AI in Higher Ed with David Joyner, PhD

Foreign welcome to Speaking of Higher Ed Conversations on Teaching and Learning. This podcast is produced by the center for Instructional Innovation at Augusta University. I'm your host, Andrew Everett. This is episode 15. Please share this episode and subscribe. New episodes are released the third Wednesday of each month during spring and fall semesters or Our guest today is David Joiner. Thanks for being with us. Dr. Joiner is the Executive Director of Online Education and the online Master's of Science in Computer Science at the College of Computing at Georgia tech. In addition, Dr. Joyner has created a professional certificate course on ChatGPT for educators, which is available on the EDX platform. And Dr. Joyner, so that folks can get to know who you are if you'll briefly share the path that led you to your current role and at Georgia Tech.

Yeah, absolutely. So thank you so much for having me. I apologize that my title is a bit of a mouthful. So I actually, I was working my PhD at Georgia Tech specializing in AI and education because I've always been a lover of education. I was a private tutor during my undergrad and graduate studies. And so I was looking at how to use AI in education when my PhD advisor was looking to do an online course in this new Master of Science computer Science program that Georgia Tech was developing online and asked me if I wanted to join him for it. And so I figured, yeah, that sounds fun. I'll do that for a year, I'll finish my PhD and then I'll move on to whatever's next. And I discovered that I loved teaching online. It was everything I loved about teaching and none of the stuff that I didn't love. And so I've stuck around since then. That was about 10 years ago now. I've gone through, developed a few of my own courses, run some things related to improving the student experience, and ended up here as the director of the program. Lately, that's taken on a big AI bent because there's so many things we can do using AI in education. And the fact that we're all online means we're already in the perfect place to use it because our students are already interacting with technology all the time. That's how I got to where I am now.

Wonderful. We thank you for sharing that. We're glad again that you're here with us to talk about something that is just causing so much, not necessarily chaos, but interest, and that is AI and how is AI transforming education for students and for educators? So we'll talk about that. And you know, we've had a little over a year now to assess how big of a change ChatGPT has had on the classroom. So what is your take on where we are now? Based on some of the initial predictions?

Yeah, and I think a lot of the initial predictions, I think, were very kind of doomsdayish in a lot of ways that this is going to just destroy everything we do. I think that happens a lot with new technologies is that whenever we confront a new technology, especially in education, we first see how stuff we've done and we've gotten really good at is no longer as relevant as it is used to be, for one of various reasons. And so a lot of the early reaction I know is kind of that doom and gloom kind of, oh, my gosh, how do I adjust to this kind of thing? And some of that has persisted, I think the past year and a half has really been characterized by the different trends in how people have become accustomed to using it, both teachers and students, that fundamentally, it's gotten a bit better over the past several months, and it's added some new features, but fundamentally what it does is pretty similar to what it was when it came out. And yet we've seen students go through learning curves of figuring out how to use it in different ways, both good and bad, and teachers go through some adjustment periods as well. We saw that in some of my classes, where early on it was kind of a, you know, starting to see a little smattering here and there. Students clearly using AI to the point where now it's become a more fundamental part. So I think that that learning curve is what's really characterized it. And there's a common battle, I think, between this perception that people are going to use it badly to replace their work, and the potential it brings to helping people learn material faster, master material faster, and get better at what we really want them to learn. I think we've seen a lot of that tension coming out. I think that's where we are now, is that there's a lot of belief in the potential, but there's also a lot of recognition of the fact that it is a clear and present problem to solve in some places as well.

What have you seen faculty doing to adjust to generative AI?

I think that there's different things on adjustments, the initial adjustments and the course you mentioned plays into this. Many of the initial adjustments we have to make are really on the assessment side that there are certain changes we have to make in order to keep our assessments reliable and effective, demonstrations of student understanding, because if we don't, we'll see a Large amount of AI generated content submitted as if the student did it themselves. Some of those changes, changes they've taken on different forms. There's different ways we can do it. I've seen some classes go to more proctored assessments and more timed assessments and more closed ended assessments, which admittedly I've done a little bit of as well in one of my classes where we added something like that because it's a class that really strongly lends itself to a lot of AI assistants. So we felt that was necessary to add as kind of another little piece of the puzzle. It's never the case that we want to throw out everything else we were doing, but it's created a need to have an additional little reassurance that students know some stuff from the depths of their own mind, as opposed to stuff they can just do with some AI assistance. But we've seen the other end of the spectrum as well, that there's this idea, if we borrow from the calculator analogy of this is kind of a calculator for writing and for thought and calculators enabled us in math classes to solve harder problems, to do more practice problems, to do more, faster, and as a result develop a better understanding quicker. I've seen classes that have done some new, more open ended assessments because if you're not yet familiar with this language or this concept or this historical figure or whatever, we now can expect you to get up to speed faster because you have this new tool and so we can expect you to do more as a result. One of my other classes I think has had the most fascinating experience with it because it's a very open ended project based class. And we basically tell students your project should represent 100 hours of work and it should be an authentic contribution to the field. That class hasn't changed at all because what you can do in 100 hours is still, you know, 100 hours of work. It's still a good project. But what you can do in 100 hours now has gotten to be a bigger project, a more ambitious project because you have access to this. There's kind of these, these two kinds of sides of it. There's the side where we have to adjust some things because if we don't, we might see students doing the GPT equivalent of just using a calculator for their entire homework assignment. But on the other hand, we can adjust what we expect them to do to raise our expectations and actually ask them to use it effectively and to use it to reach higher heights and better understanding than they could by Themselves.

