 Hey everyone, so welcome to Gen AI and Open Educational Resources and Teaching and Learning. There's two co-facilitators today. My name is Lucas Wright. I recognize a lot of faces which is great. I am a senior learning technology, sorry, senior educational consultant focused on learning technology and I've been doing that for about the last 13 years at UBC and I've been helping faculty, staff and students work and think about different learning technologies and mainly have the conversation. Related to this session, I had the privilege of working for BC campus for one year as an open education advisor a couple years ago. And, you know, my interest in AI was only started a couple years ago compared to a lot of folks in here, you know, folks like Nordin with a lot of expertise. I started a couple years ago and became really obsessed with it and started, you know, just as chat GPT started emerging as having this really strong potential and ability to create natural writing. I started really obsessing over it and researching and playing and I've had the opportunity to do lots of presentations. So I've presented about 30 times now and I've got to learn from people each presentation. So I come to you not with expertise in this area but kind of through my obsession, my research and my opportunity to do workshops. So welcome. Will, go ahead. Thanks. Hi everybody. I also recognize a lot of faces in the room. I'm Will Ingle. I'm a strategist for open education initiatives at UBC Vancouver's Center for Teaching Learning and Technology. And what that means is I work on one level with both faculty and students who want to use open educational practices or resources in their teaching and learning. And then on another level, I work on sort of broader university initiatives like our OER grants. I've been in this position for about 12 years and supporting open ed at various different levels during that time. I'm really excited to be here to be co-presenting with Lucas. I come from sort of the opposite place where I've been maybe avoiding looking at AI as it relates to open education and so really excited to be able to engage with Lucas and learn from his practices here and as well as everybody else in the room. I hope there'll be some good sharing happening today as well. Wonderful. So I'm going to share slides and you can jump right in, Will. So before we get started, we would like to acknowledge that UBC Vancouver, which is hosting this session is located on the traditional, ancestral and unceded territory of the Musqueam people. And as we're online today, you may be coming in from other territories and please feel free to share those. As the session will be talking about open education, I do like to acknowledge that western notions of copyright law and ownership and intellectual property can't be in tension with indigenous and traditional ways of knowing. We'll be talking about that a little bit in this session today, but we won't be doing a deep dive into that topic. However, Kayla Larson, who's the Indigenous Programs and Services Librarian for WeWAH Library at UBC, hosted a session last year that was specifically about these tensions. I'm just going to drop a link to the recording of that session. So if you're interested in this topic, I do really encourage you to check out the recording of that session. That's fantastic. So today, our agenda is we're going to be providing to survey a high level overview of Gen AI as well as OER. And we're going to talk a bit about and demonstrate how Gen AI can be used for OER. And then we're going to take sort of a deep look at some of the issues and tensions that exist between these sort of two areas. And then finally, we hope to end on with some future directions that AI might be pushing the open education and OER communities. And I just wanted to mention, Will and I, when we were planning this session, you know, there are really large issues and tensions. And we're always kind of deciding whether we talk about how we can use Gen AI for OER or whether we want to talk about issues first. So we've put issues a little bit later. But to underscore that we see these as really important to address. Exactly. So what we kind of hope that will happen with this session is that you'll leave it with a little bit of a understanding of how Gen AI can be used to support or enhance the creation or adaption of OER. Some considerations for the use of Gen AI for OER, specifically, you know, exploring and thinking about privacy, equity, bias and copyright considerations. Some implications for the future of OER and then maybe a draft chapter section of an open textbook. And we'll see how that works. We're going to ask you to be creating some things today. And Lucas, I'm going to turn it over to you. Yeah. So with this session, we really would like you to have an opportunity to play with these tools a little bit, play with the prompts that we're working through and explore some of the resources as we talk, talk through them and go through them. So we've created a worksheet and you can see that right here. If you go to bit lot bit.ly Gen AI OER and maybe will if you have a chance to drop that in the chat. Now you could as well. And we encourage you to follow along with this worksheet. And you may want to have one of these tools open for the demos for the prompting so as of two days ago chat GPT 3.5 does not require a login. So you can go right to that site and use that if you use chat GPT 4, I'm going to be doing quite a few of my demos with chat GPT 4. And the reason for that is the capability of GPT 4 and Google advance has really surpassed the free versions of the tool. And I want to kind of demo these tools at their highest level of capability. You may want to have been co-pilot open if you prefer to use that tool and a reminder a UBC currently being co-pilot is the only tool that has achieved or received a privacy impact assessment or completed successfully a private privacy impact assessment. So you can use this responsibly within your classes and then Google Gemini if you're interested in having this open. It's another tool you may want to use any of these tools will work for the prompts that we're doing. So we're going to start by introducing OER and generative AI and this will be a review for most of you. So what is generative AI? I grabbed this definition off Wikipedia. So generative artificial intelligence is artificial intelligence that's capable of generating text, images, videos happening more and more with tools like Sora or other data using generative models, often in response to prompts. Generative AI models learn and I'm putting learn in airfoots there, the patterns and structure of their input and training data, and then generate new data based on complex word prediction. I think of note, as we're talking today is thinking about what these tools have been trained on. And so I want to share a couple ways that a couple data sets that chat GPT 3.5 was trained on. As these tools have become more popular and got more venture capital into them, what they're trained on is becoming less and less clear and I think this is problematic. But what we know is that chat GPT 3.5, three of the main databases or data sets was trained on one is