Guardrails, Not Gotchas: AI Ethics, Practice, and Policy That Support Real Learning

Christian Moriarty

Show/hide the video transcript
– Well, we’ll get started. We’re 10 minutes past. Thank you, everyone, for joining this OneHE webinar with Christian Moriarty. Christian is going to be speaking to us about “AI Ethics, Practice, and Policy That Support Real Learning.” And without any further ado, Christian, I’ll pass it over to you to introduce yourself and take us through the session today. Thank you very much.
– Thank you very much, Dasha. Appreciate OneHE for having me, and good to see all of you from all over the place. I’m told that, seeing from the chat, everybody’s really spread out, so it’s really good to see a cool, diverse crowd and be able to talk to all of you about this stuff. My name is Christian Moriarty. I am a professor of ethics and law at St. Petersburg College, which is a community college in Southwest Florida. St. Petersburg, Florida, not St. Petersburg, Russia. Also, yes, I am Professor Moriarty. I agree. It’s super cool. And I teach ethics. It’s great. I also am a director and the treasurer of the International Center for Academic Integrity, where we are in a truly international organization that does advocacy for academic integrity all over the place. We have members in every single continent across the world with the exception of Antarctica, but of course, we are working on the penguins any day now. Today, we’re going to be talking about ethics, pedagogy, legal stuff, all of the above when it comes to AI stuff. I have like 20 minutes to cover this stuff, so by no means is this going to be expansive or anything, but I’m happy to answer your questions or engage with you after the fact as well to talk about any of the various things that we’re going to be talking about today. So first things first, I am a lawyer. I’m a licensed attorney in the state of Florida. I am not your lawyer. My license requires me to say this. I’m very sorry, but I just want to make sure that we all are on the same page that this is not legal advice. Please do not construe it as such. That being said, I’m a community college professor, so my rates are so reasonable. But joking, but either way is that happy to talk to you about this stuff afterwards. Just make sure it’s not legal advice.
I also think this is hilarious. When ChatGPT’s picture creator first came out however many years ago, I gave it just this slide’s text, and it created this image, and I gave it nothing else, and it created a book with a beak and all this stuff. It’s just so funny. I just had to keep using it forevermore. So now you’ll never, definitely ever forget that this is not legal advice and I’m not your lawyer. So to give some background about this information, about where we are in the artificial intelligence space when it comes to academic integrity and truly genuine learning, we have some data to that that has started to trickle in about the extent of the issue that we’re facing. In a study of 1,000 UK undergraduates, 95% of them said that they use AI in at least one way for any purpose, 95%. 94% say that they use generative AI to help them specifically with assessed work as something that they’re getting for a grade. 12% said that they put AI-generated text directly into the assessed work. I put that difference between 94% and 12% as the lying gap. Quote unquote, “only 12% claim that they…” That can’t possibly be true. Even in my own practice, I teach 200 undergrads a semester. I know for a fact that’s not true, that it’s, quote-unquote, “only 12.” But I think it is important to recognize where we are in this space, is that pretty much you can assume every single one of your students is using AI to some extent or another and that it is affecting them in significant ways. And further from this is that the institutional response, at least in the US, has been, let’s say, light. In a recent study, only about 32% of institutions have a campus-wide AI policy. Now, individual instructors may have it at higher rates, but it’s hard for an instructor to get support in that if there isn’t an institution-wide policy. Only 23% of faculty members that have an AI policy even consider it effective. So of all the ones that have one, less than a quarter consider it actually to be effective to prevent youth students using AI inappropriately. One interesting result out of this was that faculty at those institutions say that institutions that embrace AI are more likely to call those their policies effective. I think that is just because they’re not sticking their head in the sand and not going all ostrich on this, that we’re recognized that it’s a real thing instead of just forbidding it completely and just pretending that students aren’t using AI. Now, one of the things that’s on my mind right now and is a significant issue, and to be clear, we’re facing this right now, is agentic AI. Agentic AI is where you give an instruction to an artificial intelligence and it does things for you. There’s various qualities of agentic AI out there, Claude, ChatGPT, but there’s more than just those. Your students are using them right now, and there’s precious few ways to track them in through most LMSs. Most LMSs right now do not track the use of agentic AI.