So it's kind of raising the bar.

Exactly, yeah.

So can you share some examples of how AI is being used to significantly enhance learning outcomes?

Yeah, absolutely. I think that one of the interesting ones. So this is not exactly an example of how it's being used to enhance learning outcomes, but I think it's a fascinating example of how it's being used by students. It's actually a study we just completed in a couple of my classes. We have open ended, not open ended. We have open book, open note, open everything, tests. We've always said you can use anything you want as long as you don't talk to another person, live. You can use your textbook, you use the notes you've taken, you can browse the Internet. When ChatGPT came out, the question was, are we allowed to use this too? It was actually really interesting. Several of our students commented, well, no, of course you're not allowed to use that. That's way too sophisticated. And my reaction was, well, technically we never said you couldn't. It's not another person, so it doesn't fall under this policy of you're not allowed to talk to another person. So, sure, for now you can use it and we'll see how it goes. And the nice thing about our tools, we're able to actually see to what extent are students using these tools and how were they using them. We found some pretty fascinating results. We saw some students using them effectively as assistants in the process where they would usually have all their PDFs open and all their things and be ready to go and control f and find and search and things. They were able to then go and use these agents to ask those questions directly, but they were still using them functionally as an assistant, not as a replacement. We did see some instances of students grabbing, copying pasting screenshots and dropping it in. Just saying, give me the answer. What was fascinating was we found that the students who used AI were not doing any better overall than the students who weren't using AI. There was no change in the overall performance of these exams over time with the presence of these AI agents. We've seen that there's some interesting patterns in how students are using them on an individual level. But there hasn't been a systematic shift in the performance on these tests. We haven't seen the students who are weaker get any better because they have access to this AI. And we see the students who don't use the AI agents actually demonstrate a broader range of performances. The best students don't use it and the worst Students don't use it, and there's this middle range that does use it. I hesitate to overgeneralize from it because it's two classes that use very similar kind of assessments. But I think what we're seeing is we're seeing students use it to reach the heights they would have reached otherwise more efficiently. And my hope is that down the road, that will let them progress more efficiently as well and reach higher heights in the long run. But in the near term, we're seeing them use it to master the material the same way they would have otherwise, but in less time. I think it.

Well, from an educator standpoint, what new skills might faculty need need in order to use AI effectively in their teaching?

I think it depends in large part on whether or not we're using it in the sense of adjusting to it or actively leveraging it. The nice thing about it, though, I think actively leveraging it is the place that's a far more exciting kind of area, and I recommend everyone doing it. The nice thing about it is that the usability of it and the user friendliness and the accessibility of it is such that you don't need the technical skills you needed once. That's something we talk about in that course, is that for a long time we've been proselytizing the value of using AI in education. But for a long time it was a hard sell because it was kind of like, oh, yeah, you should use these AI agents. You should do this kind of thing. To do it, you're going to have to spend 100 hours building an AI agent for your course and integrating a canvas and talking to your learning management system, doing all this kind of stuff. But it's worth it, we swear. And now it's gotten to be much easier to sell because now it's just open this site, you know, maybe pay $20 a month. It's worth it, in my opinion. And you know, you can jump right in. And so the skills nowadays you need are far less technical than they used to be, and they can be developed almost entirely via practice. So I think the skill you develop is a familiarity with the patterns of answers that it gives to different kinds of prompts. And that helps you in a couple different ways. It helps you know how you as a teacher can use it to enhance what you do and where you can use it to make your life a little bit easier and make your work a little bit more efficient. It also helps you start to understand when students are over relying on AI and using it in ways you don't necessarily like, not in the way that you can give an assignment, say, well, Andrew, I know you used AI on this because I ran it through detector and it said you did. Those really aren't that effective. But so you can look at a more bird's eye view and say, okay, I can tell the answers in my class have shifted and students are using AI in this way as a whole, I'm going to shift my assessment in this direction. And you only get that if you use it a lot yourself and start to detect its patterns. It really is, in my opinion, at least once you've used it pretty frequently, you start to get a sense for when things are generated by AI. The analogy I like to use is that many of us have probably emailed with somebody who has a personal assistant who can email on behalf of that person. When you're emailing the person, you can feel, did I get this response from Andrew himself or did I get it from Andrew's assistant? And so ChatGPT and many of these kind of things, because they all learn from very similar data, kind of feel like everyone in the world now has the same personal assistant. And you get to know that personal assistant's kind of language and manner of speaking and manner of typing and you can start to kind of recognize it in different ways. We only get that from a lot of practice. And once you can recognize it, you can figure out how to adjust when you encounter it.