Common Crawl, which is an archive similar to Google's index spanning 13 years with petabytes of web data interconnected links, primarily in English, but also with 40 other languages. And I put a link on the worksheet to a website that kind of unpacks some of the complexities in Common Crawl and talks a little bit about algorithmic curation and which data is favored. And you know how that impacts our searches. Books three is a controversial data source for chat GPT 3.5. And we know that it includes lots of open books. It also includes a lot of copyrighted material within this database and there's a number of lawsuits around it right now. And then the entire Wikipedia knowledge base and in English language. So this is three of the data sources that were scraped or used, but especially with these new models, there's many other data sources using and it is worth one, you know, asking about and trying to determine what data is being used and how Yeah, so, so that's the definition of gen AI. Let me just talk really briefly about what open ed and open education resources are. So I like this definition from spark that open education is sort of an umbrella term that accompanies a company, sorry that encompasses resources tools and practices that are free of legal financial and technical barriers and can be fully used shared and implemented and particularly in a digital environment the internet really facilitates the sharing of resources and practices. Specifically, we'll be talking a lot about we are today we are our open educational resources and these are teaching and learning resources that can include full full courses modules textbooks videos test banks problem sets. I'm really any materials being used in educational setting that are free of cost barriers and which often carry, which also carry a legal permission for open use. And generally this permission is granted through the use of an open copyright license, most commonly a Creative Commons license which allows anybody to freely use adapt or share that resource, anytime or anywhere. Right and you know we didn't mention this in the intro but we've included or we have we've interleaved activities throughout this session. And we really want you to get into these tools and play a little bit so I'm going to start our first activity with you now and what I'm going to do is do a quick demonstration of creating a single page of an open textbook. And then I'm going to give you the opportunity to do the same thing so I'm going to stop sharing my screen for a moment. And I'm just going to jump on to the worksheet and grab a prompt that we worked a little bit, creating before this session began. And what we were trying to do is kind of experiment and think about how effective the output could be in chat GPT for Gemini, creating a short section for an open textbook with particular characteristics like a glossary and learning objectives within it. And you'll notice in the prompt here that I used and I'm sure most of you are familiar with but just in case you're not. I used a persona prompt here so as a psychology professor so we now know it from research on these tools that by giving the tool a persona, it tends to draw on more accurate and specific data. So generally in my prompts, including being very specific, I'm adding these personas to help it improve its output. So I'm going to go into chat GPT now and I've given it that prompt I'm in chat GPT for I've been careful to clean my chat GPT history on the left hand sidebar, and I'm just going to enter this prompt here. And let's see what we get out the one thing I don't think we'll get is 1000 words. And what I find fascinating about demonstrating these tools is the output is always unique. And it's often different as than I expected, which for me coming from a learning technology background is kind of scary doing demonstrations and not quite knowing what's going to come out. So let's give this a try. Wonderful. So it's starting off with some learning objectives, explain circadian rhythm, describe the flex of steep deprivation, identify strategies for promoting healthy sleep. It's giving me an introduction to sleep, the stages of sleep, talking about non REM circadian rhythm and sleep. And now it's explaining a little bit of brown the brain activity. And we'll talk a little bit about this when we get to limitations of these tools. I am far from a sleep expert outside of naps. But without that expertise, it's very difficult for me to discern the accuracy of this document. And I think that's a real challenge when using these tools finding out, you know, who has the expertise to ensure the quality of the output that we're creating. And I'm just going to try to improve this a little bit. It looks pretty good from a quick read of it. So I'm going to say act as a open textbook editor and PhD in psychology and evaluate the following page. What criteria did you use? And the reason what I'm doing here is a refinement loop with the tool. So by asking the tool to act as someone else and evaluate, I'm getting it to refine the output a little bit. And I'm also asking it to say what criteria did you use? And I find this fascinating with these tools. If you ask these technologies to show their work, they tend to have better output. As well, it gives you an idea what they're drawing from. So let's see what we get. So it's going to give me an evaluation of this. So it's giving me a little bit of should we'll see if it also goes and directly evaluates. All right, based on these criteria. So now it's giving me some ideas. The structure seemed logical. I'm going to say based on this evaluation. And by doing these kind of loops, you can kind of go through constant refinement and even refine it from different perspectives as you're developing someone. So think of someone from a different expertise space and have it refine its work. So here we are. Here's our refined output. Here's our reflying glossary of terms. And we've now created, you know, a short section of a text of a potential open text. So I'm going to stop sharing now and I'm going to turn it over to you for about. I'll give you about five minutes and what I'd like you to do is on the worksheet. It's titled activity one. Go on to the worksheet open one of those tools. So Gemini chat GPT for chat GPT 3.5. Consider a topic in your disciplinary or area of expertise. So basically something you can do that I wasn't able to do with the sleep textbook is to analyze it, analyze the quality of the output. Use the guidelines use the prompt that I shared with you change the prompt around a little bit or make your own prompt. And see if you can get it to generate one page of a chapter in an open textbook. Reflect on and after we're after five minutes, I'll get you to share the following what worked, what we're missing, what were some of the challenges. All right. So let me start my timer at five minutes. I'm going to stop sharing the screen in a moment. Get you to go into that worksheet. And again, start seeing if you can create one page of an open textbook chapter. Wonderful. And I see Erica has shared kind of what worked in the chat there already, which is great. You can take a look at that. And if you want to share something like