And then there’s a question if it’s even detectable. If you want to really depress yourself after the end of this presentation, Google “student using agentic AI videos,” and they’ll show you a student turning on an AI, pointing it at their LMS and it doing assignments for them without them even being in front of the computer. Consider how many of our institutions even have a rule that says that this is not okay. We have contract cheating rules that, funny, contract cheating rules, that seems like a 1-million-years-ago problem. But that may not be the same, and it may not even be against your rules. So we need to start considering the extent of agentic AI and whether students and how students are using it and how much that is obviating genuine learning. Okay, so existential crisis two, I guess. Now let’s move on to existential crisis three, talking about AI detection. AI detection is variously effective. The best AI detectors right now, generally speaking, are pretty good at detecting AI, but only if it is directly completely artificial intelligence in its entirety and the writing is more than a page or two. If a student lightly edits it or puts it into a rewordifier or whatever, a humanizer, that it may go completely undetected.
So the fact here is that detection is probably futile, is that even if it is effective and pure AI, all of our students are probably smart enough by this point to be able to avoid detection. But then, even let’s say it is completely effective, and when you have a rule against it and you say you don’t use it and it’s really good at detecting it, is that even really giving you an answer that you even need? If a student uses light Grammarly use, that is AI, and that will light up as 100% AI. You may or may not be okay with Grammarly. Not to mention is that shadow AI, AI that’s built into all of our programs that we use, including our writing platforms, are now in everything. And that includes grammar, the spelling, everything else. But even then, is that something that if a student uses AI, are they actually obviating their learning? And then finally, it’s unjust. Is that students that are not good at using AI or using free AI are going to be much more likely to get detected for AI, and that students who are good with AI or use really expensive, complicated AI are much more likely to get away with it. And so this means is that we’re disproportionately, inequitably hurting students from households that do not have as much opportunity to be able to be educated AI or to be able to afford it. A recent case that we have right now talking about AI detection and the legal response to it happened in New York at the beginning of this year, the case “Newby versus Adelphi.” A student was accused of using AI inappropriately based on an AI detector score of 100%. That student denied using AI and submitted results from two other detectors that said that that was actually human-written. Unfortunately, the process in this was light and/or maybe not upheld by their own process. The university’s integrity officer found a violation. The student appealed, but that appeal went to the same officer. So the court, based on this result, annulled the finding and specifically ordered the student’s grade to be restored and that to no longer receive a zero for that assignment. Now, it’s notable to take away from this course is that the court didn’t actually make a decision of whether this was AI or not. The court made a decision as whether they had a process and they followed that process, and the judge decided no.
Generally speaking, in American law, judges and courts will not question the merits, how they came to that decision, as long as they have a process and they followed that process appropriately. In this case, neither of those things were true. Compare that to another significant case in AI detection, is a University of Michigan student was writing a dissertation or a thesis and was accused of inappropriately using AI. The University of Michigan had a process and abided by that process and gave the student the opportunity to appeal and to be heard. The student sued, and the University of Michigan won, not because the court said that indeed they used AI. It’s because the court said that the student was given a process and that process was upheld. So the conclusion to be drawing from this, from the legal risk safety, is to make sure that you have a policy and you abide by that policy. So I have some suggestions for what those policies can look like. You’ve probably seen this in 1 million different places, but I by no means did not invent this. A ton of people are using this. This comes from an Edutopia article right when AI was first coming out, is that the red, yellow, green format or framework to be able to explain to students what is allowed to be used for AI and what isn’t. Red is you’re not allowed to use AI at all. Yellow is limited instructions. You can use it for some reasons and not others. And green, that you can use it, but you need to disclose and you still need to verify your work. My addition to this is that you need to make sure they have an institutional floor, is what is the default expectation of where this should be? And probably that’s going to be red. Certainly in the state of Florida, the Florida Department of Education just passed a rule that all 28 state colleges are required to have red as their floor, is that every institution needs to say the default to all students is, “You may not use AI in your assignments.” This is to give clear indication to students what those expectations are. And then the instructor can raise that to yellow or green based on their discretion and what they’re teaching.