So maybe one of the first steps is to just open an account on ChatGPT and just start interacting with it and kind of, kind of get to know it on a personal, personal level.

Yeah, absolutely. I mean, so the bottom middle monitor on my computer, because I've got a few around me, is basically always chatgpt. And it's just something I go to, you know, several times during the day. I went through just one day during a conversation like this because someone asked, you know, what kinds of stuff to use it for and I realized why. I have some examples, but I really don't know what I'm systematically using it for. Most common things I use it for are what's that word? What's that word for that thing? And describe it. And it gives me the word far better than dictionary search could. Or what's a less colloquial way of saying this? It's very fine grained little things. But the ability to just ask and get an answer that quickly is really, really powerful. And it's not replacing anything that I would do. It's just something I was otherwise Going to sit here trying to think of for three minutes and it gives me an answer in 15 seconds. And when that happens several times throughout the day, that saves a lot of time. And that's the kind of thing we want our students, I think, to do as well. We don't want you to sit there staring at a blank canvas for 30 minutes just waiting for that first sentence to come to you. We want you to have a brainstorming partner to bounce ideas off of, as long as you put it aside at the right time and let your own understanding come out.

Absolutely. I use it a lot to check my emails to make sure I ask, is this clear? Is there a better way to make this message clear, especially emails to students? And just because it's in your brain and you put it, you type it up. But then ChatGPT will suggest a few great ways to restate some sentences that improve clarity. I like that as well. And it saves time, which is really my next question is how are you using AI to save time on creating assessments and providing student feedback?

Yeah, a lot of it is exactly that. It's the fine grain. It just greases the wheels a little bit. My workflow is very often the exact same. It's just that it makes me. It allows me me to do certain things far faster. I find it's also great for we're human. We do these annoying things where we get tired or exhausted or frustrated. And those are very often the times when I think it really subs in really nicely. There are times when I'm writing an assignment or writing a quiz and I'm giving someone feedback. And it's like I've said the same thing six times before. And I'm recognizing now that I really need just one really good explanation of this that I can use for multiple people. And that's where I'll go to ChatGPT and say, can you give me a really solid explanation of this topic? And then I'll tailor it a bit for myself. But you know, when you're doing that for lots of different things, it can get exhausting to just, you know, I need dedicated feedback for 20 different misconceptions that I'm seeing coming through on different quizzes. Having it generate just the boolean the or that you can then kind of refine and tweak a little bit is just a much more efficient and I think much more cognitively engaging process a lot of the time, just because so often I think what's frustrating as teachers is we know exactly what we want to Say, in our head, the act of translating it from our head onto the screen can be a tedious thing when you're doing it for the umpteenth time or when you just have such a solid understanding that you're just like, I don't want to sit here explaining this because I know it so well. It's not cognitively engaging to me to write out what I know, but it has to be written out. It's great for those kind of tasks because the act of tweaking something to make it more like you or use the examples you want to use tends to be more, you know, at least for me, I could be, you know, I could be alone in this. But for me, at least it tends to be more. More engaging. The other place where I find it's really, really useful to use. And there was. I was on a similar kind of meeting with someone who used the term, and I can't remember who it was, so I need to find out who it was so I can credit him. But he used the term automated empathy, which I think is really appropriate because sometimes it is a case that we'll get, you know. So you'll get the email from a student who asks the 85th question about the syllabus that you've gotten that day. And you know that question is answered on the syllabus. If you would just answer it or if you would just go read it, you would have your answer. And it can be hard not to be a little bit rude if you've gotten a lot of those over the course of the day. And part of the reason why it's hard not to be a little bit rude is just because it's easier to be. It's very often easier to shut a fire off just a quick little rude, you know, it's in the syllabus, go read it kind of response that takes less effort. It takes even less effort to say. ChatGPT. Hey, can you give me a nice pointer to send to the student that they probably find the answer in the syllabus and it'll do it? And it's even easier to be kind of friendly in that regard when you're able to say something. Give me a friendly way to say this thing that I don't have the patience to say friendly right now on my own?