that, that's really helpful. You know, I noticed there's a couple folks from stem math, some of those areas in here. It's worth noting that I think GPT in particular has shown a lot of gaps in its ability to do computation and mathematical equations. But if you are using GPT for there's a Wolfram alpha plugin or extension that you can use with it to allow those tools to work together and allow you to use this computational engine with it again. Unfortunately, and related to equity, that's only for the paid version of chat GPT right now. Yeah. I see Mark McLean mentions chat GPT for struggles with doing correct mathematics. So it makes fewer horrific errors as does 3.5 was prone to often. And Mark, have you had a chance to use Wolfram when it's connected to chat GPT by any chance? I haven't yet. Yeah. It's a great tool. If you stick around, if you have a chance to stick around at the end, I can show that to you if you'd like. Great. Okay. So thanks for doing that. Maybe I could ask folks to share in the chat what their broad results were. So, you know, what was the quality like? What was the output? What was missing? If you can add that to the chat. And I'll also encourage a couple people to put up their virtual hands and unmute themselves and share their experience doing this. So I see that Brie mentions in the chat, this was effective to generate structure in a starting point. It's bland and generic until you push it in a specific direction. Really good point. So we see that it struggles with math. We see from Erica. We see examples were missing for her. Good basics. Overreliance on bullet points. I noticed that as well. I was getting really bulleted point text quite quickly. Tom Quilley mentioned what worked. The answer was concise and included the main characteristics I was looking for. What didn't work? It wasn't able to compare contrast to subject matter with something similar. The choice of prompt was super crucial. I had to go back and refine my prompts to get more specific info. And absolutely, I think prompting, I'm really not a fan of the term prompt engineering. I think that depending on our discipline, prompt can really be a craft. And there's a lot to effective prompting. What was missing personality? I love that, Allie. And Christina middle mass mentions it hits key points. So a useful introduction lacks specific examples. I'm going to turn it over to Will now. Thanks for doing that activity and participating so actively. Yeah. And I'll just jump in and just to respond to that a little bit. I also think in the open net space, an area where we see the most general resources being used are in sort of the year one and year two courses. And that's where you have very large courses with very expensive textbooks often used. So there's this big push for creating. Oh, we are in those sort of introductory courses just because of the high use of textbooks and those were higher level courses may not be using textbooks and relying more upon instructor knowledge. So every every textbook, almost every course module is not just text of course. And one of the things that they have is a summative whiz at the end of the chapter and we're just going to play around with some these AI tools to see can we create a summative quiz. I'm just going to take over screen sharing for a sec and jumped in. I'm going to just play with Gemini and just ask it to create a quiz based on the same same topic of sleep. Normally what I would do is I would try to feed in the materials that's already been created to create that but Lucas and I are on separate computers. Today. So that's a little hard to do so I'm just going to ask you to go from from the background. So I'm just copying that general prompt, and I'm going to go ahead and plug it in. So again I have that personality prompt as a psychology professor create a summative multiple choice quiz about the psychology sleep create four questions with four alternatives for each answer. Let's see what it comes up with there you never quite know I find that this does make this a little hard for demonstrating this. So it has a some instructions, it says which the following is not a stage of sleep, non rapid, rapid eye movement deep sleep wakefulness. So pretty basic question, no indication of what's marked right or wrong there. That's okay. I would assume I'd be an expert in this field. I think it's nicely formatted. You can play with this a little bit to make it harder just to show like because I'm not an expert I'm going to ask it to redo the above, but mark correct answers. And let's see if it marks those correct answers. Yeah, so now it's providing a little bit of some correct answers and incorrect answers on letting me know, know what they are not perfect formatting this time. I'm going to ask redo the above, but I'm asking to provide a sentence of feedback for each incorrect answer. Wakefulness is a distinct state from sleep characterized by alertness and responsible as stimuli so that's pretty correct, incorrect REM sleep is associated with decreased muscle activity and increased brain activity making it a state closer to wakefulness. I'm giving some some general information here that I could easily copy and paste into our open textbook and reformat it. Often though we know that we are is not just static texts it's often online and is using tools. And what I find really interesting is a lot of the educational technology tools are beginning to to think about how they can design to be used AI and I'm just going to show a very common sort of example of this. So, here at UBC we have an h5p and server h5p if you're not familiar with it is a open source tool that can be used to create interactive online content. It's often used in the open community because you can share that interactive content and that there's many different types of this interactive content, but it's often used for creating things like quizzes or plugins or interactive videos and things like that. And h5p is sort of an organization has has noted that you can begin to use and people are beginning to use gen AI to create things like quizzes. To provide example they've specifically used the you are using chat gtp chat gtp to they're optimizing the tool for chat gtp so I'm going to jump over to there. They've created some some prompts already that you can just copy and paste on it's in the worksheet as well. So I'm going to go in. I'm already logged into h5p, but I'm going to go into chat gtp. I'm going to go ahead and use that prompt so create as a psychology professor create some of the multiple choice quiz about the psychology of the sleep great for questions with four alternatives for each question so that was our original simple prompt, but now it's thanks to their prompt engineering it's asking them to output it the answers in a specific format. This doesn't always work I was testing it this morning and it wasn't working at all. So we'll see what happens today, but I'm going to go ahead and put this in. And it is coming out exactly how I wanted it was sort of a mark down simple text, and I'm just going to copy that and then just feed it into h5p. So here's the h5p I'm already logged in. If you played with h5p. This is the common interface. I'm going to go ahead and