So I think part of the issue here is that we’re seeing that students really don’t know what the expectations are, is that it seems different in every course or even per assignment. What I’m starting to do is that not only do I have something in my syllabus, but I’m starting to put this individually into every single assignment to give them clear indication every single time what are the AI expectations for this assignment, much less the course. Now, the learning objective needs to be indicative of what scale you need to be using here. So you can question, “What must the student be able to do?” If they need to use critical thinking, if they need to make their own argument, red is appropriate for this. If you want them to teach the skill, you want them to rehearse the skill that they need to develop in their profession, like their computer programmer, then yes, that yellow or green might be appropriate for that, is that you’re helping them develop the genuine skills that they’re going to be needing when they go off into their career. And so I think it’s important to note that red is completely appropriate for AI use in this. I think a lot of people are making the argument is, “AI is here to stay.” The joke I keep making is that the cow’s not just out of the barn, the cow’s in outer space. We can’t put the genie in the bottle. Pandora’s box is open. Whatever silly metaphor, analogy you want to use for this. But certainly in community college, but in any university, we are teaching baseline skills. And those baseline skills are writing skills, grammar, the ability to articulate an argument, indeed to have your own position on something. And so it’s completely appropriate to say, “No AI.”
We just have to recognize how we’re going to enforce that and how we’re going to assess that, which is of course easier said than done. So quickly into the chat, I’m curious what you think, which level is appropriate for this. A nursing student uses AI to draft a care plan for a simulated patient case and submits it only with light edits, and the finished plan reads well. If you were this nursing professor, what do you think what your level should have been for this student if you didn’t give them one in the first place and you received this? Do you think this is a red? Do you think this is a yellow? Do you think this is a green? Is this appropriate for a student to use AI for this, and to what extent? So quickly into the chat, I’m curious what everybody thinks. What level do you think is appropriate for this? A lot of yellows coming in. “Using for checking, proofing, yellow.” I’m always fascinated by how crowds respond to this. So Abigail’s hitting on an interesting thing, FERPA, the HIPAA-type stuff. If it’s fabricated data, maybe AI is used for this, but if it’s not fabricated data, it’s real patient data, then you shouldn’t be using AI for this, right? “Instructions should be explicit.” Yeah, right? But I’m sympathetic to the red arguments too, is that certainly if it’s right at the beginning of nursing practice, maybe it’s appropriate to say, “Yeah, you’re going to be using AI eventually.” I just came from a medical conference, and Epic, the AI software, is integrated everywhere. Yeah, and Barb, I think that’s a good argument, is that need to assess their critical thinking. Is it actually their own critical thinking if they’re using AI to do this? So I think both green and yellow are appropriate here, and it depends on how you’re interpreting it, how you’re using your assignment, what level the student is, all that stuff. There’s not a right answer to this.
Yeah, Heather, I think that’s something that I come to a lot too, red until we know you’re more advanced level and we know that you can use your own judgment on this. Good. See, this is exactly the thing, is that different people come to different conclusions on this, and that’s exactly where we want to be looking at this. Good, good point, Brooklyn. That’s a really good point, is that whether it’s formative or summative. So something to consider for your own AI policies is that there’s a bunch of different pieces in here to be thinking about. So what exactly is your assessment? What are you assessing? Are you assessing some sort of critical thinking or some fundamental underlying concept, or is it something more advanced where you want them to help develop that skill of using AI? I think it’s important to explain to students why we have certain rules. I think a lot of our students, or certainly a lot of mine, are coming from the perspectives that they were given rules and instructions, and they’re scared about questioning them or going against them, and they’re not even really sure why they’re there. I’ve started to taking the tack of, certainly the beginning of my semesters, but all throughout the semester, explaining why I have these certain rules, why it’s important for them to develop the skill, what I’m trying to actually get across from them. How to disclose. Certainly, you saw at the beginning of this PowerPoint, I disclose my own AI use. I’m encouraging, if I am in a yellow class, like my legal classes, if you’re not using AI in law, you’re probably doing yourself a disservice these days. And so for those classes, I tell them that yes, it’s appropriate to use it in certain ways, but you have to disclose it. And then what can actually the limits of that? If you’re using public AI, you can’t use private data into this. And then when they’re unsure, give them instructions about how they can ask and whether to ask and in what situations they can ask.
