– Thank you, everybody, for being here. Really appreciate you taking your time to come join us today. We’re gonna be talking about common AI myths and how to move past them. This comes out of a article Zach and I published in “Inside Higher Ed” recently about the five common myths of AI and why we need to move past them and engage in a new conversation. So we’re gonna share that with you today, and while we’re gonna be sort of busting some myths, we wanna also be pointed to where we can actually start building new ways of teaching and learning in the AI age. Take it away, Zach.

– So Nik and I have been working with faculty at Chico State and other campuses since the start of ChatGPT, and we’ve noticed trends along the way. And early this academic year, one of the things that we clocked was when working with faculty, they continued to believe or ask questions about things that had clearly been outdated, approaches to teaching with or around artificial intelligence that had had become dated or were were no longer relevant. And when we looked around, we discovered, ah, they aren’t just thinking that ’cause they aren’t paying attention. Those are pieces of advice that continue to be propagated on blogs, in articles, and regrettably even on teaching and learning websites and other universities. That was the start of this particular article for us, what we felt like was we needed to address, why we arrived at these sort of one-neat-trick approaches to AI in the classroom, and then the conversations that was preventing us from having, and how to move on. So we’ve got a lot to get through and not a lot of time. So we’re gonna go to the first myth. Nik and I are gonna drop some stuff into chat here and there, and we’re hoping that that can be a way that we can make this a little bit more interactive. We will have some time for question and answer at the end, but we’re hoping that you can chime in in the chat throughout. So the first myth that we wanted to address, both in the article and in the webinar is, I know it when I see it. And you’re probably thinking some things that we’re thinking, which is, “Well, sometimes I do know it when I see it.” Regrettably, that is true, but it is mainly only true for people that are bad at using ChatGPT or another large language model.

So when we trust our instincts to identify writing that’s done by AI, too many bulleted lists, bad references, so many of these tropes that we’re seeing in the chat, and I look for those same things too. Oftentimes, those things show up, but if we’re using those as a shortcut to figure out whether or not writing that a student has submitted has been done by ChatGPT or another LLM, what we’re doing is we’re really only finding the students who are bad at using ChatGPT. So if you’re an experienced user or if you’re using the large language model to produce an outline and resources and then going in and adding your own prose or using a variety of humanizing tools, those things are virtually undetectable either to us or to detection software. So the reality is detection is unreliable, and unfortunately, it is quite biased because if we’re allowing our gut to lead us in those instances, oftentimes it leads us towards places that are inaccurate. And oftentimes it just reflects our own bias. And we know from experience and from research that when we are concerned about issues like academic dishonesty, we’re more likely to target students that are underrepresented minorities, that perhaps students that don’t look like us or don’t sound like us. It’s another way that we, and bias gets introduced into the classroom. So why do we need to move past this? We need to have better conversations. So we need to not be using these kinds of things as shortcut. And one of the things that Nik and I always talk about is designing for transparency with students, telling them about our own AI usage and starting with why, explaining why it is that here’s an instance where you might want to use a large language model, why it’s important in another instance to not use it and to do all of your own work. Now, those aren’t foolproof approaches, just like anything, but they’re a lot better than trusting your gut to know it when you see it. The truth of this is we have a lot of quasi-scientific, peer-reviewed research indicating that humans are incredibly bad at identifying essays as either AI-written or human-written. So even though we are spotting some things when we’re cruising on LinkedIn or reading an article, most of the time when we’re looking at an essay that a student wrote, we’re not very good at it, and we need to move beyond that myth to better approaches. Go ahead, Nik.

– So myth number two, AI can’t do personal reflection. So we’ve seen in the discourse, both written and also just talking to people and faculty, that there’s this idea that, you know, if you ask your students to just write about themselves, like a personal experience, a personal reflection, how they’re feeling, what they’re doing in the world, that you can circumvent the use of AI, by making the students tell them at all about themselves that the AI can’t do that, because, right, the AI’s not a person. So the assumption here is that we can kind of, you know, trick the AI or at least give stuff to students that, you know, hey, you wouldn’t wanna use an AI for that. But kind of like with myth one, what we’re actually seeing is that when students use the AI, we’re finding the students who are just bad at the AI, because let’s take two examples. For example, there’s student one who just takes your prompt about a personal reflection and just copies and paste that into ChatGPT, for example. They get an output, and it’s probably not gonna look very good. It’s gonna be very general and vague and sound like it could be written about anybody, really, or maybe even written by a robot.