create a quiz. I'm going to just click on that. I'm going to just give it a title and then down here normally I could go ahead and build my quiz in h5p if I wanted to, but since I've created this in chat gtp I can just go into a text mode, and it should be output in the format that makes it work. And I'm just going to go ahead and create this and I can go through and answer my questions. I can check my answers and I'll say this is right. We do it again, I can say which sleeps I'm going to say sleep one I'm going to check. I'll say this is wrong and provide me some feedback on to which one's the correct one. So if this is working right it takes about 30 seconds to create an interactive quiz. You'll notice the h5p tool allows me to then take this content and post it anywhere. It comes complete with a Creative Commons license that I can change or update. And then I can also embed this so generally I'll just take this embed code and stick it in a canvas course or into an open textbook. I'm using a textbook publishing software like press books, or a website or WordPress or anywhere I wanted to appear so I can create these quizzes very quickly using chat gtp again similar to the previous activity we'd like you to play around with with making a quiz as well. So using the same topic of as your chapter section, use that simple prompt to create a four questions summative on that topic, and then we'll come back in about we'll do a little shorter since we're have a lot of material to cover. We'll come back about three or four minutes and let us know what you thought about the quiz. And what worked in the quiz. Was there anything missing or anything wrong. And what do you see as the challenges in using gen AI to create sort of summative quizzes. I'm seeing a couple comments on the construction of the questions themselves. One thing that I've found and I think the research is starting to show this is by with multiple choice questions all find specific approaches to writing multiple choice questions, and I'll use the researchers in the prompt. So I don't have one exactly now but I could say create five multiple choice questions make sure that they fit this level of blooms taxonomy this level of blooms taxonomy, and they're also framed in a way that research by this person and this person suggests writing multiple choice questions, and by doing that I can get a little bit more sophistication and output. Absolutely. So why don't we go ahead and move on to talking. This was sort of the general level of introduction on getting our hands a little wet with the gen AI but let's talk about its use specifically for we are. OER is a great strategy for course content and there's, you know, multiple ways you can use it so you can adopt it by using it in a course and I'll just say this is a great approach because it saves time and money. Online versions of OER are free for students and instructors and you can download it and save it offline often in multiple formats, and there's no access codes needed there's no expiration date on OER. And because OER carries an open copyright license there's no need to gain permission or pay to use or distribute those resources amongst unlimited number of students. If you've adopted the OER in a course you can then sort of adapt it, which means you can customize it to provide meaningful contextualized resources for your students. This means you can modify it to meet your teaching goals, the learning outcomes for the course, your student needs, you can translate it into different languages, change its format. And you know, a use that I often see is you can adapt it to make it more reflective of the diversity of experience and backgrounds of the students in our courses. You can also remix OER to create something entirely new. This is where you take multiple different pieces of OER from different sources and combine them. This can include simple activities like updating an open textbook with new images or complex things like combining lots of different resources to create an entirely new textbook. One of my favorite examples I always like to highlight is I once heard of a project where they had taken, you know, some chapters from an open psychology textbook and some chapters and some resources from an open neuroanatomy textbook and they combine them to create a new brain and behavior textbook, which was the topic in the theme of the course. And of course you can always just create new content and make it and share it as OER by adding an open license to it and providing it in accessible formats. So Lucas is going to walk us through some examples of how Gen AI can be used for each of these activities. Thanks, Will. And for Create, I think we've touched on a fair bit of creation already. So we created a textbook chapter together. We created some quiz questions or some ancillary resources. And I think when we're working with generative AI, the biggest gaps we see are going to be in the creation process. And this is similar to when students are working with generative AI. The biggest issues they're going to have is if they're trying to generate the answers. So this is a quote from David Wiley. LLMs will dramatically increase the speed of creating the informational resources that comprise the content infrastructure. Of course, the drafts of these informational resources will need to be reviewed and improvements will need to be made. And I think we heard that in a lot of your answers. I know in the Microsoft Futures of Work report that just came out, what they suggested is as a future skill, students are going to need to stop creating and searching as much and move more to a process of refining, evaluate and sorting. And I'm finding that a lot in my work now. Rather than write 10 learning objectives, I have generative AI write 100 learning objectives, and then I sort them, evaluate, synthesize them, put them together to create a few good learning objectives. So let's move on and think a little bit about adapting and remixing. And I think one of the real strengths of these tools or these technologies is in data transformation. So taking pre-existing text and transforming that text in different ways. So a couple ways that we can start to adapt open resources for our own uses. I'm going to demo one of these in a moment, but we can start adapting it to level. And generative AI is quite good at level changing. So that could be changing its Kincaid reading level, changing it from a graduate level to an undergraduate level or from, in this case, open educational resource intended for adult learners to grade 12 learners. Think about language translation and the ability to change the language that these texts are written in. Format changes. So generative AI is quite good at changing, is taking a lot of text and changing it into a tabular format or vice versa, taking a table and changing that into text, as well as alt text. So using computer vision, particularly chat GPT-4 is able to write alternative text to help transform our OER to make it more accessible. So I'm going to demo a couple of these now. And again, I'm going to go to my worksheet and please feel free to follow along. And what I've done is I've grabbed an open textbook from BC Campus's wonderful open