So I think, easier said than done, but there’s some assessments that we can make student thinking visible. This is by no means an exhaustive list. And certainly none of these are perfect. Version history is one way to go about this. Google Docs has a really good thing about this, is that you can enable a lockdown Google Docs that actually explicitly shows you the history of student writing in it, if there’s been any copy and pasting, stuff like that. In-class checkpoints, if you do have in-class students, that you can have students do tasks in front of you so you can verify them. Verification tasks, I’m a doctoral student at FSU right now. I have a professor that specifically asked me to do reflections on my AI use. I think that’s been helpful in thinking about it myself. I really think proctored assessments are important. For my synchronous, I do two proctored assessments every semester to make sure I know it’s them and that it’s actually their own thinking. And any various things to look at the actual thinking process rather than the only just the product. So, that being said, process-based is designed, but not enough. We have to recognize that while process-based design can help, that process-based design can itself be faked by artificial intelligence. Really good artificial intelligence is really good at faking human voice, and it can make up AI logs or anything else. And that there’s a significant literacy divide of students that know how to use this to its full extent and those that don’t. I think a solution to this is that we need to start requiring AI literacy as a core curriculum.
We at St. Petersburg College have started implementing this. We created our own faculty-driven AI literacy as well from the libraries to be able to help students that start thinking about their literacy and start questioning their AI use. And it’s indeed some students don’t even know that they’re using AI because it’s so built into so many of the software that they’re using. So we also have to recognize that there may not be a solution at all to this, that this is all a moving target. There’s never going to be a singular solution, that students are, despite all of this, are still going to cheat because it’s easy to do and it’s hard to prove. My argument to you is, “Guardrails, not gotchas,” is start thinking about ways in which we can verify genuine student learning through means of making honesty easier than it is to cheat. Students cheat for two main reasons. One is that they think their fellows are cheating so that they need to cheat in order to compete. And two, they do it because it’s easy and they can, and they know they can get away with that. And so, we genuinely need to put students in a position, we need to incentivize them where it makes it easier to be honest than it is to be cheating. That is very much easier said than done. But I think that is incumbent on us, is that expecting students to be honest and having integrity when their livelihoods are on the line, I know I’m not even really blaming them at that point, is that if they’re competing, they need to be able to have the opportunity to do their best work. And if that means that they think everybody else is cheating, well, they’re probably going to do that too. So we need to put them in a position where they are most likely not to cheat. So thank you all for listening to me. That was like woo. Asking an ethics legal professor to talk in 20 minutes is so hard. I’m one minute over.
– You still made it.
– This was brilliant. I loved it. It was short, to the point.
– Yeah.
– And I hope our audience loved it. We’ve got some Q&A and…
– Awesome.
– We’ve got 10 minutes for Q&A, so this is the most exciting bit. So we’ve got two questions in the Q&A.
– Sure.
– Would you be able to read them out?
– Yeah, sure. So, “I was drawn to the gotchas part of the title for this session. I’m curious about your take on faculty who place hidden text and prompts.” That’s such a good question. I’m personally against it. From the legal side, it’s not technically entrapment. You’re not forcing them. You’re not coercing them into doing something wrong. They’re doing it themselves. But I also think that it’s ethically specious in that you are still setting your students up for failure. You’re not lying to them, but you are setting them up for this. You know that they’re going to do this, so you’re trying to do the gotchas, right? I personally don’t use the hidden text. Also recognize that the really good AI can detect it and will alert to a student, “Hey, there was hidden text in here. Is this sure you want to do?” You’re taking advantage of older AI, not so good AI, from a weakness of it that’s called prompt injection, is that if you can get someone to accidentally or purposely give a prompt into it that it wasn’t intended, and it gives you a bad result, I think it’s incumbent on us to be honest with our students and to put them into a position that they can succeed. Now, students are still going to cheat even in this. I have to recognize that. I’m not naive. But I also recognize that I want us to have a collaborative conversation, a collaborative relationship with our students, and to make sure that putting students in an opportunity to do the best work without pre-assuming that they’re going to do something wrong. Again, easier said than done, and I very much recognize the temptation.