Yeah, absolutely. What about some faculty might have this question, should I even mention AI to my students? So what do you say about that?

Yeah, so I think that students know about it in a lot of ways. Teaching about AI is about Teaching is very similar to teaching about a lot of other things we teach students about that we don't want them to do quite yet, but maybe one day they will and you can draw your own conclusions about what kind of things. I'm drawing an analogy to there. But it's the kind of thing, if they know it exists, they talk to their friends about it, they know what it can do for them, and so they're aware of it. Not telling them about it, not teaching them to use it effectively, is just going to lead them to use it ineffectively. Either they're going to use it to do their homework for them and not realize how it can help them do their homework, or they're going to do something that we see. In one of my classes, one of my classes, we tell students on all your homework and all the work you do at home homework. You can talk to your friends, you can collaborate, you can look up answers, you can do whatever you want. Because what we really care about is that you develop your own understanding however you want. And so if you need to seek help, you're perfectly fine doing that. We don't check those problems for cheating or anything like that. So feel free to collaborate, get together. We think learning should be a social experience. That's also a kind of a. An allowance that you don't have to worry about over collaborating with your friends, because we're not checking that at all. Now, does that mean you could go through copying things from your classmate? Yes. We're not going to catch you. As a result, we calibrate our tests to be significantly more difficult because the tests are really the only thing we can rely on as an accurate barometer of your understanding. We see that exact kind of thing. We see students who will go, and they'll use ChatGPT and say, hey, ChatGPT, how do I write some code to do this? And it will give a pretty good answer a lot of the time. And then they get to the test and completely, you know, completely vomit. And we know this happens a good bit because one of the. So my classes, all the classes I teach are online. And one of the benefits of teaching online is you get to see kind of a fine grain, almost creepy level of detail into student performance. And so when I have a student email me, say, you know, I worked so hard, I got 100 on all the homeworks. How could it be that I got a zero on the test? I'm able to go in and see, well, yeah, you got 100 on all the homeworks. You started Each problem at a certain time, you know, 8, 26, and you submitted a complete working solution. 40 seconds later, I couldn't do that. So the fact that you were able to do that on every single homework tells me something about how you went about it. And I'm not surprised you struggled on the test because you weren't really doing it yourself. Not saying you cheated or anything like that. You did, but it's not something we were detecting. But you were harming yourself. That's what happens when you don't tell them about it. They'll go and use it their own and do it in unhealthy ways. Whereas if you actually give them some guidance and say, hey, here's how you can use AI effectively, If you're stuck on this, ask it this kind of question. If you hit an error and you don't know what the error means, ask it what that error means, what some common causes are. But then put it away and go back to your own work and adjust your own work and fix it so that what you're doing is coming out of your own understanding, not out of the system directly. My experience has been when you give students that kind of feedback and you acknowledge that this is a thing that they have access to, they're more likely to use it effectively and healthily. The ones who will use it to just replace themselves, to circumvent their own understanding, they were going to do it anyway. You're not bringing anyone new into the fold. You're educating some people who are going to use it badly, and you're introducing something. Some people will use it effectively who might not have known about it otherwise, but most of them are already going to know about it.

Do you have any tips for how to talk with students who you suspect may have used AI?