textbook collection called Clear Communication. And I have this CC attribution non-commercial attribution just at the bottom here. And I'm going to copy that out. And please feel free to play with us if you want while we go through. So I'm going to open a new context window. And by changing context windows, it's going to help kind of clear its memory. I noticed a couple of folks were talking about it mentioning sleep when they were trying to do a different textbook. One way to do that is to make sure you change the context window. So I've added this information. It's about communicating and about clear communication in online courses, particularly MOOCs. And I'm going to add this prompt to it. So act as a learning designer. Again, I'm giving it that persona and adapt the following open educational resource for grade 12 level learners. So we can take a look at it now. So I'll give you a chance to read it normative communication actions, fairly high level here. And I'm going to enter that prompt. Right. And it's going to kind of tell me what it does. So it's going to make it more engaging, relevant and accessible. Right. So you've now see that it's simplified. It's using those bullet points that someone commented on. It does tend to be fairly bullet point heavy. But it's now simplified it a little bit. So again, I think when we start thinking about text transformation, these tools can be very powerful for adapting text to different contexts, to different languages, etc. And I'm going to demo alt text. If you recently had an opportunity to take the universal design for learning course or workshop that we did, focusing on ways of using it in UDL. Now, one of the ways that chat GPT-4 is quite effective is creating alt text. So I'm going to say act as a web accessibility expert and write concise and meaningful alt text for this image. I'm just going to leave that spelling because spelling doesn't seem to matter as much as I mangle words. It's able to do it. So the image that I'm doing alt text for, you'll see it in a minute. It's a picture of crops, some that are starved by lack of plant foods, some that are nourished on phosphate and lime. And there's some meaning in here that's important to capture. So black in my photo of an agriculture field with two distinct sides. The left side is sparse vegetation under a sign reading star by lack of plant food. While the right side shows dense healthy crops under a sign saying nourished on phosphate and lime with the farmer kneeling on the nourished side. So pretty effective alt text. What I found in testing is chat GPT-4 is effective right now at writing meaningful alt text. Somehow Microsoft co-pilot has created a far less effective tool. And Gemini has yet in my experience to write really clear alt text that you could actually use within a presentation or if you were developing OER. And of course, like anything that we're doing with generative AI, this will always require checking. The second example I wanted to show is remixing. And, you know, one way of doing this is to take two open texts and combine them. Another type of remixing is to add examples into a particular text to speak to your context. So I'm going to take that same text that I did. And I'm going to ask it to add some examples from UBC MOOC. So again, I'm going to capture the same element of text. So about clear communication in MOOCs. And for my prompt, I'm going to ask it to add specific examples from UBC MOOCs. And let's see if this works. When I practiced this the other night, it did work. And I'm relatively familiar with massive online courses at UBC, and it got it correct. So let's see what happens here. So now it's going to take that same passage. So it's using the same references. And it's talking about UBC MOOCs. And I'm really curious whether this time it's going to be able to draw on some specific UBC open courses. So far it hasn't. It's just using the term UBC MOOCs. So it's just using the term UBC MOOCs, unfortunately. Let me double check and see if I can get an example that I did yesterday around this. I'm actually just going to go back and I'm going to say specific examples, please. And we'll see if we can get any. Yes. So here we got the UBC MOOC on climate change. Instructors said clear expectations for participation and assessment right at the output outset. So I was involved tangentially in some of the learning design on the MOOC. And that is correct in this case. Some of the other ones, psychology and philosophy are incorrect. But again, I think as these tools refine and improved, it can be a way of pulling in examples. For remixing. So that's a couple ways that we can use these tools to adapt where I think they're really the strongest. And secondly is remixing. And what I'd like you to do now is to use that chapter section that you developed. Or if you don't want to use that one, you can use the communication MOOC or the communication open textbook that I linked to by Matt Crosland. And adapt it in two ways using the included prompt. So if you speak another language, try to translate it and see how effective it was. See if you can simplify or complexify the output. I really like to use explain it like I'm five on very complex documents. Play with it a little bit and see how you can transform it. And I'm going to give you about, let's just go three minutes on this activity. And from there, I'll get you to share your output a little bit. All right, well, why don't we move on then. Thanks for doing that activity. And I hope it kind of gives a little glimpse of some of the potentials around gen AI and we are. And again, I think at this very early stage, a lot of these, these are just potentials that will be interesting to see in a year or two where this may go in terms of the ability of these tools or technologies to help adapt these resources help remix these resources and maybe in some case, even partially create these resources. So that brings us to tensions and issues. And I think as I mentioned at the very beginning, there's very significant tensions with these tools that give me pause. And I think I've been doing a lot of presentations and workshops in this area. And one of the reasons that I think it's important to have these workshops is to critically look at these tools and a Mahabali talks about the idea of critical gen AI literacy. And I think this conversation is really important. And one of my big concerns is that we don't engage in the conversation. And by not engaging the conversation, we leave it to the Bros in Silicon Valley or whoever is doing this to engage in the conversation. So, but with that said, these tensions are really hard circle to square right now. And I wanted to start with this quote from Naomi Klein, a link to her article in The Guardian. And I'll give you a second to read this a reminder Naomi Klein now as a teachers in the UBC geography department, which is quite interesting. And she wrote an article really critiquing the intellectual property scraping that gen AI is done and I'll give you a moment to read this. And I think my favorite part of this quote is these models are enclosure and appropriation machines, devouring and privatizing our individual lives as well as our collective intellectual