Inside the Google AI for Learning Forum: How Google Is Building the Future of Learning Artwork

Edtech Insiders

At Edtech Insiders, we interview founders, investors, operators, educators, and thought leaders about the future of education technology. You’ll hear from leaders across the education technology industry and learn about all of the important edtech trends happening globally.

All Episodes

Edtech Insiders

Inside the Google AI for Learning Forum: How Google Is Building the Future of Learning

November 27, 2025 • Alex Sarlin • Season 10

Send us a text

This special on-site episode of Edtech Insiders was recorded live at the Google AI for Learning Forum in London on November 14, 2024, where we sat down with leaders shaping Google’s next generation of learning tools, including Shantanu Sinha, VP of Google for Education, Tal Oppenheimer, Product Management Director, Google Labs & Learning, Julia Wilkowski, Pedagogy & Learning Sciences Team Lead, Google, and Maureen Heymans, VP & GM, Learning, Google. Together, they share how Google is designing AI-powered tools grounded in learning science and built to scale across classrooms worldwide.

💡 5 Things You’ll Learn in This Episode

How Google is embedding AI into Classroom and Gemini.
How Learn LM’s pedagogy powers Google’s core models.
How tools like NotebookLM and Learn Your Way personalize learning.
How Google measures learning effectiveness.
What’s next for AI in education.

✨ Episode Highlights
[00:01:36] Shantanu Sinha on putting educators in control of AI
[00:02:32] Gemini’s role inside Classroom
[00:06:53] The next wave of reasoning and multimodal AI
[00:09:36] Teachers creating simulations and learning artifacts
[00:14:42] Tal Oppenheimer on personalized textbooks in Learn Your Way
[00:17:29] Building learning features across Google surfaces
[00:35:03] Julia Wilkowski on Google’s five learning science principles
[00:41:43] Multimodal learning and educational video generation
[00:52:03] Maureen Heymans on using AI to enhance video-based learning

😎 Stay updated with Edtech Insiders!

🎉 Presenting Sponsor/s:

Every year, K-12 districts and higher ed institutions spend over half a trillion dollars—but most sales teams miss the signals. Starbridge tracks early signs like board minutes, budget drafts, and strategic plans, then helps you turn them into personalized outreach—fast. Win the

This season of Edtech Insiders is brought to you by Cooley LLP. Cooley is the go-to law firm for education and edtech innovators, offering industry-informed counsel across the 'pre-K to gray' spectrum. With a multidisciplinary approach and a powerful edtech ecosystem, Cooley helps shape the future of education.

Innovation in preK to gray learning is powered by exceptional people. For over 15 years, EdTech companies of all sizes and stages have trusted HireEducation to find the talent that drives impact. When specific skills and experiences are mission-critical, HireEducation is a partner that delivers. Offering permanent, fractional, and executive recruitment, HireEducation knows the go-to-market talent you need. Learn more at HireEdu.com.

As a tech-first company, Tuck Advisors has developed a suite of proprietary tools to serve its clients better. Tuck was the first firm in the world to launch a custom GPT around M&A. If you haven’t already, try our proprietary M&A Analyzer, which assesses fit between your company and a specific buyer. To explore this free tool and the rest of our technology, visit tuckadvisors.com.

[00:00:00] Julia Wilkowski: So we've tried to align on five learning science principles, which distill the body of learning science, research, and product teams use these to guide their product development and also evaluate how they work. So these five principles are managing cognitive load, inspiring, active learning, deepening metacognition.

Inspiring and motivating through curiosity and adapting to the learner.

[00:00:24] Alex Sarlin: Welcome to EdTech Insiders, the top podcast covering the education technology industry from funding rounds to impact to ai developments across early childhood K 12 higher ed and work.

You'll find it all here at EdTech Insiders. Remember to subscribe to the pod, check out our newsletter, and also our event calendar and to go.

Deeper check out EdTech Insiders Plus where you can get premium content access to our WhatsApp channel, early access to events and back channel insights from Alex and Ben. Hope you enjoyed today's pod.

We are here at the Google Deep Mind Event in London, England. It is. Absolutely amazing. We're here with Shantanu Sinha. He's the VP of Google for Education. Welcome to the podcast. Yeah, great to be here. Shantanu, you have been doing EdTech for such a long time. You have been so thoughtful about it. And Google Classroom is basically the surface, the number one Surfacer EdTech is happening really throughout.

