Building AI Responsibly: What the EU AI Act Means for Teaching and Learning

Hello, everyone. I think, we are looking at sharing there we go. Thank you so much. Hi, everyone. Welcome. Thank you for joining us, and thank you for watching us back in your free time as well.

We are having this conversation today for one very, very specific, quite urgent reason, August twenty twenty six. We are exactly two months away from the EU AI act's primary transparency and compliance enforcement deadlines for high risk systems coming into full effect. This isn't theoretical anymore. It is our shared reality right on our doorstep. If we can go to the next slide, please.

I am Frank Colombo, one of the Instructure's customer success managers here in EMEA and I'll be setting the stage today. Joining me are three phenomenal leaders from our Instructure team, Jody Saylor, who bridges the gap between our product vision and pedagogical best practices, Zach Pendleton, our chief architect who leads the engineering of our AI solutions, and Melissa Lobel, our chief academic officer here at Instructure for all things pedagogy. We do have a short q and a space at the end of today's session, so please circulate your questions in the dedicated q and a area and we will try to answer them at the end. Now let's address straight away the elephant in the room. Jody, Zach and Melissa are joining us from our American team.

You might be wondering, Fran, why are your US based leaders interpreting European law for us? The answer is simple, they are not. They're here because the EU AI act isn't just a regional hurdle for Instructure. It has fundamentally shaped our global AI architecture. Having our chief architects and pedagogical leadership here is a signal of how deeply this regulation resonates across our entire organization. These aren't just guardrails that we bolted on after the fact.

These are some of the foundation pieces of how we decided to build our Ignite AI suite in the first place. So let's delve deeper into what we will be covering today. So first, we are going to demystify the EU AI act and what it specifically means, for your tech stack. We know the August twenty twenty six deadlines, for risk for high risk systems are top of mind at the moment. Our goal is to break down these regulatory requirements so that you understand what that means for teaching and learning environments.

Secondly, Jody will help us bridge the gap between these new regulations and everyday pedagogy. We are going to show you how regulations and good teaching actually go hand in hand, and you'll see how our closed loop system ensures that your institutional data remains yours alone and is never used to train external models. Finally, Zach will chart a secure path forward by showing you some of the Ignite AI in action and how Instructure is prioritizing the AI advancements. Most importantly, we want you to we want to make sure that you know how to take advantage of our global free access period running through September the thirtieth, of course, twenty twenty six, so that you can start or continue trialing these tools with peace of mind. Before we dive in, I want to establish clear boundaries here.

I'm not a lawyer. Instructor is not your legal counsel. The information that we are sharing here today is meant to empower you, but it must be followed by crucial conversations with your internal IT, leadership and legal teams. It is your responsibility ultimately to assess that your local compliance is met and please take take away any any learnings today and back to your teams. One of the first thing that your legal team will tell you is that the EU AI act is extraterritorial.

Now this means that you do not have to be physically located in Rome or Berlin to be impacted. If you are an institution in the UK or the Middle East or even Australia and you have distance learning with students residing in the EU or if you use AI tools to process the data of EU based staff, this regulation applies to you. So this governs the market and the end user, not just where your headquarters are located. It is truly a global standard that is kind of hiding inside a a regional law. If we go on to the next slide.

So what actually is the EU AI act? So for those on the call today who haven't been glued to the regulatory tech news for the last two years, Let's break it down quite simply. The EU AI act is the world's first comprehensive legal framework for artificial intelligence. Its goal isn't to ban AI or to stop innovation. Its goal is actually to make sure that AI as AI becomes a bigger part of our daily lives, it respects the fundamental human rights, safety, and transparency. A helpful way to think about it is like a digital road safety training.

The axe recognizes that not all vehicles on the road pose the same danger. An AI spam filter is like a bicycle, low risk, but an AI system that screens job applicants or grades student exams, that is heavy machinery. The law says that you cannot just hand someone the keys to heavy machinery and hope for the best. You need a theory test, what the act calls AI literacy. So users understand the rules of the road and you need specific driver training, what the act calls the human oversight so that the person behind the wheel knows exactly how to hit the brakes if the machine veers off course.

Next slide, please. To enforce this road safety, the act categorizes AI into four risk levels, minimal, limited, high, and unacceptable. Some things are entirely banned. The unacceptable risks, for example, you cannot use AI for emotion recognition or biometric categorization in the workplace or the classroom, full stop. But why do education tools often land in the high risk category? Under the act, AI systems used for education and vocational training as well are designated as high risk when they go beyond simply simple delivery and actively manage an educational path.