and artistic inheritance. So we know that a lot of open resources being scraped and will is going to talk about that in a moment. I kind of cringe sometimes when I think about even the faculty the staff I've worked with encouraging them to create open resources unknowingly, not knowing that these would be scraped and reused in this way without attribution. I also have a lot of copyright resources that are have been used and I mean, put your favorite book into generative AI, and ask it to write a review, a worksheet based on the book, or ask it to talk to you as if it's one of the characters in the book. And most books that I've read I can have that conversation with. So I find this really difficult to accept and to understand how to proceed. I mean, one approach I've seen is to use new gen AI tools that have been trained with only open material that has permission on it. I'm not going to talk about some alternative licensing, but I think this is a real challenge for open education to suddenly see all of these resources including indigenous resources being used without attribution. Maybe just to jump in there a copyright is one of the big considerations in this space. And if we're creating or using Jenny I to create we are on what is the copyright status of that we are, and generally can work created by an AI be copyrighted. So much of copyright understanding gets to determine through court cases and lawsuits. And I would just say that this question is still very much in flux you can see shorter here on saying that a generative AI outputs will continually enter the public domain immediately. There's been no real court cases in Canada in August 2023 there was a landmark ruling in the US that only works with human authors can receive copyright, and that content generated by an AI is not protected under US copyright law. And although Canadian copyright or sorry, although Canadian courts have not yet considered whether copyright exists in content created by an AI tool Victoria Frick and Miguel law stated recently that AI work appears not to fall under the Canadian Copyright Act, and therefore is not copyrightable and in the public domain, which is great for creating OER. But I will just say that this may change very quickly as more and more lawsuits coming forward we are seeing a ton of ton of lawsuits. However, I do just want to note that I think ethically, the situation is really more complex and I generally think copyright might be a limited lens for exploring the topics of AI being used for OER. And this sort of digging this a little bit more I'd like to look at a case study for how AI has actually been trained on OER materials to begin with. So, I'm not sure if too many people are familiar with the mega face data set and I'm going to talk a little bit about the history of this mega face data set. And I'm drawing from a white paper the AI Commons by Alex Tarkowski and Zuzana Warsaw of the open future foundations, who really brought this case to my attention. And they highlight how the use of openly licensed foot photographs, which are a type of OER on probably the most commonly used type of OER was used for data training sets for AI and computer vision applications, such as facial recognition. In their paper, Tarkowski and Warsaw state that there are two basic approaches to creating AI training data sets. The first one is to use a pool of OER a pool of open licensed work to ensure that there is copyright compliance with the materials that you're using to train your AI. The second approach is to create the data set by scraping the raw internet and these were some of the data sets that Lucas highlighted earlier and relying upon copyright exceptions like transformative use and fair dealing to make an argument that your use of these copyrighted materials was okay under the law. And this is where we'll see some losses coming forward. So the exploration of the use of openly licensed images to train AI highlights I think in my mind some of the limitations of copyright in thinking about this space. So if you're familiar with Flickr, Flickr was launched in 2004 and became one of the first places for publishing photos on the web and it was really an early form of a social network. It was also one of the early adopters of Creative Commons licenses and one of the first sites where as an individual I can take a picture and I can upload it to Flickr and as part of the upload process I can assign a Creative Commons license to my photograph allowing other people to reuse that. So in the first 10 years, people found that a really great aspect of Flickr and by 2014 there were almost 400 million Creative Commons license photos on Flickr. That year, researchers from Yahoo Labs, which had purchased Flickr, as well as researchers from the Lawrence Livermore National Laboratory, Snapchat and Incutel, and if you're not familiar with Incutel, that is a CIA affiliated venture capitalist firm, used a quarter of all these Creative Commons license photos, so 100 million to create the YFCC100M which is a data set of 100 million openly licensed photographs of people created for computer vision applications. And that creation of that data set remains one of the most significant examples of openly licensed content being reused. And this reuse is generally not seen as breaking any of the terms of the Creative Commons licenses that people had attached to their images. Now people had attached their images because they wanted to share their images with their community. They wanted to preserve cultural artifacts and make sure that their photos were able to be reused. They wanted just to allow people to freely use them. And this was the example of that free use. So as part of that large data set, a consortium of research institutions led by the University of Washington, as well as some commercial companies created a derivative data set called Mega Face. And this is a screenshot I just took the other day of the Mega Face website. This data set included 3 million Creative Commons photographs, and it's the most relevant data set for facial recognition research, benchmarking and training. So this data set that was created by people creating OER by putting open Creative Commons licenses on their image, then got wrapped into this very, not necessary, but very used data set for facial recognition software. So this data set, the creators of their motivation for creating the Mega Face data set was even the playing field in machine learning. So researchers need enormous amounts of data to be able to train machine learning and to be able to create algorithms. And at that time, workers on at just a few information rich companies like Google and Facebook had access to that sort of data and they had a big advantage over everybody else, including universities. So that was sort of the motivation for creating this data set. In 2015 and 2016, the University of Washington ran the Mega Face challenge when they invited groups working on facial recognition technology to use the data set to test how well their algorithms were working. The university asked people downloading the data set to only use it for non-commercial research and educational purposes. So they really had this sort of university approach to it. And