Yeah, a lot of it. So I think the first thing is that you have to know how confident you are in that I think I alluded to. There are detectors out there. You can drop text and say, to what extent was this text generated by AI? And those things, they really aren't that effective, and it's not a referendum on them. When I say that they have a harder task than the generators. They are tasked with detecting text that was novel. When the entire idea of the Pardon me while I take my cat's collar off so she'll stop jingling. There we go. Sorry about that. But the entire idea is that their role is to detect the kind of text that was not generated by an AI. But the AI is programmed to generate text that Looks just like all the other texts out there. Detecting something novel is always going to be really big challenge. First thing is you have to know how confident you are in that because that flavors the conversation. The example I used before, where I was able to look at the submission logs and things like that, that really strongly affects the kind of conversations I can have. Because technically, I don't know if you use the AI. What I do know is that you did this problem in 15 seconds that you couldn't have done in 15 seconds. So something happened. I know you didn't do it authentically if it's text and things like that. What I find is that even if you don't know, you might not be perfectly confident that the answer is AI generated, but what you are confident about is it's kind of a weak answer. The AI generated answers that we usually see, they're better than that student probably could have done on their own. But they're not great answers. They're usually very vague. They misunderstand the context of the question very often they don't use the vocabulary correctly. There's a distantness. And honestly, having worked in this area for a while and seen patterns of certain things, the thing they most remind me of is contract cheating in a lot of ways. Contract cheating, for those who don't know, is when you just go to a website and say, hey, I'll give someone 30 bucks to do my homework for me. You give them the homework assignment, and when we get assignments written that way, it's very clear, relatively clear, I should say. This was written by someone who all they saw was the homework prompt. They didn't see all the conversations they had, they didn't see my material, they didn't see the way we teach it anything like that. All they saw is this prompt, and they don't know any of that other context. And so very often you can't prove that in a way that lets you come down and say you used AI, you hired someone to do this, you get a zero, but it's a bad answer anyway. And so the goal is not accomplished because the assignment still got a poor grade, may not have gotten an F the way it would have otherwise, but got a pretty poor grade. The other thing, though, I think that's really important in that equation is that this is where it's important in any class to have different measurements of student understanding so you have a full picture. One of the disheartening things that I've seen in some classes is faculty who have long had really good projects really good essay assignments that happen to be really well situated are well suited for AI assistance. And so they've gotten rid of them because, you know, well, my sensors just use the AI on those and I don't want that. And I think that's throwing the baby out with bathwater to use a colloquialism that I would find an alternative for if I could ask you to do right now. But you know, I don't think you don't want to get rid of assignments that lend themselves a lot of AI support. You just want to make sure you have something else in your puzzle of assessment that will help you measure that. The class that I mentioned where we added some new practice assessments is a very heavily project based and essay based class. Where it used to be, students would write pretty much an eight page essay every single week. And we were starting to see a rise in assignments that we identified as likely being generated by AI. Not in a way that we could come down and penalize them, but in a way that's kind of like, okay, these are, we're drifting a certain way that we're not comfortable with and so let's add something else just to give us another insight into student understanding. And that's why we added that we didn't rid of those essays. We removed a couple of them just to balance the workload. But we shifted some things over to just a different way of assessing so we can have some confidence in overall understanding. Even if a student really heavily relies on AI on one type of assessment and gets a real unfair advantage because of it, that's going to come out in other things that they do. Not just because they can't rely on AI and other things, but also because they didn't develop an understanding through those assessments that would then come out on those other other kind of things.

Can you share some other strategies that you've developed to adapt your online courses to include students using AI?

Interestingly enough, I'm in a fortunate position where, like I said, I teach computer science and my classes are all AI or HCI related. HCI being Human Computer interaction. So how people interact with computers, they all very closely deal with this. For us, a lot of it has been making it a first class object. One of the assignments we added to that class, I was just describing where we added some proctored quizzes is an assignment that just says reflect on the use of ChatGPT in this context. What are the potentials? What are the drawbacks? What should we do? How can we safeguard it? We integrated talking about AI as a first class object into the course material. In a way, it's very natural for us because it's an AI class and HCI class, it's part of the topic. For other classes, though, I think you can do some similar things by looking at some assignments that basically say, not just ask students to use AI, but ask students to reflect on how AI addresses certain things. In history class, for example, there's an activity that we recommend in the course about asking AI agents to summarize a historical event and then analyzing its summarization for different biases and different perspectives and what kind of view of the event the AI is embodying and using that as a reflection of how well does AI understand history, things like that. One of my favorite examples is when I accidentally got ChatGPT to summarize a book it insisted didn't exist, but just different little wrinkles of the prompt would take it from either no, that book doesn't exist to here's what that book is, is probably about, and I'll talk about it as if I understand it because the book came out more recently than that particular data set reflected. And so to then look at what was it about the AI, what was it in this equation that let it figure out that that's probably something that happened in that book? What does that say about its understanding? And that I think spans a lot of different subjects. Look at how I could see assignments like AI industry, AI in English, AI in literature, just to understand what does it mean to apply AI to these areas. And the accessibility of it means you can do that even without having a deep understanding of AI itself and how it works. In fact, that becomes a way of learning how it works.

What are some unique ways you've seen your students use AI?