If an AI evaluates what a student has learned, and uses that to steer that professional, future or livelihood, then the EU categorizes that as high risk. And honestly, we agree. The stakes in education are really, really high. So that's why the requirements for these systems, transparency, human oversight, data governance, and AI literacy are so stringent. Next slide, please.

If looking at these requirements feels overwhelming, I want you to take a breath because compliance alone is not our goal today. Our goal is to actually show you that these regulations reinforce what good pedagogy already demands. The transparency, the human in the loop oversight, the data privacy, this is exactly how we designated our own Ignite AI suite. Of course, we could say that today we have done we will have done our job if you leave this session feeling confident in the architecture and safety of our tools that you feel compelled to start and keep trialing Ignite AI at your institution and that that you don't have to build a compliant AI ecosystem from scratch, that we are here and we have been investing our resources and keep investing our resources to make sure that we keep building one for you with the EU AI act in mind as well. And soon, we will be able to present you with the evidence, so do watch this space.

We want you to, test it, poke holes in it, and see how it solves real day to day challenges for your educators and students and without compromising on security. And with that, I will hand it over to Jodi to explain how these regulations connect to good teaching, and then Zach will show you how we, in structure, approach AI development. Jodi, take it away. Thanks so much, Fran. Before I just jump right into the specifics, I want to make one big point.

A lot of institutions and organizations treat the EU AI act as a compliance problem to solve. And as Fran kinda mentioned already, we think that's the wrong frame. When you read what the act the act actually requires, it's not asking anything new of educators, actually. It's asking AI to meet the standard that good teaching already sets. So as we look at the next slide, let's walk through specifically why.

The act requires that AI driven decisions be explainable. So similar to what, Fran, you are already saying, People know and must know when AI influenced an outcome and can ask why. This likely sounds familiar. As educators, we have always understood that students learn best when they understand and know how they are going to be assessed. What are we looking for? Do the rubrics, the feedback, and the great explanations actually show what the student needed to know? These really are not new concepts.

These are things that we've always done as educators. The act now is just extending that same expectation to AI. If a tool can't tell a student why it recommended a path or why their, work was flagged, it doesn't meet the bar that good teachers have always held. On the next slide, we see also that there are two specific requirements that fit naturally together here. The first, as mentioned by Fran, is human oversight, which is really the idea that a person must be able to review, intervene on, and override any AI decision that affects a learner.

So keeping that human as the in that driver seat. Second is bias and fairness controls. These requirements are that the AI systems work equitably across different groups of people. So educators already are carrying both of these responsibilities. They have forever.

You interpret the data in context. You notice when a struggling student's numbers don't tell the whole story. You, as the educator, advocate for equitable access. Also, things like universal design are for learning aren't new. They've been the cornerstone of good practices for decades.

The act now is just making it a legal expectation for AI as well. And on the next slide, we see a couple of other obligations that you are likely already working on. So first and foremost is data minimization, which refers to only collecting what you need and protecting what you have, and that maps directly to our student privacy practices that institutions already follow under GDPR and similar frameworks. If your privacy policies are solid, you're already thinking in this way anyway. And then the second is AI literacy.

This is an obligation that's really worth spotlighting in my opinion. Article four of the act requires that anyone deploying AI has sufficient understanding to use it responsibly. But there's a bigger opportunity here than just compliance. Building AI fluency into how your institution operates for faculty, for administrators, for your academics, and also for students themselves is how you turn a regulatory requirement into a genuine institutional strength. Digital fluency alongside subject matter expertise is what learners will need.

We, as educators and educational leaders, have a role in building that. So these components, the privacy, human oversight, equity and fairness, and AI literacy are not just regulatory checkbox for us either. They are things that we are considering and thinking about as we make design decisions. And with that, I'm gonna actually hand it over to my colleague, Zach Pendleton, to show you exactly how we tie these to our AI development processes. So, Zach, over to you.

Fantastic. Thank you so much, Jody. Thank you, Fran, and thank you all for being with us this afternoon. I want to talk a little bit today about how we develop software and how we think about the development of AI features, because I I think that sheds some light, on how we approach, both the EU AI act, but also the responsible use of AI in the classroom. So all of our AI features, are built on a foundation of data privacy and of regionalization.