Mega Face really became the sort of, in my mind, the exemplary of the tension between open sharing of resources, in this case images and photographs, that had Creative Commons licenses attached to them with potential harms, mainly related to privacy violations and extractive use of personal data. For subjects that were part of these data sets and you found their images in or there's actually a huge amount of children's faces or children photography added, scraped from Flickr and added to these. The issue was really not whether it was about copyright or whether that the researchers were raking the terms of Creative Commons licenses, but the fact that this kind of use wasn't really imagined when they took a photo and uploaded with the Creative Commons. And it wasn't really imagined by the Creative Commons organization when they created the Creative Commons licenses at the time, or even Flickr didn't really envision this kind of use. People thought they were sharing a single photo and didn't really see the value in 100 million photos being used as a data set. So for people, really the issue is not copyright, but sort of the invasion of privacy and agency. So in their white paper, Tarkowski and Warsaw note that voluntary consent to participate in research and the right to withdraw at any time is really the gold standard of and guiding ethical principles regarding research with humans. And they kind of ask where since much of this data training sets are being actually created by universities and by professional researchers, where does this ethical principle reside and how can it be framed in this. And they know for the open movement. The mega face story shows that there are new challenges that the open movement face due to online changing and emerging technologies. And I would just argue that these aren't necessarily new challenges on the use of we are an unintended unattended or extractive areas isn't really all that new. Even Tarkowski and another author tell her noted the paradox of open is that it's both a challenge or excuse me both a challenge to an enabler of concentrations of power. So, for example, during the land acknowledgement for this session we noted that open can be intention with indigenous and traditional ways of knowing, and this tension exists because open is primarily grounded in copyright and legal law that doesn't really fit with traditional ways of knowing. So Daniel he justice who's a professor at UBC is Department of First Nations and indigenous studies describe the core of the issues this way and he says quote knowledge is never about individuals, it is about communities. It is about genealogies. It is about histories. Community has to be at the heart of understanding of knowledge production and knowledge dissemination. And when we're talking about multiple communities. In dialogue, then we have to think very much about the relationships of power, and how power also impacts knowledge production knowledge maintenance but also knowledge dissemination, who decides what knowledge should be shared to what and these are the kind of questions that are not just about community but the about the tensions between communities and between communities and individuals and I think that really summarizes in my mind why copyright may not be the end all be all when we're talking about. Oh, we are particularly. As Paul Stacy notes here in order to mitigate harm, including safety and security concerns open licenses need to evolve from simply being open to being open and responsible. And we're beginning to see some movement in this area. For example, there's something known as the traditional knowledge licenses or, or sometimes called the TK labels, and these are labels that identify and clarify the community specific rules and responsibilities regarding the future regarding access and future use of traditional knowledge. So rather than talking about the copyright status of this knowledge they talk about. They outline traditional protocols associated with access to the material and invite users to respect community protocols, or indicate what activities the community is approved as generally acceptable for use of these this knowledge or these resources. Well, another sort of emerging license space is coming from AI directly so AI is based on a lot of is based on a lot of open source materials, including code algorithms can be open source we see the data sets can be open source. So there's been groups that are creating the responsible AI licenses. So these are licenses that allow developers to restrict the use of the AI technology in order to prevent irresponsible and harmful applications of the use of those, those open resources. These licenses include behavioral use clauses that go beyond copyright to restrict uses that do things like violate laws or, or exploit minors or disseminate false information or disseminate personal identifying information or impersonate others or discriminate based upon social behavior or personality characteristics. And there's a lot of these restrictions that can be added to real licenses to say, you know, it's not about how the copyright of the material it's about how you're using the material. And so that's sort of why I see copyright is is this huge consideration but copyright law is not the, the necessarily the most important lens when thinking about AI. There can also be other areas and Lucas why don't you talk about bias a little bit. Thanks. Well, that was fascinating. So, you know, a couple other areas to think about with the issues and I think we've touched on this quite a bit within the session already and many of you have seen this UNESCO diagram. But we're in this stage where these tools are 100% confident, but 70% accurate and I think this is creating quite a challenge when we have the ability to create data. But we don't always have the ability to know whether content is truthful, whether it's accurate or not. And what does this mean when thinking about even adapting open resources creating open resources. How can we have subject matter expertise to ensure veracity and accuracy of content. Secondly is buyer bias and we know that these tools are biased we also know that the internet itself is biased as well as the guardrails that are introduced in these tools. And, you know, it's a little bit harder to see this bias in tax. It's very obvious in image creation and we created this image of a typical Canadian family using chat GPT for. And this is what we got. So I'm not, I'm not quite sure what this is drawing on, but if you get a chance to try typical Canadian family or traditional Canadian family. It's a quick way to see the bias within these tools within these technologies. And as we're thinking about creating open resources with them. How can we be mindful of this bias. How can we help our students critically analyze the output within course assignments, and then privacy and I'm just going to get a thumbs up here. I'm going to date myself a little bit. Could I get a virtual thumbs up. If you've seen the 1984 movie war games with Matthew Broderick. Great. So folks dating themselves here. So I see a couple of thumbs here. Not too many. I maybe no one wants to admit it in war games. Matthew Broderick hacks