Next question is, “I’ve been tasked with updating our university’s academic integrity policy, but the institution itself lacks AI guide.” Great question. That’s a whole separate session, is AI governance, and I’m happy to talk to you about that too. “Each professor decides what’s permitted in each of their courses, and such stipulations are made in the syllabus. I’m struggling to academic the integrity policy given the wide variance.” Yeah, so I think if you can, one way to start with this is thinking about your institution’s AI philosophy. Certainly at my institution, this is where we started, and this has been the recommendation I’ve been giving everybody else, is you get the stakeholders in a room and start thinking about- You’re going to need a lot of AI governance policies. What AI can you use? What can’t you use? But you need to have the overarching, what are the virtues? I’m very Aristotle on this. What are the virtues we’re starting to look for? Transparency, honesty, integrity, whatever. And then all your other policies then can flow from that original philosophy document. And so yeah, please contact me after. I’m happy to go over the different strategies about doing this. I’ve also been doing consulting work, going to various institutions and helping starting the process. If that’s something on your mind, I’m happy to work with your institution of doing that stuff too.
Karen asked, “I incorporate free AI in the first few assignments in our first-year seminar class. My students often resist.” Yeah, I get that. Is that I’m pretty sympathetic to this, is that AI is the biggest copyright infringement machine of all time. Not to mention the electricity that it uses, the environmental issues, all that other stuff. One of my colleagues did the math on this, and it’s like every time you press Enter on ChatGPT, it’s sort of the equivalent of flushing a toilet in amount of water. That’s a lot of water. And don’t quote me on that. I’m not sure quite true, but I trust my colleague’s judgment on that. I think this is valid to recognize this, is that there are significant AI considerations to be made here, and that just because AI exists doesn’t mean we have to use it. That doesn’t mean we should use it. And so I think it’s worth having that conversation with them. “What is your issue with it? What can we look up to better understand this issue?” And if they’re very strong about this, particularly in a first year of seminar, I might be okay with saying, “Okay, you don’t have to. Let’s figure out alternative ways to do this.” And/or at the very least, giving them institutional AI in some way that is less likely to use their copyrighted data, because that’s effectively, when you use open artificial intelligence, you are giving away your copyright to a certain extent to be trained on their platforms. And I understand the trepidation against that. Chad, thank you for that and putting that in the chat. That’s really good use. The conclusion that website makes is that it’s not as much as people think, but it’s still not nothing. And I think that’s like, of course, like, what’s it called? Golf courses use way more water, right? But I’m also sympathetic to people not wanting those places in their own towns because they’re loud and that they increase electricity costs and all that other kind of stuff.
– We’ve got one question in the chat as well, which is kind of related to what you have just been speaking about.
– Yeah, yeah.
– About collaborative policy, certain AI policy collaboratively with students.
– Yeah.
– And what’s your take on that?