Many, many, many schools. As you sit in this amazing space and learn from all these different leaders, what is top of mind for you in terms of Google Classroom and how you bring educators along for this AI journey that we're all on together?

[00:01:36] Shantanu Sinha: Yeah, great question. To me, I think the most important thing is how do we.

Put educators in control in this moment. How do we give them the tooling that they need? How do we give them the capabilities that they need? How to make sure that they're really leveraging AI appropriately, both for themselves. Yeah, making their life easier, getting back time, but also enriching the classroom and driving richer.

Student experiences for their class.

[00:02:01] Alex Sarlin: Yeah. Google Classroom has become such a go-to for user generated content. You know, teachers use it, they build their whole classes around it. They upload so much into it. Yeah. And now there's this whole Gemini suite and all of these really fit for purpose tools that Google is building that can be built into classroom.

Tell us about how that strategy is evolving. It's a sort of. You know, building these amazing standalone tools, many of which are being demoed here, but they can work in an institutional context or for individual learners at home or for adult learners. That's a really interesting model. I'd love to hear you talk about it.

[00:02:32] Shantanu Sinha: Yeah, I mean, I think one of the things that's so amazing and exciting to me and somebody working in this space is that. AI can do so much. Yes. It has such broad capabilities and it always like blows my mind when, you know, we kind of come in and we think about the, you know, these four or five use cases and you talk to your educators and they're like, well, I'm doing these like 50 use cases.

Right. And there's just so much power behind it and I think that that makes it really exciting because. You know, we have a number of different products and they can all support users in different ways, whether it's the Gemini app directly, which is, you know, remarkable kind of engine that can connect to products, you know, with extensions and other things, and provide really rich kind of AI first experiences or within your existing products, classroom docs, vids, Gmail, all these other.

Surfaces that you've always been using. Yep. But how do we really make sure AI can enhance that and empower it in a broader way? So when we think about the space, and it really is changing how you think about product development to be honest, because you're going to a world where Gemini's kind of inside of almost everything to a certain extent.

And you think about, you know, one, what are the capabilities we want the model to be good at? How do we think from an educational context? You know, we have a Learn LM White paper, and how do we make sure that. We're infusing pedagogy into that core model. And it can be good at things like driving active learning and these like pedagogical principles.

Yep. And that's really important to get into that core model, but then when we do that, all of our products benefit from that. Yes. Now, if you're taking a. Notebook, lm, you're trying to generate a quiz on it. It has all the benefit of that. Or if you're doing it in GEM or if you're doing it in classroom, or if you're doing it in forms or wherever you may be, may be thinking about this.

We brought that in through all of those products. So it is a different way of thinking about the world. And you think about capabilities first in certain ways, and you think about how do we. Leverage our different surfaces to really make those come alive for our end users.

[00:04:27] Alex Sarlin: Yeah. You, you mentioned putting educators in the driver's seat, and I feel like that's such an important stance right now.

You know, people are trying to figure out how AI and education fit together. Some are embracing it. They have those 50 use cases. Yeah. Others are feeling a little nervous about it, feeling like it's coming from different directions. How do you build that sort of feedback loop where you're working with educators, you have so many educators on classroom at any given time.

Yeah, yeah, yeah. You know, how do you build that feedback loop so you're learning from your. Users and that they are sort of learning from you about how Google is thinking about the next generation of AI tools?

[00:04:58] Shantanu Sinha: Yeah, I mean I think this is the core of, you know, we had Google for Education done for a really long time and I think like our, you know, I think the ethos from day one was to really work with educators.

We recognize we're not the experts on it. We have a lot of educators on the team. Teachers we're passionate about the space, but we also understand technology's changed. They were in the classroom. Yeah. And we've, from day one, really focused a lot on how do we pilot with. Teachers, how do we actually get feedback from teachers?

How do we make sure that we're really empowering them? And I think we have a bunch of different tools to do that. So we have a really strong UX R user experience research arm that really goes in and, and talks to educators and brings that feedback in. We have a strong pilot program where we're building new features.

We'll run pilots with classrooms, see 'em in those settings. We have customer advisory boards where we're talking to educators. All of that feeds in. And then we have data, right? We will, we'll look at like how. Features are used and like what happens in these different contexts to see like, okay, what's, you know, we thought this was gonna work one way, but what's actually happening and where are people actually, you know, where does it actually have any impact that we thought it would?

So I think, you know, leveraging all of those different types of signals is really what kind of drives our process. And having a mechanism where. We really are listening and understanding what's truly happening in classrooms.