In the class that I mentioned before about where we have the open ended, not open end, the open book tests. Early on it was just very simple, asking ChatGPT if it had thoughts on on different course content. And what do you think about this? Do you know the answer to these things? Because we're in a computer science program. More recently, we've seen more students coming into exams with AI agents. They trained on the course material so they, before the exam loaded up all the PDFs of the course transcripts and all the readings for the course, in some cases even some of the videos themselves into the database so that it already had a knowledge not just of the material as reflected on the Internet as a whole, but of the way we teach it. It kind of gets to that thing I was saying about it lacks the context of how you teach the course and the kind of examples you use. They load it up in advance and then they ask questions to it during the test that reflect that. And my attitude to that so far has been I'm all for it. I think they learn as much about what I care about them learning about through the process of training an agent, evaluating and doing all that stuff in advance than they would just by studying it themselves. It's a different understanding, but it's still a valuable understanding. The challenge is going to come, and I'm sure this is coming when instead of it being, I built an AI agent that does this when it instead becomes, okay, last semester, Andrew used an AI agent for this task, and I'm just going to grab his agent and I'm going to use his agent on my test. Then you get all the benefit of going through that process and none of the. The learning that you would get by developing it yourself. That hadn't happened quite yet. But I'm sure the day is coming and it'll be an interesting question to figure out how we adjusted that. I could see some different things that there's certain things we can do to adjust the wording of questions to specifically trip up the way large language models think. And so if you're really reliant on it, you're not going to notice some of those little wrinkles. But I think most of that is going to be a place where we say, even if you're going to use it in these kind of ways, there are other assessments where you're going to have to demonstrate your own understanding in different ways. And we trust there's other ways of being a good barometer of your real understanding. Sorry, go ahead.

Oh, go ahead and finish your thought.

I think it is interesting, though, that even as we start to see that happening, we still haven't seen a systematic change in performance on these exams. And so I think it is okay. My belief is that the only elegant explanation for that is that students are using it to do as well as they were going to do otherwise more efficiently, which I'm. Of all the possible outcomes, I'm kind of okay with that. If it was a case of students who were going to fail otherwise are getting B's, that's a problem. But we don't see that happening right now. We actually don't see any observable change in performance whatsoever, which means that either they're using it and it's not really helping them, which I don't think it's too good to not be helping them, or it's helping them do just as well as they would have otherwise in less time. Which is, I mean, in a lot of ways is exactly what we see in other areas of education that, you know, certain students come in and they're going to do whatever it takes to get an A. And if we reduce what it takes to get an A, they're still going to do, you know, exactly what it takes to get an A. There are others who, you know, they're going to come in, be perfectly satisfied with a C. So if this gives them a more efficient route to a C, then they'll just take their C and be happy with it. What we have to do as educators, I think, is adjust our assessments so that that C still represents C level knowledge. It might be more efficient as a way of getting to C level knowledge, but you still only are getting the grade that you deserve based on your understanding.

Interesting. How can AI be used as a tool to help promote equity in learning?

I think some of the more fascinating places and the really beneficial places are in this area of AI for accessibility. There's so much potential for it to be a major equalizing factor. Digital education already plays a lot of these roles. We've seen lots of cases of students who could never have succeeded in an on campus program due to some of the structural obstacles to success, be able to succeed in online programs like ours, Like Western Governors has another fantastic program that really draws a lot of people who can't succeed in a traditional campus environment just for reasons related to the accommodations that they need or some other kind of obstacles to participation even. It's just a case that they can't move to campus and they can't commit to the kind of schedule. So flexibility of digital learning has already done a lot in that area. But historically, especially for accessibility for students with more profound obstacles, there's been a lot of effort that had to be put in on the course level to make material accessible. And it's expensive to do. And so it's legally everyone has to do it and morally everyone should do it. But it doesn't change the fact that in some cases it can take three times more effort to make material accessible to 1% of your students than to the other 99%. No matter how dedicated we are to making it equitable, that's always going to present just a challenge. AI, I think what it can really do is provide an equalizing kind of mechanism because it can create tools that exist on the clients on the student side to adapt any kind of material flexibly to their needs. We see more things around bot sic you can drop a diagram in, it will describe the diagram to you in plain language and things like that. Where I'm really excited about this, we've seen this happening more recently, which is just really inspiring, is finding out students were able to succeed in our classes with certain obstacles without us ever even knowing about it until they had already succeeded. Because that's really what equity is. It's not a case that, you know, it's not that equity is giving everyone the exact same experience, but if we're able to provide an experience such that using just the tools everyone has available to them, they're able to self adjust to what they need to succeed in the course without ever feeling like they have to ask for special treatment or ask for special accommodation, or they have to incur any kind of inconvenience on the core staff to reach what they need to reach. That plays a major kind of equalizing role. Because now everyone can, I won't say do it themselves because that puts too much pressure on. You have more obstacles to overcome. So you have to figure out a way to overcome them. But it does provide people with a tool that can let them succeed in things that they might not have even tried otherwise because they didn't want to feel like a burden. They didn't want to feel like they needed special treatment to do what others were able to do.

What are some of the common questions faculty ask you about AI?