– Sure, so when we first updated our last policy, what I call the roadshow is that we had a core group of people that had a bunch of stakeholders in the room, deans, provosts, faculty members, students. I think students are really important to consider in this. And we started with that core group. And then when we started developing that policy, we started to go to as many different other groups as possible, started crashing people’s meetings, departmental meetings, student government, all that other kind of stuff, to start selling them on this idea and also taking their feedback. You’re much more likely to get buy-in, and people are going to abide by the policy if you explain it to them, get their feedback on them, and make sure that everybody’s understanding of it. There’s nothing quite like getting an email on Monday morning saying, “Hey, there’s a new AI policy,” and you’re like totally blindsided by that. So I think doing this as a collaborative process. Now, this is difficult. It takes time. It’s going to take away a lot of time from a lot of other things, but I think it’s incredibly important to conduct, particularly right now where we need these policies, not just academic policies, but governance policies. What AI can you use? To what extent institutional use? What student data can go into it? What can’t? The number of people, faculty members and staff members, that tell me, “I’m putting student data into my public ChatGPT account. Is that okay?” And I’m like, “Oh my God, no. No, don’t do that.” So it’s important to be able to have these policies and be collaborative on them so you’re much more likely to get them followed, and they might give you feedback that you didn’t otherwise consider.
– Great. Thank you. I’ll synthesize the next question for you. So basically the next question from the anonymous attendee is about using AI. So giving alternatives to AI assignments for students, but that requires a lot more work and grading workload for faculty.
– [Christian] Yeah, it really does.
– So what’s your take on that? Should faculty require to offer alternatives for AI assignments?
– I’m going to go, like, show my age. I remember when plagiarism detectors first became a thing, and everybody was freaking out about those, legitimately so. And that we, at the time when they first came out, we had an institutional policy that we had to allow students to have alternatives. It’s funny how what’s old is new again. And I think that’s legitimate to a large degree, is that I think students are right to not want to give their data to certain programs where they don’t know where their data’s going. So I think it’s fair to give them those alternatives, but it’s also fair to students to be able that they then need to do significant work to show that what they’re doing is actually them. Maybe that’s a proctored situation. Maybe that’s additional annotated bibliographies or whatever else that is more grading. That is more time. What I’ve seen is that I give my students that opportunity. Zero have taken me up on it despite them having problems with it. I think because they see that it is indeed more work. Now, is that coercive to a certain degree? “Use AI or I give you more work?” Yeah, it is. That sucks. But that’s unfortunately the situation that we have ourselves in.
– Yes, thank you, Christian. Are you still okay to answer two more questions?
– Yeah, no, please. No, my class is at two, so I got time.
– All right. So hopefully we’ll get through all the questions. So the next one is about universal design for learning and how it ties into some of the things you’ve been talking about.
– Yeah, sure. So universal design, I think, is a significant consideration when it comes to artificial intelligence, because universal design, I think, doesn’t cleanly fit with artificial intelligence. I agree. I think it can exacerbate inequities, that AI use is not always accessible. Screen readers might struggle with it. Not to mention the cost that comes with using them to one degree or another. I think Cal State universities are paying over $10 million a year to be able to give institutional access to their students to institutional AI. That’s significant. And I come from a community college. We don’t have that kind of money. We can’t provide students that kind of stuff. And so that means that the students that have more money and can afford $20 or $200 a month or whatever version they’re using are going to be put in probably in a much more significant advantage. I think an equity argument is that we need to provide AI to our students. Very much easier said than done. And then, of course, all the other ethical quandaries that come with that. And so, not to mention that students from indeed those backgrounds you’re talking about, first-gen, majority Indigenous, don’t have the background to be able to use this and that they’re at a disadvantage themselves. So I do think it’s incumbent on us to provide AI, to do AI literacy courses, appropriate, inappropriate use, and all that other kind of stuff. Very much easier said than done. Costs a lot of money, but it’s important for us to do if we’re going to have true equity going forward.
– Great. Thank you. I kind of answered the one of the questions that was in the chat. It was about asynchronous online courses, especially those that do not use high-stakes testing and would not benefit from those solutions. So what do you think of fully online interdisciplinary BA program in professional studies for nontraditional students?
– So I’m not quite sure I understand the question. Do you mind repeating it?
– So what ideas do you have or strategies for online asynchronous courses? So I think
– Oh.
– What it’s asking is like-
– Yeah. “How do you go about…”
– Sorry, I misunderstood.
– Yeah.