But now imagine student two who takes that prompt, but instead of just copying and pasting it in there, they spend some time to give the AI a second prompt within that, that tells a little bit about who they are, what they are, the experience they wanna highlight. They can just give it a bit more context about who they’re as a person, or even make it up if they wanted to, and feed that in there. And then the AI will kind of, you know, work with that source material to produce a reflection. And these are probably the kind of things we do see in discussion posts. Many people now complain that even when you ask students to write about personal reflection, the discussion posts end up sounding very generic in general. And in that way we are simply, you know, observing the students who are not doing this well. So it’s not a really good way to catch your students using AI or prevent them from using AI. Now, there’s many reasons, pedagogically, that we want to use personal experience and reflection in our assignments for metacognition, for connecting the world to, you know, individual lives and the lives of our students. So that’s a worthy goal, but it’s, kind of, what we are saying is that we need to do that goal for a different reason, not to stop people from using AI or for catching them for using that, right? Getting back to reflection for learning, thinking about how we can communicate in full transparency to our students why we’re assigning the things we’re assigning and what kind of benefit we get from having these conversations about connecting our lives to the source material that we’re learning in our classes. So yes, on connecting experiences. Yes, on connecting experiences to concepts. And yes, on helping our students refine their judgements about the world, but using personal reflection as a way to circumvent or negate AI use, that is not a real good strategy to use.

– Thanks, Nik. I’m gonna have a quick aside about this one. This is actually the thing, the specific thing that Nik and I noticed in working with faculty that made us think, “Boy, we’ve got a lot of groundwork to do because we’re still using some advice that’s quite outdated.” So you may recall that in the early days of ChatGPT, a common approach was to ask about current events. And that’s because the training data for the very first instances of ChatGPT-3.5 were frozen in late of 2021. So if you asked it about a current event like Russia’s invasion of the Ukraine, it would not know what you were talking about. And so it was an approach in professors early on to frame a lot of their questions and a lot of their essays that we’re asking students to write in relation to current events. Obviously, that is not the case anymore. These systems are plugged into real-time news and social media networks and drawing information in real time. ChatGPT has been able to access the live internet for more than two years at this point. So the idea that then we’re gonna ask about current events and then that is gonna be a way to work around a large language model is just frankly no longer accurate. That said, as you can see, if you click through to the article, this continues to be advice that is propagated on blogs and, yes, even on teaching and learning websites. So obviously AI can access current events. One of the things that we’re asking people to do instead, if this was an approach that was attractive to you at some point, is to move past that and use relevance, not restriction. Explain to students why a topic is relevant to them, why it is important to their future professional or civic life. And this doesn’t mean don’t talk about current events. Like Nik just made the point in terms of personal reflection, there is absolutely redeeming pedagogical value in connecting course material to things that are going on in the world. It really makes material come alive for students. But the idea that we’re gonna ask them about something that’s happening currently, and by doing so, circumvent their ability to use a large language model to accomplish that task, is not true and has not been true for quite some time. Now, Nik’s gonna talk about something with no pedagogical value.

– Yeah, okay, so myth number four, the Trojan horse. So you may or may not have heard this. Back in 2024, a teacher went viral on TikTok, I think, that was embedding in hidden instructions into her essay prompts using small white font almost invisible to the human eye. So they could say something like, you know, I like broccoli, or tell me about broccoli, or whatever it happened to be. So embedding in their prompts on their learning management system small little text, and then students would just copy that text and use AI to do the work. And all of a sudden in the middle of their essay that, of course, they didn’t proofread, the teacher or the instructor would find something about why broccoli is the best vegetable, things like that. So it’s essentially, you know, what we see here in this image, a classic Trojan horse, which, of course, is in Greek mythology, but is also an actual criminal way that, you know, hackers and people doing bad things on the internet have been using Trojan horse code to infect computers for a long time now since the internet and probably before. And so this is actually kind of replicating a model of, you know, something that people would use to hack into people’s computers or take control or take data from computers into our pedagogy. It’s also, you know, related to a sort of gotcha-style of teaching, whereas I’m gonna sort of outsmart you or plant something in there that’s gonna reveal that you have been a dishonest person.