That means that, anytime we deploy a feature that uses AI, we give you the same guarantees about where that is hosted and how data is used that you have with the rest of campus. So in the EU, that means that you're hosted in the EU, and all your AI, features when you use them are hosted there as well, right alongside your data, to ensure that, you're not dealing with data crossing borders, or AI systems that exist outside of, the EU's jurisdiction. Now, on top of that, what you'll see frequently inside of Canvas, is this idea of AI, overlaying or augmenting, your existing workflows. I think this is, really important for a few reasons. First, I I think it makes these features easier to find.

If you know how to do something in Canvas, and there's an AI option, for that, you'll see that presented to you right in the UI where you expect it. And that's going to be, consistently branded inside of Canvas as well. You know, as was mentioned earlier, transparency is a really important component of the EU AI Act. And so we we want to, be as clear as we can be, that AI is being used, when it is being used so that you, as an educator, can make the choice to to use that or to skip it. The other really powerful thing about having AI right in the existing workflows is that it gives you a standard pattern for asserting your autonomy and ownership over the or ownership over the outputs.

So in this example, for instance, you know, the AI outputs show up, in the same forms and fields, that a human would have typed them into, making it very easy, for you to edit those or update them or, you know, even reject them if that output is not something, that you want to move forward. Another thing that is really central to our, our commitment to transparency, to openness are our nutrition facts. So these are small model cards like the one you see here, that are available anytime we use AI inside of Canvas, and they're going to share with you the information that you need to make a decision about whether that AI feature is right or wrong for you in that moment or in your classroom. So that includes things like, the name of the model, what data is sent to it, and how that data is used, at what the expected, results are of using that feature, but also what the expected risks may be. Because, when we use AI, we know that it's not deterministic, which means I may get two different outputs, with the same input.

And so we have to expect that there may be cases where the output isn't something I want. And understanding what the expected risks are of the feature, I think, can help us make better decisions. See the slide. There we go. And, because not every feature is going to be appropriate for every school, for every classroom, or for every instructor, it's also really important that once you have that information, you make the decision, you then have the tools to enable or disable those features.

So all of the Ignite AI features, are off by default. But can be enabled, and configured at the account level and at the course level. And that means that you have, maximum control over, what AI looks like in your Canvas account and can even go so far as to allow it for instructors that are interested in it, but also allow instructors who would prefer it off to turn it off inside of their courses. And all of this, you know, I mentioned it was built on a foundation of, a, yeah, a strong data residency, and privacy, pattern. There's also, you know, from the kind of the center of the architecture here, a real story about openness, innovation, and integration.

I I don't think, you know, we don't want we don't want anyone, to feel like they're locked in, to our AI features and that they can't tell the story they want to tell to their teachers and to their students. And so, in the same way that we deploy everything in region from the very beginning, we also build on top of this open platform that we expose to you and to all of our other partners, so that you can choose what tools work best for you. You can build your own tools, or you can bring in other partners, if there are things that you prefer. And that that foundation has things like those listed here, over five hundred open Canvas APIs and a strong LTI one point three foundation. So I I'd like to talk now about some of the features we do have available inside of Canvas and how the principles that undergird the EU AI Act show up in them.

So the first I wanna talk about here is our Ignite AI generator for rubrics. So this is a feature, that allows instructors to quickly generate rubrics for existing assignments. You'll see here what I had spoken about earlier, clear branding and a feature that's available right inside of the existing UI. So when the instructor asks the system to generate a rubric, that shows up now in the criteria builder that they would have used manually. So if there's something that the instructor wants to change about that generated rubric, they would do it the exact same way they always have.

It's just one button click away, which means that, first, the results are very auditable, which is a key component of the act. But they also, the UI is designed to provide human oversight and to encourage humans to use that oversight to get the exact artifact that they want. But in this feature, right, it doesn't replace human decision making, but it it does speed up a workflow, that otherwise can take quite a long time. The next one I wanna talk about is our summaries for discussions feature. This one is a little different.

So the the idea here is that if I'm an instructor that wants to use discussions but is having a hard time doing that in my classroom, maybe because I have a lot of students or I'm in a large online class, where I can't read every student post, I can use something like discussion summaries, to get a sense of where students are struggling, what questions they have, what concepts they're consistently discussing or or talking about in a way that allows me to then take that information and and adjust my instruction or clarify things that may be confusing to students or or so forth. So here it's less about efficiency, and it's more about, augmenting my ability as an instructor, to understand my students and to meet their needs. The next feature is a a little bit different, but one that I'm personally very excited about. So the Ignite AI agent doesn't just overlay one workflow inside of Canvas. It overlays all of Canvas.