– Yeah, so about half my enrollment these days, I have 200 students a semester. About 100 of them, 120 of them are asynchronous online. So I’m right with you on this. I require a minimum of two proctored assignments. So one of them, what do they do is that it’s actually using proctoring software. A camera’s pointed at them and to be able to show evidence for me that it’s actually them doing that. I had a student this past semester that clearly was using AI because they were using a second computer, and I could tell because the two keyboards sounded differently. And so I think it’s really important for asynchronous to have proctoring. Now, I understand that there’s some arguments of equity and access, but at the end of the day, if we are giving them a credential that shows that a student has these skills, we need to verify that they’re the ones that actually developing the skills. I think it’s particularly important for math education to be able to show that they can actually do it without relying on calculators or artificial intelligence. But I use it for my ethics courses too. The other proctored assignment I do is I have them do recorded video of themselves giving a final presentation so I can actually verify it’s them. Now, that being said, is the PowerPoint that they use and the script that they’re reading from all AI-written? Yeah, maybe. And maybe I can’t fully suss that out. But at the same time, that’s certainly better than nothing. I know that it’s them. I know they’re thinking about it. And AI can’t quite fake those videos just yet. Wait two years. I’m sure that’s going to change. But I think presentations and showing that critical thinking, I think is one of the most best ways that we can deal with this with asynchronous online.
– Great. Thank you. And last quick question, “How effective is Turnitin?”
– I will not comment on individual products because I know not to get in trouble. What I will say is that if you’re talking about AI detectors, every AI detector gets better or worse over time. The best one six months ago got worse recently. One of the worst ones got better recently. That being said, AI detectors, generally as it stands right now, is that they’re pretty good right now at making sure that there’s no false positives. Is that the best AI detectors, generally, if it’s positive, it’s more than likely positive, 99 point whatever percent. That being said, they’re getting worse at false negatives. I think they’ve taken the feedback that an accusation from a detector is really bad, and they started to adjust their algorithms, and they only show that it’s AI if it’s for a fact AI. That being said, a student can very minorly adjust AI outputs or even put it through a separate AI to avoid AI detection, and that will then come up as 100% human, even through the best AI detectors. We also have to recognize that just because it is AI doesn’t mean they did something wrong. Grammarly is going to show up as 100% AI. That doesn’t mean that they used it to obviate their learning or their arguments, that just you fix their splice comma or something. So it’s my conclusion that I’ve drawn on this, is that I personally use a detector that’s paid for by my institution, but it’s advisory only. And that’s the recommendation I have for your policies too, is that detectors, well, if you do use them, seriously consider whether you should use them, but if you do use them, it cannot be a sole use of evidence by itself. You must be able to provide some other sort of evidence to show that this is something that is indeed obviating their learning, like a falsified, fabricated source, weird writing that doesn’t go along with their traditional writing, any of that kind of stuff. Certainly, with my face-to-face students, I start every semester with having them write longhand something for me, a short paragraph to get an initial assessment of the writing, and it’s something for me to compare to. Again, that’s not perfect either. None of it’s going to be perfect. Certainly, a bunch of students are lawyering up these days, and so you need to recognize that you make sure you go through that process and do it appropriately. But that is to say that it’s just complicated these days. And if you want to talk about this more offline and something that’s not recorded, I’m happy to do that with you, but you’re more than free to contact me.
Generative AI has become a standard part of student learning workflows, but most institutional policies and practices have struggled to keep pace. This webinar recording examines two interconnected challenges: building AI governance frameworks that are transparent, defensible, and practically usable by faculty and staff; and ensuring those frameworks do not entrench existing inequities in access, surveillance, and due process.
It was facilitated by Christian Moriarty who is a Professor of Ethics and Law at St. Petersburg College, the Ethics & Governance chair of the Florida Artificial Intelligence Learning Consortium (FALCON), and a director and treasurer of the International Center for Academic Integrity (ICAI) in the United States of America.
Useful resources:
Download webinar slides (opens in new tab, PDF, 786 KB)
DISCUSSION
How can universities develop AI policies that promote responsible use while ensuring fairness, transparency, and equal access for all students?
Please share your thoughts in the comments section below.