The most common one is how do I detect it? And my response to that is usually that's the wrong question to ask because a, you can't and you'll never be able to, and you probably don't want to. If you're asking the question how you can detect it, you're probably reflecting a course design that's going to be kind of outdated very quickly. In fact, in some ways already has. The question of how do you detect if a student's using AI to me is very similar to how do I prevent my students from using a calculator on their math homework? And at a certain level, maybe you do want them to do that, but you don't detect if they did. You modify the assignment so that even if they are, you're still reassured about the understanding that they're demonstrating in math. The entire idea of, you know, show your work is very much that kind of, you know, kind of purpose is I want to see that you did it yourself, not that you Just got the answer out of some system that could give you the answer. Now, of course, now that things like Wolfram are around, it can actually show you all the steps to solving a problem that's gotten, you know, even more difficult. But the idea that we want students not use that is kind of backwards because this is a great teaching tool. The question is, how do you design assessments and how do you motivate students to use it as a teaching tool, not as a doing tool? How do you get them to use it to help them learn, not replace their own learning? That's the most common question that I get. The second most common probably. And this is because we're all human. And I don't mean this to say this question. You know, judgingly, this is, you know, teachers are the most overworked profession, in my opinion, in the world because we're, you know, there's so much live performance, but there's also stuff to do after hours and things like that is how can I make it, use it to make my life more efficient? So how can I use it to help me? And that's where I think it's really, you know, it's really inspiring to see some of the use cases because its usefulness is very often in very. It's really well suited AI in general to the kinds of things that we don't like to do. So it's well suited to the idea of, you know, I have a lesson and I need to give just like a quick little attention check quiz at the beginning of next class. And writing 10 multiple choice questions is not fun, but it does a great job at that. Writing a really well thought out bit of feedback for the 25th student to demonstrate the same misunderstanding of something is not as cognitively engaging. And it's unlikely that you're going to give the same quality feedback to the 25th person you need to say it to in the first person. But it can do a great job of, you know, it's tireless in that way of, you know, give me a great example of why this is an issue in this essay. And it'll, you know, do that. And then you still, you know, you still tweak it, you still tailor it, you still make sure it fits with what you believe and what you want students to get out of it. But it provides that, that foundation. So I think it has a lot of things it can help with in that regard. It can be a really powerful teaching tool. And when we say it's a great teaching tool, so much of our focus is on how students use it, but it can be a great tool to help teachers reach students more efficiently.

So as we wrap up, I want to ask our continuing the conversation question, which is what is your advice for educators who are interested in learning more about using AI in their courses?

First and foremost is just use it yourself. Just make a habit. So in computer science, there's this real fun term called rubber duck debugging. And that's actually the reason why the COVID of one of my book, actually two of my books have rubber ducks on the correction. Now three of my books have rubber on the COVID is it comes from that rubber duck. Debugging is basically the observation that when you're encountering a problem with some code, very often the best way to reason through it is to explain it to somebody else. And just the act of explaining it surfaces some of the misunderstandings for you. And so just explaining it to a rubber duck or whatever you have on your desk is an effective way. ChatGPT is like a rubber duck that talks back. It's a verb deck that's always there. It's always available to you and can give you some real thoughts. It's not going to give you a whole lot that goes beyond what you yourself understand, unless you're asking certain kinds of questions. But it's always there. So my recommendation is just start using it frequently. Just the little questions you have over the course of the day of how can I make this better? What's the term for this? What's another way to explain this? Can you give me three examples of this? Those are exactly the kinds of things that it does quite well. And that familiarity you develop with that. Trying to teach nowadays without developing that familiarity yourself is like being a driver's ed teacher without a driver's license. You can't teach others how to use it without having a really strong understanding yourself. Even if that's not the topic you're teaching, knowing how to teach it and how to teach with it and what it's capable of doing is really important. The other thing I recommend is also talking to others, talking to people. The course that you mentioned, in part, it's some best practices that we've observed or some hints on ways that can be used that appear to be novel. But it's really also meant to be the foundation of a community of teachers who can share ideas and talk to each other and share their experiences and then share what they're seeing. Because we're all adjusting to this new reality really quickly. We're all going to Learn far faster. If we're sharing notes and sharing ideas, students are tech savvy, especially our younger. They're gonna get much better with this way faster than we are. But we're gonna see the kind of things that they're turning in and we can share notes and figure out, okay, we're seeing students really going in this direction. So how do we adjust our assessment? How do we adjust things to make sure they're still understanding this really important kind of topic? So I think talking to others about it and sharing what you're seeing and hearing about what they're seeing is the other really important kind of way to go about it. If you're really interested in kind of getting more in depth, there are some great YouTube videos out there. There's some great lightweight courses. The first of our course talks a bit more in depth about how it works under the hood. One of my favorite analogies for how people need to adjust to AI is again, the driving analogy, that AI is going to be kind of like a car. Everyone is impacted by the existence of cars. Even if you yourself don't drive, the fact that cars exist, you are impacted by them. So you can be the teacher who doesn't use AI in their teaching, but knows about it, knows what it can do, knows that when a car is speeding towards you, get out of the way, you look both ways before you cross the street. There's that level of understanding of AI. There's the driver's ed level of understanding of I can use it, I know nothing about how it works, but I can turn the steering wheel and turn the car on and put my blinker on. And then there's the kind of level that I think most of us want to reach, which is the basic level of maintenance, a level of familiarity, the driving equivalent of, I recognize that my tire pressure is a little low, I need to top it off when I get home, versus my oil light is on. I need to pull off to the side of the road right now because otherwise something's going to happen, you know, go disastrously wrong. That level of, you know, just kind of, I can replace my own wiper fluid, I can replace my own oil. That kind of maintenance level of AI is, I think, where we. Where many people will want to get, because that's the level at which you can put a question to AI, get an answer back and say, okay, it said this, but that's the kind of thing AI doesn't get right very often. One of our examples in that course is going back and forth to chatgpt about the hexagonal cloud pattern on. On Saturn. Saturn, Jupiter, maybe both. But, you know, asking, you know, where did that come from, what causes it? And watching it drift from giving really good explanations of what we know about it to completely making up sources about, you know, where the research was done, who did the research, creating people. And so to understand at what level does it, you know, start to go off the rails a bit. So there's some great explanations of that that will help you understand a little bit about how it's working, why it's probably not as scary as it seems, and how to detect when it's being used in ways you don't necessarily want it to be used.