So there's a new button in the top right of the UI that opens up the agent. And the agent is a chatbot available for educators and for administrators. This is not student facing. But it connects directly to that open platform I spoke about earlier. So, the agent understands all of Canvas's APIs, which means that I can ask it to do things on my behalf inside of Canvas, and it can plan and execute those flows for me.

So, you know, here for instance, the user has asked for ten assignment ideas for the content of the page that they're looking at. The agent can put those ideas together, and then as a human, if there's one that I like, I can ask the agent to go ahead and create that assignment for me in a module that exists in my course or create a new module. And and then the agent will take its plan, what it what it wants to do inside of Canvas, present that back to me for approval and authorization. So, you know, here, again, we're seeing AI used not to replace human judgment or a decision, or, not in a way where it runs on its own. Right? It's deeply tied to what I'm asking it to do in real time, and I'm being asked to give approvals in real time as well, to make sure that as the human, I remain in control, and that the the system's not making decisions for me.

But it's able to extend what I want to do in Canvas, and make it simpler for me, as an educator to, to complete complex workflows or common tasks. Now I I think a key component, of our AI strategy, is also about equity, and access. So if we go back one, we've got a few features that I I think make it easier for students to to view the content, that we may already have inside of Camp. So the first of those, which is on the previous slide there, there we go, is our translations feature. So this exists inside of discussions, announcements, and the Canvas inbox.

But it's a great tool, for students who may be learning the language of instruction, and may not speak it natively. So as you'll see in this video, when a student asks for a translation here, they get the original content and the translated content side by side. Because, again, the goal here is not to, short circuit, human thinking or learning, but to provide a way for students to reinforce their language skills at the same time that they're learning the content. Now this one also is a great example of a feature that may make a lot of sense in some contexts, but not in others. This is probably not the right feature in a language class, for instance, where I I do want students to be, productively struggling with the language.

And so this is a a place where using those very fine grained access controls, may be really important. Now, we also provide tools to make it easier for educators to do the right thing and to make their content more accessible. We recently released our Ignite AI content accessibility checker on the next slide. And so this is a course level report that identifies common accessibility issues, things like missing alternative text, headers that are in the wrong order, tables that may be missing captions, and then allows educators the ability to fix them from one central place. And those fixes can be powered by artificial intelligence.

So using a large language model, for instance, to generate alternative text for images very quickly, or to suggest table captions. So again, the the goal here is is not to replace human oversight, not to replace human thinking, but to find a way to expand what those people are able to do by transparently layering AI into the existing flows and using it to suggest both where there may be issues in the existing content and what the remediation may look. And the last feature I wanna mention is a a student facing feature because, you know, increasingly, we're seeing educators build really great patterns for how AI can be brought safely and effectively into the classroom and put in front of students. And so we're beginning to to to follow some of that work that's being done in classrooms by providing those features natively in Canvas. A great example of that is our new Ignite AI study tools.

So, you know, this one is designed to help students adopt practices that have been shown to drive better outcomes and improve learning for them. Ignite AI study tools does that, by allowing students to quickly summarize content that they may be learning inside of Canvas. It allows them to generate personalized quizzes, so that they can have more opportunities for formative assessment and for testing their own knowledge. And it also helps students to quickly develop flashcards so that they can, use techniques like spaced repetition to improve their learning and their mastery of the concepts. And with that, I I will hand it back to is it to you, Fran? It is me.

Thank you so much, Zach. Before we open the floor to our q and a and the rest of the session, I wanted to remind everyone that we do have a purposeful AI in learning blog series that we have starting this week, which will also highlight a free AI literacy course that is coming to the on demand portal on July the fourteenth. So do check that out. Also, using the same link that you use to register for today's webinar do make sure that you also register for tomorrow's webinar the second part of this kind of EMEA AI series where the team will show you a little bit more the practical use of our ignite AI suite. So I just wanted to plug that in really quickly before I hand it over to Melissa who will be handling our q and a and discussion ahead.

Yes. Thank you so much, Fran, and thank you all. We've got quite a few questions that popped into the q and a, so I'm excited to dig in. For this lovely group of folks. I'll guess at who the best person is to answer, but don't hesitate to jump in if you wanna add or anything to whoever answers the question.

So and I'm gonna go ahead and start then with a question that came up at the earlier on in the webinar. I believe this was when you were presenting, Jody. So I'm gonna I'm gonna, ask you this question. And Allison asked, can you share some more examples of high risk and unacceptable risks in the context of this conversation? Sure. Absolutely.

