, sometimes not so much, and that's all okay.

SPEAKER_01 6:04

You got to test and learn, you gotta play around with it, see what happens. Test and learn. And honestly, sometimes having our kids do it is almost the best, right? I mean, for me, it's it's rewarding because I'm like watching a mini prompt engineer grow into her own and figure out how to do this with his tools. And by the time she's older and using them in her adult life, it'll just be just secondhand knowledge to her for how to engage with these tools and get the kind of output that she's looking for. We still struggle. And so it's really funny watching the kids wrestle with it.

SPEAKER_00 6:34

It's great. I love it. It's like using like phones or iPads. Like, I feel like it's just no, they're like born now and they're like, yes, I know how to use. And we, and yeah, same thing with prompting. Like, you you get them into it early. You and I I think with that too, you talk with them about the dangers of it, right? Like you don't want to put in certain types of information, of course, package for a child, but don't tell it your real name or where you live or anything like that because you don't know what happens to that information. So you get some like learning lessons in there too, about like digital literacy and what it means to be a responsible human engaging with these tools. And I think that all goes a long way because that's something that we're seeing is a big gap in society, and especially in the US, is like just that critical thinking surrounding these tools. So I think once children start using it, and with guidance, of course, and start thinking about the right and the wrong way to do it, it becomes like social media where like my daughter knows she can't have a TikTok, she plays Roblox, right? And like we have pretty set regulations about hey, you can't chat with people, you can't share your information with people because uh there's dangers associated with it. And so she gets it because that's the limitation for her, and it's been there since the beginning. Um, and so I guess that is just me agreeing with you ultimately, and that but watching the kids play with this can it it's really neat how they tackle problems and they come up with their own. I wouldn't have thought about framing it that way, but they're they're they're little child brains that are making these all these weird connections and things. So it's cool. It is, it's incredible. Neat to watch it. So next question for you has using AI changed the way that you think about your job or your expertise?

SPEAKER_01 8:22

I think that expertise is shifting in the way that we think about what that even means, right? Expertise used to be like, I'm the one with all the answers, I'm the one with all the data points to answer the questions quickly. And now having just access to data to quickly figure out two plus two equals four, not as valuable as it used to be. And so I think that we're seeing a shift here. And it's less about remembering information and more about how do we answer really good questions, how do we think really critically about the answers that we're hearing, evaluate those answers, and synthesize ideas. So it's definitely like shifting the things that I would place value on when we define what it means to have expertise in a certain field, right? I think we're moving much towards more critical thinking and be able to synthesize output and be able to understand where there might be problems, because to your point, it does. It still hallucinates all over the place. And I've run into several personal and professional use cases where that's been the case. And so we need to make sure that we're evaluating it very carefully, but then recognizing also how it can rapidly accelerate the work that we're doing. Because what I've found with the AI is it has quickly moved me off of staring at a blank page and instead just having a conversation with the tool to explain what it is that I'm trying to do, who my audience is, what the context is, right? I do find that the more context you give it, the better the answers are that you get out of it. But then again, it's really about evaluating the output of these things and understanding is this helping me get closer to solving the problem I'm trying to solve here? Or am I going down some weird rabbit holes with this? And when I even look at my own teams, I've got the Voice of Customer team, our CX Insights group, who are reviewing student comments by the hundreds and thousands every single day. And this just isn't scalable. It's not reasonable for them to continue doing that sort of manual work. And so they've leaned in with the AI tools and have figured out working with our data science folks, but we've built a large language model called Learner AI at SNHU, where we can now take all of the thousands of student comments, run it through this AI tool, get all the comments categorized and tagged, and then get a recommendation from the tool as to whether or not we should be taking action with this student. Do they need help from us tomorrow? Are they stuck? Can they get over this obstacle? What do we need to do to support them? Or is this just neat to know feedback that we can bank, but we want to keep that because we want to know all the good things and all the bad things too, right? So for them to be able to manually process all these, we just weren't able to hire at a rate that we would be able to do that and keep our voice of customer work scalable. So the AI is really helping out tremendously there. And their roles have now shifted from manual review of comments to now actually teaching the model, hey, you got this tag wrong. Actually, you should have used this tag instead to increase the accuracy of that model, right? Or going in and saying, oh, actually, on this one, you said you would recommend that we take action on this comment, but we don't need to. And here's why. Or if they identify a new tag that maybe the model hadn't come up with before, they can go in and give three or four different examples of how that tag could show up in other forms of student feedback so that the next time we encounter that feedback, the model goes, Oh, okay, I recognize this. I've got the right new tag for this. They've now become almost like little AI trainers and graders of the AI's output to make sure that the work is good. And there's going to be plenty of work for them to continue to do. That job isn't going away anytime soon, but it does fundamentally change the kind of approach that they're using to review all that feedback. And then we've used it in different ways across our teams. We've built some custom GPTs that can help us with conversations around prioritizing the problems that we need to solve at the university. In what I'll call the old days, we would all sit around a table and we would take votes on which things we thought were the most important problems for us to solve and how we would score certain things in terms of the feasibility of fixing this problem, the student demand to fix this problem, how much it would actually help students succeed or persist if we fixed it, technological feasibility. All of those are the things that we would score manually in the old days. And now we can actually build a custom GPT that can do a lot of that work for us. And instead of sitting around a table again, staring at a blank sheet, we're now looking at the output and saying, okay, that feels right. I'd maybe tweak this one up a notch or that one down a notch, but we generally find that this is directionally accurate in terms of how we would have prioritized these solutions. So I think there's all kinds of different ways that we start seeing it change our jobs, but it certainly isn't taking jobs away. And I think we need to be thinking about, again, the skill sets that we bring, the expertise that we bring to the table, and how that will complement the skills that AI have now made so much easier for us that bringing that skill to the table on a resume may not be as valuable as it used to be.