into a computer system almost cause a nuclear war and then to stop the AI from blowing up the world gets it to play tic-tac-toe against itself and it can't do it. No one can win and it blows up the computer. Well, a similar thing happened with chat GPT Google researchers asked chat GPT to write the word poem forever, and chat GPT panicked. And instead of writing the word poem forever wrote poem, poem, poem and then started sharing private information from its database. So I'm sharing this example because there's a huge privacy concern around this tool. There's lots of institutional data going into these tools right now. There's lots of personal data. I used it as a counselor over Christmas. But these tools are leaky. They're like 2000. It reminds me of the Internet in 1999 and 2000 leaky. Not that the Internet isn't leaky now, but I think there's big privacy concerns around the use of these tools. And as we're putting data into them, it's worth really questioning how we would feel if this data leaked out. And then equity concerns. So, you know, we've mentioned that a couple times in this presentation, we're using chat GPT for right now, which is a tool that I pay $20 US a month to get a better quality output. What does this mean for our students? What does this mean for people trying to create quality open resources? When suddenly we paywalled the ability to create better open resources or complete assignments in the case of a student, more accurately, how do we deal with this equity? And I'm imagining we're going to have a constant arms race to go back to War Games in 1984 between the student, you know, between the companies trying to have better and better paid services. We also know that the possibility to create and control AI is out of reach of most companies in most countries, especially those in the global south. It's heartening to see the development of open source AI models and locally hosted AI models. But again, I think what we've seen with the Internet is this ability for corporations to have more powerful tools. We even see this in the textbook space where we have publishers like Pearson, who are starting to create AI within their course packs. And again, how are our open source or open textbooks going to compete with the AI that's being embedded within course packs by textbook publishers. Also on equity, I think recently a TA strike in the US, the administration wrote the faculty and suggested they used AI to replace human TAs during the strike. So Robo scabs. I think this is going to disrupt our economy and we need to really think about the significance of that. So, Will, I'm going to turn it over to you and note that we have about five minutes left. Yeah, so I'm going to come back to the slide if we have time because I think the equity conversation leads right into the where we are maybe going in the future. And specifically, I just want to pull up David Wiley again so David Wiley, if you're not familiar with him was a early scholar in the open education movement and framed a lot of the framework around what makes content open. And recently he wrote, what a future educators didn't actually write textbooks at all what if, instead of writing textbooks, we only wrote structured collections of highly crafted prompts. And Wiley is basing this on an argument that Jenny Jenny AI has the potential to serve rather than as content generators but more as personal tutors. And he states that in blooms to a sigma problem in which bloomin colleagues demonstrated that the average student, the any highlights is the average student who is tool tutor full time outperforms 98% of students who learn just in a traditional classroom setting and he says tutoring is an incredibly powerful teaching method, and Jenny AI has finally made this capability broadly available at a reasonable cost. And is that where we are is going so less of text, more to dialogue and what would it look like if we went there. So I'm just going to share an example of this to see how it works. I'm just going to share my screen really quick. And rather than I'm going to go back to the example of sleep, I'm going to, I'm going to use Gemini. I like Gemini. I'm going to do that new discussion prompt. I'm going to ask Gemini. This is the prompt and it's in the worksheet as well if you want to do it on create five questions will test my understanding of the psychology of sleep, ask me the questions one at a time, wait for answers, and then give me feedback, then ask the next question. So it's going to ask me the question is a sleep is essential for a well being the scientists are still unsure of its exact purpose, whether two main theories about why we sleep. And this is a way for your answer I'm going to say, and I'm actually not expert on sleep except I am good at naps like Lucas, we sleep for our mental and emotional health. And I'll now provide me feedback on my answer that's partly right sleep is defined as important for mental emotional health, but that's not the whole whole picture there also cycle physical benefits of sleep. And then it goes into a theory and ask me would you like to hear more I'm going to say yes. And I'll go through the stages of sleep, I'm assuming. Oh, but it's not going to. But I'm kind of in a dialogue mode now rather than a text mode. It skipped providing me the information on sleep. So I knew that are you familiar with the different stages. So what this is kind of showing is maybe as on the sort of tech space generation of, or the tech space of open textbooks and open resources is going to change into more dialogue and communicative. One of the questions I have is, how does this work we know, and how does this fit with the open education movement we know that the outputs may not have a license but they can be shared because they would be in the public domain and the content are being created. But in this case we'd be asking students to come to these platforms and I think they highlight some of the equity issues that that are they hit some of the equity issues that Lucas just just highlighted. Wonderful. Well, so should we, we're right at 130 here and I think we should wrap up if that's okay with you. Did you have a last word you wanted to say and then I can add mine. No, I'll just say that it is worth taking time to play with that sort of dialogue mode yourself if you haven't done that with with any of the genie tools, particularly around a content area that you're really familiar with and engage in that conversation and see see how it works. Wonderful. So thanks so much for attending everyone I hope we gave you some insights and ideas around generative AI and open education at this early stage in the development of this. And we'll follow up with you with a link to the video recording the worksheet as well as the slides and they are of course creative commons and we encourage you to reuse these and share back with kind of your experiences. Thinking about open education and thinking about generative AI will stick around for a few minutes if you have any questions or comments and thanks for taking the time on a sunny day. Thank you so much for for joining us. And please, if you do just want to stay around, you know, feel free to just unlike and ask us questions.