Fantastic. That's great advice. I also want to talk about your new book, A Teacher's Guide to Conversational AI. What can we expect in this publication?

Yeah, so it's interesting. The course and the book are actually very, very similar. They're structured the same way. They follow the same kind of pattern of a unit on the fundamentals and then how we design assessment in an age of AI, how we use AI to help students directly and how we AI to help us as teachers. It came from the fact that when I was developing an Introduction to Computing course, it was a course and a textbook at the same time. And we found that 90% of the work involved in producing a book is the same 90% of the work involved in producing an online course. It's just at the end, do you typeset it and edit it and create nice little visuals, or do you film it and get in front of a camera and do that kind of thing? So they're very similar and they're meant to be kind of a question of what is your. What kind of experience are you looking for with this kind of content? If you're looking to join a community, if you're looking to get a certificate, if you need continuing education credits kind of things, the course is a great way of going about that and getting in touch with other teachers who are going through the same kind of things. If you're looking for kind of a quick reference thing, I think that you can, between classes, just say, what's an idea for how I can do this? The book is a great way to go about that. Books are far easier to leaf through and peruse, whereas a course is far better for intensive learning and assessment. But it's organized the same kind of way. It covers some of the foundational principles of how these work as they relate to educators. It strives to give you the level of understanding of how large language models and things like ChatGPT work, just at the level that you kind of need to know, kind of that maintenance level as well as some of the vocabulary. So when people talk about GPT and GPT3 and GPT4, you kind of know what people are, are talking about. And it kind of builds from the idea of, you know, what many of us start off by doing is figuring out how we have to change our assessments. And then it kind of uses that as a starting point because that's kind of the, in many ways that's the fearful reaction of I have to adjust to this reality because if I don't, my students are, you're not going to learn from me very effectively. And it uses that kind of as an entry point to say, okay, you started off with the changes you had to make. Now let's talk about the opportunities you now have available to you because you can use this in other ways as well and ends with all the ways that it can help us be more effective teachers, us more efficient teachers. Let us reclaim some of our time that we're spending on some really time consuming tasks. I think reinvest that energy into creating a good learning experience.

Wonderful. We'll include links to that on our show page. And thank you for this great conversation on AI and higher ed. What platforms can listeners connect with you on?

Honestly, nowadays the best place to find me is on LinkedIn. We can put a link to that in the notes, I'm sure as well. I found that's the one that's become most durable for me as every other social network goes through. It's kind of, it's chaos. That's the one that's persisted for me. So I recommend finding me there or if you have a question or anything like that, feel free to email me. You'll find my email address on my website, davidjoyner.net it's the best way to get in touch with me. Just give me a little bit of patience in getting back because my email does tend to be a wasteland, but I do see everything and reply to things that need to be addressed and answered.

All right, well, thanks, Dr. Joyner. I certainly appreciate you joining us. It's been an interesting conversation. I appreciate it.

Yeah, absolutely. Thank you so much for having me.

The conversation continues on our facebook page@facebook.com Augcii Please contribute your thoughts on today's topic and other episodes. You can find more on our show page at Augusta. Com. Please take a moment to rate Review subscribe and share this podcast. New episodes are available the third Wednesday of each month in spring and fall semesters. Speaking of Higher Ed is produced by the center for Instructional Innovation at Augusta University.