And I understand the drive, and I have friends and people I know that use this Trojan horse technique because they’re at their wits end. They don’t know what to do with the assignments anymore because so many students are using AI to cheat. And so to them, this is a compromise, but that’s one they are willing to take. But teaching our students, or at least applying prompt injection techniques that we find that people who are up to no good are using already to insert into LLM prompts to do nefarious things, it’s not really a good form of pedagogy. It’s unreliable, and it does undermine trust in our students. It creates an adversarial relationship between the police, you know, as the instructors, and the students, as sort of the criminals that I’m sure a lot of us don’t wanna have to do, but maybe sometimes feel like they should do or they need to do in order to gain some control back in their teaching and learning practices. So to move past this one is, I think, that many people would say, “Well, just don’t do it,” but you know, what do we do instead? Well, being explicit and transparent about when and when not to use AI and helping students learn what is potentially good about it and what is not so good about it and the appropriate use of it, or not, is a better approach to taking with our students to engage in that conversation as, you know, full adults that we all are. And having those sort of building those relationships and trying to have come through with a sense of trust. That does mean that we do have to then take that time to design new kinds of assignments or make those tweaks and also kind of make sure we’re always telling our students the purpose and the reason we’re asking to do things. And I think for a long time, many people, and myself included, lost sight of that. Like, we would just assign stuff, and we didn’t really communicate why those skills and the knowledge or the text or whatever happens to be in your class is important. Why is it meaningful to their lives? Why should they take the time to get off of TikTok or whatever they’re doing to put that time into your assignment? Or, you know, when they’re off of their shift at night and they need to do their homework, why is it meaningful to them?

– As we move into myth number five, Nik and I do want to acknowledge that we’re not here to shame folks that have engaged in these sort of pieces. After we published this, I got a message that was almost a confession from a faculty member at a different university, and that’s not our intention at all, is to make fun of or shame folks. So we recognized that most of these approaches were born outta a sense of desperation that faculty felt when the technology emerged. Suddenly a bunch of stuff that we’d been doing for a long time did not work anymore, and we did not know how to respond. So there was this scramble for these quick solutions. We just know that it’s been three years, and it is time to move past the one neat trick and into a different genre of solution. The last bit that we want to talk about, the ‘Claude’s not in our classroom’, is one that sort of tugs both ways for us because this is the idea that if you ask students about something specific that you talked about or used as an example in the classroom, that then a large language model won’t have access to that context. And that’s true whether you’re talking about solving a STEM proof, writing computer code, or writing an essay.

Obviously, similar to the personal reflection myth that Nik addressed earlier, this only works if you have users that are not at all sophisticated, because if a student is in class, they can easily add that context to a prompt that they are at, again, you know. You’re asking a question about the history of the relationship between the different branches of government in the United States. You can just say, “My professor used this specific example of FDR during the Great Depression. Make sure that you include that.” So students can include that context in prompts. And so that said, obviously integrating these sort of things together, these sort of layered questions that draw on class content, it’s absolutely a best pedagogical practice, and we want faculty to continue to engage that, to ask questions, have conversations with students that span assigned reading materials, classroom discussions, and deliverable assignments. It’s just that that’s not an effective approach to addressing large language model usage in the classroom. We instead are gonna encourage folks to take the time that you would normally spend to do this, and instead use that time to build AI literacy in your classrooms. Talk about appropriate usage. Talk about times when students should be doing their own work and why. As Nik said, for the last couple years, a lot of us got a little bit lazy, and this includes me, in outsourcing the authenticity of student work to technological solutions like TurnItIn. And so, in some ways, large language models are forcing us to reconcile with that, that we might have struggled with for years and return to fundamental questions like, why it is that we’re doing what we’re doing? The more we’re able to articulate that to students, the better off we will be as teachers and they will be as students. Nik, why don’t you take us home?