[00:06:14] Alex Sarlin: Yeah. I know you have a background with Khan Academy. Yeah, and Khan is sort of relaunching their core product in some really interesting ways.

Right now it feels like we're sort of entering almost like the third inning of the AI chapter in education as we sort of define it. We're three years in and feel like Google has really been quite out front in some really interesting ways in actually shipping products, testing them, piloting them, getting them into.

Many surfaces. Gemini is everywhere. Learn LM is under the hood. Yeah. As you think about, you know, the next year, what do you feel like is sort of the next around the curve about where AI is going to go next in terms of you, what's the big trend that is about to happen that people may not know about yet?

Do you have an idea for that?

[00:06:53] Shantanu Sinha: Yeah. I don't know if I have a crystal ball on any, nobody does any of this stuff, but I would say like AI continues to impress in pretty remarkable ways, right? Yeah. I think I, you know, as somebody who's thought about the space and you know, I actually studied artificial intelligence when I was in college and like to kind of see like the rate of progress, like it's not slowing down.

If anything, it's speeding up. That's true. And I think. It's starting to unlock different. So I think a few things that you've already seen is with reasoning models, hallucinations is coming way down. Yeah. Accuracy is going way up. Yep. So the quality, and that changes a lot when you pick an educational context.

If something's right 90% of the time, you might as well be right, like Right. That's not useful for me. Right, right. Like I need you to be right all the time. And I think like we're really getting to a point where, particularly if you ground it on concept like notebook lamb, where it's like high, high quality experiences that you can really make in these educational contexts.

So I think. That's changing. So the accuracy of these models, and you could see it in the benchmarks, right? Yeah. Like you see like what Gemini Deep Think could do in terms of like the top math exams, and it's remarkable. It's really amazing. I think you're also seeing amazing stuff on the multimodal Yeah.

Side of it, the video, both on the understanding and the generation. Yep. Of it and it's getting faster and cheaper and to do, and I think that's really, really powerful. Yep. Because that starts to open up a lot more possibilities, particularly when it gets faster. Right. And you know, already we have things like live translate on meat where you can take a audio of somebody speaking and within just a few second delay, it's translating it in your voice.

Yes. Right. Back to it's remarkable. Right. Amazing. And that's really just. Like, and it speaks to that multimodal understanding, voice, understanding these models have with the capability that it's had around translation and then the generation and like it could do all that in a really short period of time.

So I think that's an amazing trend and I think that starts to, you know, I think translate's a great example of what can happen with that. But if you think about it in educational context, just so many really. Amazing experiences that can happen if you can use video voice, all of that together. I think another thing that a lot of people are talking about is the gentech stuff.

Yeah. And like how can you really take a goal and like put AI against that? You know, one of the most obvious ways that people might first see this on things like vibe coding, where it's like, okay, I wanna build something. I think that already has amazing applications. Like I was just thinking, you know, if you're a teacher today.

And you wanted to have a simulation or you wanted to create this amazing exercise for your class, you know, back in the day you would have to like find the web, hope somebody built it. Oh yeah. Now you're just like, I can imagine it, like I wanna see pendulums and do this physics lab. Totally. And just tell it to Gemini and it's gonna make that for you.

Yes. And you can use that in your class. So I think like. That generation and kind of those type of agentic experiences is remarkable. When you can put that in a really strong educator's hands, they can do amazing things with that. It's true,

[00:09:36] Alex Sarlin: and those artifacts are shareable. They can be shareable across classrooms.

It could be potentially shareable among schools. It's a really exciting potential when any individual educator can create. Simulations can create videos, can, you know, become content creators. Yeah. Legitimately. Yep. Yep. It's an amazing world. I wanted to ask you about social and collaborative ai. I feel like it's a theme that's, we've talked about a lot on the podcast and I think it feels like something that may be just over the next curve.

AI as a facilitator between teachers and students, you already do some of that in classroom AI as a facilitator between students working in groups. Is that something that's on your agenda, sort of as you're thinking about what's next for ai?

[00:10:12] Shantanu Sinha: Yeah, I mean, I think we're thinking about all the different.

Ways that educators are trying to use this. Yeah, and I do think, like as you're saying it, I've been kind of amazed at how broad, you know, people have tried to leverage this. Yeah. And you know, in many ways as a brainstormer, as a discussion coach, like there's really interesting things there. What's important to us is that we're empowering educators to, that we're trying to build general purpose tools.