And Fran, feel free to jump in here as well because I know you spoke to this a bit as well. This is a great question, and it's really worth separating the two categories because they're quite different in practice. So unacceptable risk means the banned outright use, so you cannot use it as of February of this year. So the two most relevant that I would say education are emotional reg recognition and social scoring. Emotional recognition is any AI that watches a student's face or behavior to infer how they're feeling, which is prohibited in educational settings.

And then social scoring is AI that builds a behavioral profile essentially for a student over time and uses it for more broad judgments about them as a person. This is also one that is banned. There are products that have existed for some time in this space, and so the act is now drawing a line on not using them. High risk then is different. It doesn't mean don't use, but rather to use it carefully with oversight and transparency.

So those human oversight transparency are the the keys here. And with transparency, I would say that would bring in the bias and all of those things that we discussed as well. So ensuring human oversight will be the most critical here. The examples and and you heard Zach just speak to maybe a couple that you'll be questioning. Does this go into the high risk, space? But the examples that tend to surprise people when I talk to them in education here are automated proctoring that flags or penalizes a student based on detected behavior.

Also, adaptive learning systems that decide whether a student is ready to progress or not or steers them toward a particular path. Admission tools that score or rank applicants using AI, and AI generated feedback that directly influences a grade without allowing for that educator oversight, before actually influencing that grade. So, again, here now, high risk doesn't mean that you need to stop or not use it. It means be very intentional, transparent, ensure that there's always that human oversight. And, honestly, like I mentioned before, I think this is the standard that educators have set always.

So it's just really holding us to holding the AI, excuse me, to that same bar that we expect of ourselves as educators. Also worth noting, we always recommend, of course, that you work with your local team to ensure appropriate governance and approval of any tools that you want to bring into your educational settings. Wonderful. Thank you, Jody. And, Fred, I've got two questions for you at a high level around Ignite AI in general that I think will get these out of the way so that we can kinda dig in a little more deeply.

We've got a couple of product questions related to Ignite AI. The first one was from Ellen who asked, this is the first time she heard about free access to Ignite and how does she get access? And then a second question came in, and that was, from, Anya. And that was if we use it if we do use it during the free trial period, so we sign up during the free trial period and we decide not to subscribe, will any of the course updates that were made made by Ignite AI stick, or, or do they have to worry about those going away? So I'm gonna pass that to you, Fred. Yeah. Thank you, Melissa.

So the first question around how to, access the Ignite AI suite is to go into onto your instance as an admin, click on settings, then feature options, and then you will see what you have there available to you and that is either been enabled or disabled for you. The best thing to do is to put ignite in the search bar and all of our ignite AI suite will appear underneath and then you can see and you can just simply toggle on and off and enable the ones that you would like to trial. If you don't want to do this in in kind of the the the production the prod environment straight away, do test it in beta first. Absolutely do so, but that is how you get that. And the second question around what happens if I start using it and then I don't you know, I I I decide not to opt in once the the the free trial ends.

Ignite AI is really a layer on on Canva. So you can't really do you can do so much with it, but you can't really create something that then you lose off the back of that. So it's it's really kind of an a tool that will allow you to save some time in your everyday task and and just overall better the experience of your students and instructors. So nothing that will get created off the back of the use of the AI function will be lost because it's essentially it's not a a separate thing. It's a it's a kind of a a layer on top of Canvas, so to speak.

Wonderful. Thank you so much. Okay. Jody, question for you because I know and and you highlighted this a little bit in the conversation today, but they are very passionate about accessibility. So Suzette asks, how is the AI course accessibility checker different from the accessibility checker checker available as a feature option already without Ignite? Yeah.

So I think the one that is being referenced is the, RCE or rich content editor, accessibility checker that is available just by nature in core Canvas. And that is actually within that rich content editor that it is looking for your accessibility challenges that you might have that you need to address so that you are fully accessible when you're building out content. The accessibility checker that Zach just shared is actually at the course level. So you are looking at it scans the entire course, and then it provides you with a report that shows where you might have any accessibility concerns that you need to address, which then you can fix from there. As Zach mentioned, it is looking for a a specific set of things, and we are hoping to, extend that over time as well.

But just wanna call that out as well that it it does have specific things that it is looking for such as the tables, the links, the descriptive links, I should say, the color contrast, alt text, that sort of thing. So looking for similar items, but at a higher, level, not just within that rich content editor. Fantastic. Thank you, Jody. Zach, I have one for you, because I think Christophe's question, and I'll read it in just a minute, is getting at some of our philosophy around what we're building and why we're building it.