– So yeah, what happens when we move past these myths, or what can happen? The first one is we can spend less time on detection. Most of us probably, yeah, the vast majority of us did not go into teaching to be sort of police officers and control students and want to spend our time, you know, becoming a detective about who’s real and who’s fake. It’s not what we wanna do. I get why sometimes it feels like the only thing we can do, but by moving past these, we can kind of give up on some of that time that we’ve dedicated to detection and being a detective and also being an enforcer. And it’s no fun to call a student in and then tell them they’re getting a zero or they’re gonna fail and so and so forth. So with that more time we can get to focus on learning design, and this is the hard work. It takes a long time, but you know, revamping, re-looking at our assignments, coming up with new kinds of assignments, deploying old classics in new ways that are open, transparent, and we give our students a real sense of purpose and meaning about why they’re doing the things they’re doing in our classes, it’s helped me become more excited and invigorated to teach, actually, because I’ve been going through all of my assignments and all my classes and my student learning outcomes and kind of just starting over on a lot of that stuff. It’s definitely taking time, but it’s actually making me, I think, become a more attentive and a connected teacher to my students. And I’m learning about who they are in more ways because I’m connected. I’m asking and trying to figure out what it is that they want and need and what it is that I want to need in my teaching as well. And this also gets us to clear AI use expectations. You know, the world is messy, and the students, some use it, some don’t use it. We don’t have a great picture of this, but simple statements like, you know, banning it or just ignoring it are not gonna help us figure out clear norms about ways in which we can and cannot use these tools or should or should not use these tools going forward. So I think open transparency, dispositions where we have ongoing conversations and re-negotiations with ourselves and with our students around these tools are a way that we can kind of move forward past these myths and really open up more dialogue with our colleagues, with our students, with administrators, everybody who’s concerned about the fate and future of higher education.

With that, we really think that AI literacy is so important, and it’s a kind of a wide open concept, so we don’t need to define it here, in so far as to say that we kind of wanna help our students understand what these tools are and what they’re not, what they’re good for and what they’re not good for. What are the limits of the technologies? And we’re very well positioned in the university to take advantage of all of our knowledge of pedagogy, of the empirical world and our concepts and our theories. It’s kind of like our mission to help students understand new things and new technologies. And so I think putting more time into that in a very, you know, rigorous kind of way at the individual level, but even at the system-wide level or the university level on developing, you know, what it is, how to understand and approach these technologies, and what they are and what they’re not is something that we can and could be using with our time rather than trying to think about how to catch students or prevent them from using them in ways we don’t like. So that gets us to the last thing here is really about collectively and individually working towards new forms of curriculum, updating old policies. You know, for example, at Chico State, our writing proficiency policy dates back to the 1970s. So we’re assessing students’ writing proficiency on policy that was written really before the personal computer even existed. And maybe it’s time to go through those kinds of things and revamp them, get rid of some things, add new ones, but that time can be spent in these really important conversations about curriculum, not just at our level as instructors, but at department program level, college level, university level, all of those kinds of things.

So ultimately we think that the goal is not to win against AI. The goal really though is to prepare students for learning and work in an AI-present world because it seems pretty clear that this technology is not going away. And so we need to focus on how to help our students understand their place in this world. So thank you so much for being here with us and listening. We’d love to hear your comments and questions, so post them there. You can find us through email here, Nik Janos and Zach Justus. We’re also on LinkedIn where we post a lot. We write a blog called “Melts Into Air,” so you can find us on “Melts Into Air” to follow our writing. We also host a podcast called “Unfixed,” where we examine all of the ways that AI is disrupting and changing higher ed. And the last plug here is, please, if you’re interested in finding out how AI is intersecting with higher ed, we have a newsletter that comes out every two weeks. You can sign up for that and a link in the chat there. But thank you so much. Again, thank you to Dasha and Olivia and OneHE.