That are customizable. Yeah. So like already within classroom you can go in and make a custom gem and you can give it very specific instructions on exactly how you want it to operate and assign that to your class. And again, you might think, okay, the first thing people would do is try to make an AI tutor or this or that.

But educators come up with all kinds of different things. Yeah. And all kinds of richer, yeah. Different types of experiences in their classroom. And I think that is really, yeah. Powerful. And I guess when we think about it, we think about, okay, how do we put that control into the educators? How do we we'll ensure that we can unleash their creativity?

Yes. How do we give them the controls that they need in that context? And then how do we make sure our models are good for, right. Those experiences. If it is like, where does it have to push on? The learning science or the pedagogy and how do we make sure, you know, we're not just hoping that these models are good, but we're actually ho climbing and evaluating the models to ensure they're actually meeting the needs of those educators.

Yeah.

[00:11:31] Alex Sarlin: And then you can learn from all of those use cases that educators are doing really interesting things in the classroom that you become ideas that are part of the mix. I mean, you mentioned, you know, just AI is getting so impressive and I feel like, you know, as a whole society right now, uh, technologists, educators, we're all sort of getting our head around how fast this thing is moving.

Yeah. You know, it's just moving so quickly. Things that a year ago seemed. Really far out are here. Yeah. And I think a year from now there'll be even more things here we don't expect. I'm curious, you know, as the entire field, obviously Google is at the forefront of ai outside of education as well. Mm-hmm.

Doing all sorts of things, uh, as the entire field evolves and as you know, new capabilities come in, how do you think about sort of bridging the gap between like the cutting edge AI things and then the classroom use? I feel like you're sort of play this really amazing role sort of helping. Bridge, you know what is absolutely brand new in the space.

Maybe agentic, maybe you know, robotics. I have no idea. Yeah. With what's actually happening in the classroom, how do you see your role in sort of bridging that interesting gap?

[00:12:26] Shantanu Sinha: Yeah. I go back to like, I see our role as putting educators in control of that experience, right? 'cause this stuff's happening.

Yeah. And students sort of running off and using this, no question in technology on their own. And they may be using it appropriately, or they may be using it in ways that the educator be like, Hey, I don't know if you really did the right thing or you learned the right lesson from that. There is. And I think like for us, we wanna put the educators back in control of that.

And we wanna make it easy for them to use these tools, but use them in a way where they can guide students so it has the best outcomes. And I think, you know, when you think about that spectrum, you know, there's core parts of what education does that isn't changing, right? Like I think in some ways a lot is changing, but there's certain things that aren't changing.

That's true. At the end of the day, we're really trying to make sure that we're. Having an education system that's creating well-rounded thinkers who can have understanding critical thinking capabilities or good collaborators, good communicators who will socialize well with their class. Like these are the things that are core to education.

And I think these are things that are, aren't changing significantly. And sometimes technology can help and sometimes technology gets in the way of that. But I think like what we are most interested in is how do we make sure that we're. Empowering educators so they can leverage what's great about that, to build those richer experiences for what they're trying to do in their classroom.

And I do think it does open up whether it is what you could do with video creation now. Yes. You could do with vibe coding, with like all of a sudden your project based learning can be so different than what it was before. And like everybody with an idea could probably make something really remarkable in different ways.

And that's an amazing world that we're, it really has that we're moving in and we want to encourage more educators to be able. To move there as opposed to, you know, trying to catch up to it. We want to get educated back in the driver's seat Yes. Of AI and ensure we're building tools that really like empower them that way and

[00:14:10] Alex Sarlin: driving sort of at the front edge of what AI can do, which is I think, really exciting part of how this works.

Final question for you, one of the. Products here. That caught my eye, because I think it goes right to the core of this concept of personalized learning is the Learn Your Way project. Yeah. Which is coming outta the Google research team, but that's basically sort of personalized textbooks based on student level, student interest.

It can generate educational illustrations. I'm curious, that seems like a natural, potentially a natural fit for classroom, but given Google's, you know, structure, I'm curious if that's something you foresee going. Being available for students in Google Classroom anytime soon.

[00:14:42] Shantanu Sinha: Yeah, so I think before, one of the things that makes it really exciting and interesting now is how it's changed our product development process.

And, you know, the Google research team can really push the frontier of these capabilities, right? And ultimately it is about those capabilities at the model. Like is the model good at. Taking your text and translating it into a quiz that's really accurate and really getting to the nuances of what it, or can you take that text and re-level it to this grade level with the right vocabulary?