And I know that was another question that we've heard quite a bit. So I think this will maybe give some context. So here you go, Zach. I think Ignite has a problem on both sides of assessment. Students generate the work, teachers generate the feedback.

That starts to look like an educational orbors to me. The part I worry about more is the language itself. AI writing style becomes the norm students write toward, and that becomes the next training signal. Over time, the range of academic invoice narrows. The human reviewer doesn't fix this because the reviewer gets pulled the same way.

And we're already seeing students push back, so it doesn't only, you know, hit this person's worry, Christophe's worry. You've described this as a template or wireframe for teacher content, but a template is also a default. If everyone starts from the same one, not literally, but in the sense of the seed, the work converges. How are you thinking about that on the product side? And I think this like I said, that's a really good starting point. Think how are we thinking also about our product develop more more generally? Yeah.

That I mean, that's a great question. Certainly, you know, what you've described is, I don't think it's something that any of us want. I is the the classroom version of, of the dead Internet theory where machines are are generating content for other machines to read, and humans are are, you know, not not in the loop at all. I think that, you know, as we approach AI in education, I think it's less for us about replacing human work, inside of a flow, but asking instead, right, what would a human be doing, in this in this use case if they had more time? If I had more TAs, if I had a forty eight hour day, are there places that I would invest as an instructor that I can't invest in today? And I I think you see that in things like our, you know, discussion summaries feature, for instance, where it's able to expand places for an where an educator could can do the right thing or can do the thing that they wanted to do. You know, on on the student side, I think you'll you'll see we've we've been really conscientious about this and trying not to create tools that encourage students to use AI, to build their submissions, or to kind of replace their their thinking patterns.

Alright? I think that the the value of of struggle inside of learning is is real, and we wanna preserve that. I I think, again, what you see there in things like Ignite AI study tools is us, using AI to push students towards better behaviors that they they then do manually. Now, you know, to your to your question about defaults, I I think, look, it it is a hard balance. Okay? I I think we certainly, you know, we know that there are people that could take something generated by NAI and just accept it without reviewing it, without editing it, without making sure it's their own. In the product, I I think you you'll see us, without getting into specifics feature by feature, trying really hard to build user interfaces that don't just have a human in the loop.

Right? I I think that's a very easy thing to say. Alright. We've got a human who pushed a button. But that that the interface encourages the human, to exert their own, authority, autonomy, and influence on that process and to not just choose the easy thing, right, or or the default. Thank you, Zach.

And just to add, some of the work Jody and I do is also about in a in an AI enabled, teaching and learning environment, how do we think more fully about the skills that are being developed in the kind of work that students are doing? So rather than very discipline specific skills, how are we thinking particularly about student, just as Zach mentioned, how how they have opportunities to build judgment, to build adaptability, to build agency as as Zach was mentioning. So we we partner with Zach and team to make sure that we're thinking about just what he described, and we're also thinking about it from a broader AI enablement perspective. Not even just how the features are being developed, but how are we thinking about encouraging our, you know, faculty, teachers, instructors, and students to be using these tools as well. And so I know as part of the follow-up, by the way, for a few of you asked, you will get the presentation and the recording. So don't don't worry.

You'll have copies of this. But also as part of that, we'll make sure to include that link that was shared earlier to community and some of the other resources because we are trying to continue this this kind of deep conversation there as well. Okay. So, Fran, I've got a product question, a a a high level product question that relates to that risk conversation you and Jody were having at the beginning of the webinar. So I believe it's Francesca.

I'm gonna that's probably a a poor translation of how to pronounce your name, but I'm trying my best. How would you rate Ignite AI's grading assistant for SpeedGrader? Does it pose a high risk because it helps respectively influence the grades of student submissions? And I lost the unmute button. Thank you, Franziska. Yeah. Franziska is actually one of my one of my clients.

So hi. And yeah. So it is, I would say, in the higher risk. Definitely not in the kind of unacceptable, but automated grading and evaluation does fall under the the high risk. So it's permitted but heavily regulated.

So AI systems that are used to evaluate learning outcomes, assess student submissions or grade exams definitely come under fall under that category. So if the AI's output directly impacts a student's final grade or academic standing, then it is it is high risk, and that is why we kind of you know, we're very, very intentional in keeping the educator in the loop for the final grading decisions, which is obviously crucial for the compliance advantage. Wonderful. Thank you, Fran. I have two more questions.