– Brilliant. Thank you so much, Nik and Zach. I really appreciated your, kind of, reminders, educators about kind of building trust and kind of having open conversations. So we do have some questions in the Q&A. In fact, we have seven questions. And so are you okay to answer, take it in turns, or?

– Yeah, Zach, you wanna go first?

– Yeah, sure, there are several questions, sort of, around some of the same issue. Michael’s is short, but addresses some of the other things that are popping up. Do you recommend using in-class writing assignments to see what students really know? And I’ll start by saying this partially addresses the the first question up top from an anonymous attendee as well. So I wanna try and do both those together. Nik and I are a big fan of what we see as partial solutions to this. So we reemphasize telling students why throughout, but we fully acknowledge that there are no silver bullets for this. That’s kind of what we’re here to dispel. It does seem like doing work in class for courses that continue to meet in person and continue to be live is a worthwhile approach. So having students do some of that work when that work is critical that they do it in class. And we’re seeing an institutional-level change as well. You might have seen that, I think just yesterday, Princeton University announced a return to proctored final exam for the first time in 133 years. So we’re definitely seeing that as a partial solution. We’re also seeing different things like faculty moving to different modalities for their final product. You know, not submitting papers, but instead submitting a podcast or submitting a series of interviews or an interactive multimedia piece. Now, obviously, AI can be used for some of those things. We live in a world of partial solutions. We haven’t found any silver bullets. Nik, we have quite a few questions. I think that Denise, sort of, asked a global one here. Can you say a bit more and define by what you mean by move past these myths?

– Mm-hmm, yeah, so I think on that one, our main thesis here is that the techniques that we kind of critique today, we think that they’re just not good at pedagogy for detecting students using AI or stopping them to use AI. It’s not that they can’t work. It’s just that they are distracting us from not only new conversations about higher ed, but the kinds of things that we need to begin and continue to work on refining, you know, our learning goals and objectives, deciding how many and what kind of classes should be online asynchronous versus what and kind of classes should be in person. What about, you know, programs and majors that require on these, like, really high stakes examinations, and like, how should those be handled, versus other departments like myself where we can kind of get away with a lot of smaller, you know, very, sort of, scaffolded assignments that are maybe lower order? And so, that’s what we mean by move pass, is not only just to change the conversation to some of these, you know, more critical things. It’s also to leave behind this idea that we can just, sort of, police our way through this, but then also to help the conversation move into action around curriculum design at the individual faculty level, at the program departmental level, at the higher levels of the university itself.

– I’m gonna try and answer Dan and Tracy’s questions together. And then Nik, I have one that I am hoping you can do specifically because you’ve addressed this issue. So it’s super clear that as this technology continues to evolve, we are gonna have to change our approaches and that that’s not gonna be a one-off event, that faculty universities as well as professional development organizations like OneHE, we are all gonna have the task in front of us continually updating our material. And that’s not just to find different ways to move past or work around large language models. The thing that we have to recognize is the world has changed, not just our classes. And so when we are adapting to this environment and these new changes, we’re gonna have to be in conversation with the people that employ or work with our graduates to find out, what do they need to know now in terms of agents or in terms of new capabilities that just wasn’t even on our radar two years ago? And so yes, agents will continue to change how it is that we teach, and I do think that course revision and updating is gonna be something that we’re gonna be have to doing much more regularly now. And I gotta be honest, as somebody who likes to do that work and talk to employers, that’s a future I’m looking forward to. Nik, I know you’ve addressed this directly here from anonymous attendee about people who are ethically opposed to using AI for a number of reasons when you’re trying to use it in class. Do you think you could address that one for us?