SPEAKER_00 13:30

I love that point. I'm not even going to add anything to it. Mike Draw. I do have one more question for you. And this is probably more of a fun one. If you had a magic wand, wish you could have granted about AI for something that it can't do, or you're not aware at least that it can do, what would it be?

SPEAKER_01 13:52

Yeah, I think about this one an awful lot, and I'm almost scared to say it out loud because there are a lot of risks around what I want it to eventually be able to do. So let me be clear. I do not want it to be able to do this for me tomorrow. But what I find is really great about it is that it can remember stuff. If I'm like searching for a restaurant in a new area that I've never been to before, it will remember my daughter's allergies and it will say, hey, actually, here's a really good one that she's gonna find plenty to eat, but maybe steer clear of this place or that place or the other. That memory that it has, it's not always on, right? Like sometimes you have to go back and remind it, oh, hang on a minute, that restaurant might not work. Remember, she's got a peanut allergy, right? So find me something different, right? So it's not perfect, but it does a good job of remembering a lot of the time. And it can connect dots across my life a lot of the time as well, but it isn't perfect at it yet. And what I find with the AI tools right now, it feels almost like a place that we go. It's like a tool that we go and visit once in a while, and then we come back from it, right? And I'd almost like to see AI be more present in our complete lives in a way that I would say, let's give this example. If I'm asking AI a question, how would I approach refinancing my house? It's gonna give me all kinds of instructions on how to do that, right? All kinds of information, great instructions. But I just wanted to refinance my house. Can you actually just go do that thing for me? Or at work, I've got this use case right now where I've got skip level meetings with all of my team. I run them about once a quarter. My team's growing. I don't know I'm gonna be able to do them once a quarter anymore. I might have to reduce them to just two or three times a year. So I'd love to be able to say, hey hi, go look at my calendar, look at all my skip level meetings. And if I've got them scheduled quarterly right now, change it to once every four months. Also check it against workday to look at my org chart and see do I have skip level scheduled with everybody in my org? And if anybody's missing, go ahead and schedule those two. Get in the Outlook, find the best time for all of us, stick it on there, and don't worry because I can move it all around later. And just like click, do the thing. It can't do all that for me yet. It might be able to suggest some really great ways for me to do that efficiently or effectively, but it's not going to do all that for me. And I love thinking about the time in your personal life when you can just like eventually sit back and say, hey, I want to go to Fort Lauderdale, book me a flight, get me this rental car, I need a mid-size SUV, get that kind of hotel, make sure it's got a roll-away bed for the kiddo, and book it between this date and that date. And you just sit back and let it do the thing. Because right now we all have to make all those bookings ourselves. But again, a lot of risk in all of that. Like, how much access do we want to give it to our credit card? How much access do we want to give it to all of our accounts to log into these websites and book flights, book hotels? That stuff starts to get a little bit scary, but it's going to happen. I don't know. I guess I think about what it might be like to just walk in my house and say, hey, check my email. What's going on in my bank account? Tell me everything I need to know in my life and what's in my fridge for dinner tonight. Make this all easy for me. But not quite there yet. Oh, that's okay. Before just giving it the keys to everything, because that is a little terrifying to me.

SPEAKER_00 17:17

Yeah, I don't think you're gonna have to wait too long to see some of that. We're starting to see it already.

SPEAKER_01 17:22

We certainly aren't we. I love it. Feature by feature, it's starting to creep in. It's good fun.

SPEAKER_00 17:27

Yeah, I like the point too about like how much access do you want to give it? Because I think we all have our own risk tolerance, and some people will be like, yes, totally fine, take all my information, and maybe nothing bad will ever happen. That's right, okay. Maybe it will. And some people are much more aware and privacy-oriented, um, and that's cool too. Absolutely, absolutely, yeah. Matt, this was a really great conversation. Thank you so much for taking the time on our inaugural How Do I AI episode podcast. I've been on a lot of really good things that give us uh something to think about after we're done. So thank you for having me. It was such a pleasure.