So the next question is from Ricard, And this is I think, Zach, we'll we'll have you at least start to answer this one. Could you explain a bit deeper about what open platform means on AI, like MCP, etcetera? Yeah. Absolutely. So it it starts with machine access. So, right, having, all of those existing Canvas APIs.

But then I we layer on top of that, I I think, two important things, when we talk about AI. So the first is, providing a consistent way for, large language model systems to be able to access all of those APIs and understand them. Right? So in in the past, I would have had a human developer who can read documentation and and build custom code. We wanna make sure that large language models can do that same thing. And so we do that with our model context protocol server, which you you mentioned in your question.

This is a standard that essentially just wraps, APIs and makes them discoverable and callable, to large language models in a standard way. And so, you know, the the power there, and I I think when we talk about openness, is that, you know, we can use something like that to build a feature like Ignite Agent, but any anybody else can use the same thing to connect to the large language model of their choice, and to use that to control Canvas. So, you know, that's available to partners. That's available to customers as well. The other thing that I think matters a lot, when we look at bringing large language models into education in an open way is this idea of context.

Right? Does the LMS provide, safe and secure access to what is being taught, to what the instructor's goals are, to what the student may be struggling with, so that, the systems that are plugging in, have the information they need to improve the likelihood of a correct outcome. And so, you know, as as we look at what the platform is moving forward, I I think a lot of it is making sure that that context exists. And that's a combination of of things like model context protocol, but also making sure that we have the right APIs available. So things like Ignite AI search, which can work with, you know, not just exact, kind of keyword matches inside of a course, but it can map the relationship between concepts. So if a student has a question, about a concept but can't quite remember a term, a large language model can take that student question, fetch relevant course content, and then use that to align its response to make sure that that response is correct, to make sure that it extends and amplifies what the instructor is trying to teach, and and doesn't just kind of give a a generic response using prior model training.

So, again, I I think, yeah, that context and then having that that really kind of extensible machine layer are really important pieces of the platform moving forward. Great. Thank you so much, Zach. And I have a follow-up question. I'm gonna keep with you, Zach, that was asked about the accessibility checker, but I'm gonna make it a broader question because I think it'll be good for everyone to to understand our approach from a product development perspective in your slides.

And I think also Fran talked about how we we note when we're using Ignite AI features. It was those nutrition facts. How do you find them? And the follow-up question was from Suzanne around the RC, you know, which accessibility checker is she enabling and how will she know if it's using AI or not? And, you know, maybe just talk just a little bit, Zach, about how you know when there are features, the the, you know, the the the symbol of Ignite AI. How will you know if things in the product are things using AI versus not? Yeah. So we do use consistent branding for Ignite AI.

So it's going to be a a, like, a a gradient that goes from purple to blue, and you'll see a a little sparkle icon typically there as well. As regards that accessibility checker in particular, I I think the confusion comes from the fact that the checking is not done by AI, but some of the fixes are. So you can turn that feature on, use those reports, and not ever really use AI. But there are some options to use AI to remediate some of the found issues. So those would be branded using that same branding.

Fantastic. Thank you, Zach. And I have the last question that came into q and a, but then I'm gonna ask the the whole group one more question to to round this out. But, Jody, maybe, I'll have you help answer this one. So Christophe really appreciated the context of how we're thinking about keeping a human in the loop, where and how we're thinking of not, trying to create a biased system.

So he had a follow-up question, which is great. Thank you for the previous answer. Very thoughtful. Have you considered making generative generated feedback not usable as is as in provide forms of bullet points? And I think that a broader question here from a product perspective is, are we are you know, when we do use AI, are we are we enabling that automation that creates high risk, or are we always pulling someone back in? So, Jody, I'll pass that back over to you. Yeah.

Really great question. So none of our especially this one specifically about the the grading. Right? None of that is surfaced to a student automatically. The educator actually must push the buttons to say, yes. I want to release this to a student or to change it.

We have talked a lot. And, Zach, feel free to jump in here too because I know you've been part of these conversations as well. We've talked a lot about what does it actually mean for them to be in the loop, and are there some things that we can force to ensure? We did talk actually about bullet points and thinking about if we give bullet points for feedback, does that actually force an educator to change the language before they actually hit the submit button? And what we found is no. A lot of educators actually do use bullet points in their feedback. And so, the way that we have been designing and and thinking about this, though, is what are the actions that an educator would need to take? So they need to be reviewing.