– Yeah, this one’s really difficult, and I think it’s an emerging, you know, issue amongst instructors. I’m just hearing from my colleague sometimes, “Oh, yeah, I had a student who doesn’t want to use it.” And I did have a student last semester who was in my environmental sociology class who said that they didn’t wanna do one of the assignments where they were doing, sort of, like a roleplay with ChatGPT around sort of environmental politics and values to kind of explore different perspectives on an issue. And they said they didn’t wanna use it because of what they perceived as environmental impacts around energy and water and so on and so forth that we have been hearing a lot in the news. And my first reaction was, “Well, some of their concern is obviously very real.” Some of it I think, you know, studying energy and the environment myself, I could say that some of it was also a little bit of misconception about the water use, for example, in AI data centers. And so I thought about it quite a bit, talked to my chair, and we came up with the idea that, well, maybe it would be best for the student to do that work who’s clearly says they don’t want to use AI, so I’m gonna trust that they’re not gonna use AI when I give them the alternative assignment and I had them kind of do some research going through the different perspectives and data on the environmental impacts of data centers and writing kind of a value perspective of, you know, what it means to sort of be ethically opposed to using AI, but also in relation to data about material requirements or the lack thereof, in some cases, of data centers that underpin the systems of AI. So it was a way for us to explore the conversation in more depth by taking them seriously. It was a different assignment I had to grade outside of all the other ones, but I worked with that student individually. Other than that and the number of years doing this in most of my classes, I don’t have a lot of students who who come to me with that, but it was a way that we could have a conversation and they could produce something that was meaningful to them, which was to explore that issue in a much deeper way than they had wanted to in the other assignment that they said they didn’t wanna do. So I thought that was a good compromise for both of us in that case. There was a question about asynchronous education.

– Yeah, Nik, I tried to answer that in text, but that and Chris’s question kind of go together. And this is more your wheelhouse than mine. I just want to acknowledge upfront, this conversation about online classes, especially about online asynchronous classes, oftentimes gets glossed over, and we don’t want to do that. We think it’s critically important. We’ve addressed it directly with a four-day intensive at Chico. Nik, do you wanna offer any quick advice for folks? And I have encouraged them to email us as well.

– Yeah, yeah, please email us, and I see Dasha has put some resources there. This is the biggest one, the asynchronous online teaching. It is the thing that is most broken. And I do think that universities have to come to some decisions about how many classes they want to offer online. But my department here at Chico, we run a very old distance education program that goes back to, you know, mailing things in. But we transitioned to the online era about 25 years ago, and we have now the second largest distance ed program at Chico State. We have, you know, most of our majors now, 70% of all of our sociology majors are distance ed. So we put a lot of time into it as a lot of what we’re doing to redesign our classes, our learning objectives and all that. There is no easy answers for this one, but we’ve mostly all moved away from discussion posts, for example, where most of us have embraced Perusall, which is a social annotation and reading tool where students and professors are watching and reading texts, and they’re commenting like a social media thing and having a conversation about the reading or the videos. That’s where I’m integrating more of my AI-type assignments that actually use AI around Socratic dialogue to supplement their reading, to supplement the lectures, to supplement other forms of activities that they’re doing. In my field, I have them actually go out into the world and take pictures or draw things or do things with their hands. Not every field can do that, but whatever’s appropriate to your discipline, finding ways to connect the different ways of learning that aren’t just relied on writing has been really important. We do more videos now, so I do video assignments, and this doesn’t scale. If you have a, you know, a 400-person class, you can’t do a video assignment where they’re all doing a five-minute video. But in my class, I had 30 students, and I found it actually more enjoyable to watch the videos of them talking about the theorists and connecting it to the world in their lives than I have ever had reading student papers back even in the good old days 20 years ago. The videos connected me online to the students in a way that I had never had before. I got to see them in sort of three dimension as real people who have voices, and you can talk, and you see them thinking. That has been really fantastic for some of us here in my department. That might not work for all, but you know, the last thing I’ll say on this is just really about experimentation and being open to trying new things but open to also failing at some of those things and moving to the next one. We don’t know what works yet, but if we don’t try for new things, then we’re just gonna be stuck reproducing a lot of these same myths here and kind of being miserable. So for me personally, I’d rather keep trying and working with others who wanna try on building those next things, those new things. And sometimes we get it right. Sometimes we get it wrong.

– Well said, Nik.

– Yes, very good end of this very insightful and very honest conversation. We really appreciate your energy in sharing your knowledge and expertise with us and with the OneHE communities. So thank you, everyone.