They need to actually push a button saying, yes. I agree with this or to to make adjustments. Zach, anything that you would add there? No. I I think that that's great. You know, I I think a lot of this, you're going to see show up as feature by feature.

You know, if and I that's one thing I really love about the EU AI act is this idea of risk classification and additional requirements layered, on top of those risks because it acknowledges that not every use of AI is the same. And and so not every user experience or confirmation or flow needs to have the same level of rigor. Right? So we we've mentioned earlier grading assistance. A grading assistance workflow, I I think you'll see us introduce a lot more friction, into the UI, to encourage human, oversight participation and review compared to something like a discussion summary, where, you know, oh, man. Well, could I have a human go read every post and then confirm that matches the summary? It's like, I could.

That would defeat the purpose of that feature, and I don't think it would provide a lot of additional value. Right? The risk there is very minimal. So Great. Thank you so much, Zach. And I'm gonna ask one last question, and I'm gonna ask all three of you, Fran, Jody, and Zach.

And if any other questions came in, we'll make sure to follow-up after. I know there was a question that came in around signing up for other webinars, and there was a deeper question around custom custom LLMs. So we'll make sure to follow-up with you individually with those questions. Here's my last question. So I'll start with you, Jody.

How will how will we think about the act as we move forward from a thought leadership perspective? And then I'm gonna ask you the same question, Zach, from a product perspective, and I'm gonna ask you the same question, Fran, from a customer, you know, experience perspective. But, Jody, how do we think about it from a thought leadership perspective? Where's the app fit in our world? Yeah. I think as we are thinking about our thought leadership, we are absolutely thinking about every region across the globe that we work with. And how do we consider what you are thinking about as you think about the ways that you want to, consider AI in your educational settings in responsible, scalable, supportive ways for students and educators, particularly? And so, we will continue as the academic strategy or learning strategy team to review and and watch, you know, how are how is this shaping and how are things changing as we consider not only the EUAIA Act, but other acts that we see that are coming about. We do have a team member that is following all educational policy who does help us to to stay on top of all this as well.

And so we are thinking about and considering how can we help to highlight the ways that you can follow best practice, first and foremost, from a pedagogical and and educational perspective, and how do these things then enhance the experience that you are trying to create as an educator. So long answer short, we will continue to watch this act as it continues to, come into full force as of August of this year, as Fran mentioned, and also to see how are you as institutions talking about how this is shaping your educational experiences and how can we help to further that even more. Great. Thank you, Jody. Zach, how are we thinking about it from a product perspective? I I love that answer, Jody.

Well done. I I think, you know, what I like about the the EU AI Act, I I hope this came across the webinar, is that, for me, it it's less about, another checkbox or another regulation that we just have to meet. And I think it's a real invitation, to, you know, us as a product provider, to participate actively in using AI to build the type of future that I I think we all want. Right? One that is human forward, human centric, and really is more about students and teachers than it is about the the tools that we may use, to serve them. And so I I think, you know, we'll continue, to push forward with that type of vision, and I think the EU AI Act is a really great, check, on on making sure that we're doing the right things, in in the right ways.

Thanks, Zach. Fran, finish us up. I mean, it's really hard to follow that, Rick. Both Jody and Zach, I feel like from a from a customer experience perspective, you know, I can reference the the the whole idea that originated this webinar in the first place is really to make sure that you you know that we know. Right? You're not you're not on your own having to navigate this very, very important regulations, and we are not just simply a provider of some really cool features.

We, you know, we set the pace and we set our, you know, our product road map around this. And we want you to know that, and we want you to, you know, make sure that you can have these conversations with us as well and and that you don't feel on your own because, you know, you can be quite isolating as well, you know, trying to figure things out. And, you know, the community piece of, you know, of being within the instructor, I guess, ecosystem, it also means means that the the lack of kind of loneliness in navigating these these regulations and these important conversations. Thank you so much, Fran. And I think I just wanna underscore what you just shared.

Him and his team are ready to be supporting you. We all are ready to be supporting you as you're learning through your own journey with implementing AI in general and supporting the EU AI act. So we're a community together, and we really appreciate this conversation and all of the great questions. We'll continue to be connecting with you. And, again, you'll get the recording from this session as well as the presentation.

And don't hesitate to send us questions as they come up. We're here to help. And with that, I think we're right at the top of the hour. And so thank you everyone for attending, and we wish you a lovely week. Thank you, everyone.