And you know, these are really core capabilities and I think Learn Your Way is a really great example of us taking arbitrary text and doing these different customization that really makes it yours, right? Whether it's your own interest, your own reading level. You know, different form factors of content.

We think that's a very important capability and I think that's, you know, when we launch something like that in labs, it allows us to experiment to understand it, build those capabilities that ideally we upstream into our models for everybody to brought to benefit from. Then there's different surfaces that have.

Things kind of like that. Yeah. That we wanna push the envelope on. So, like you said, classroom's obvious. One, we have a new Gemini feature in classroom where educators can do things like take texts and re-level it and, and, you know, make sure that we're putting educators, giving them the ability to say like, well, let me take that content that I found on the internet.

But that was like on Google Scholar written for, you know, post grad. And I want that to be relevant for my middle school. Little class. Exactly. Can I actually take that and make a nice lesson with that? So we have some things there, but as that capability's being built, we'll have more and more notebook, EMS and other obvious application of it.

So I think, you know, we try to focus on that capability, get, try to get that capability as strong as possible. Yeah. And then we try to make sure as many products as possible can benefit from it.

[00:16:24] Alex Sarlin: That makes a lot of sense. And those transformations can be used in many different contexts, right? By teachers or students

[00:16:29] Shantanu Sinha: themselves.

Yeah, exactly. It's, and for a lot of this stuff, there's a really strong synergy between what the educator wants versus what the student wants. Because ultimately, you know, I think people will sometimes look at that as being on different views, but like the student wants to learn ultimately what they need to.

Do well in the educator's class and what the educator is gonna be assessing them on. Exactly. And I think like having that educator input becomes so important because it's like, okay, this is what a good quiz is. This is what a good instructional material is, and that benefits students. Even if they're doing it on their own and using Gemini to try to do some of these conversions.

Yeah.

[00:17:04] Alex Sarlin: And educators wanna engage students. They wanna keep their interest and they wanna make it relevant. And all of those things are baked into these tools. It's really amazing. Being here is just spectacular, you know, showcase of all the things that Google is doing in ai. Shantanu Sinha is the VP of Google for Education.

Thank you so much for being here with us on EdTech Insiders.

[00:17:20] Shantanu Sinha: Yeah, it's really great to be here. Thanks for having me.

[00:17:23] Alex Sarlin: We are here with Tal Oppenheimer. She is the Director of Product Management on the Google Learn Team. Welcome to EdTech Insiders.

[00:17:29] Tal Oppenheimer: Thank you. Thank you. Excited to be here. Yeah,

[00:17:31] Alex Sarlin: so first off, we're here at the Google DeepMind office in London.

It's an absolutely incredible event with amazing people and learning all about what Google is doing and what the ecosystem looks like at large. Tell us a little bit what Google has been doing in the learn space. It's been really innovative and very wide region.

[00:17:47] Tal Oppenheimer: Yeah. Well, I, I'm glad to hear it. We've been hard at work across many, many different teams at Google to really focus on how we can help students, but also everyone, yes.

Just, just learn more effectively. And we try to do this based really deeply on kind of learning principles across the board. We have kind of a deep pedagogy team with a lot of expertise, but we also think about for everything we build. How do we not only take those learning principles, but how do we also assess their effectiveness in helping actually kind of drive those learning outcomes?

And we do that across so many of Google's products because we know many different products are used for a lot of different learning journeys across the board.

[00:18:16] Alex Sarlin: Yeah. So we've covered how Learn lm, the Model for Learning was incorporated into Core Gemini. Yep. There are also these learning benchmarks that, that you measure different features across.

Yeah. Can you tell us about that approach? It's a really interesting one.

[00:18:28] Tal Oppenheimer: Yeah. So across the board we have kind of both like foundational benchmarks that we look at from the model level, but we also, for every new feature that we build, we think about, you know, what is this feature specifically trying to help with?

What are the specific journeys that we try to help with a set of queries that we specifically try to support. For example, a homework query that's trying to help with math. It's a very different set of responses and support than a more kind of open-ended maybe history question or, or learning experience.

And we wanna make sure that our products can work across all of those things. And so, you know, the classic, especially in the world of LLMs, you build your eval set and your your query set, but you also wanna make sure that you are assessing that not only on kind of quality of answer in terms of user perception or human perception.

But also effectiveness on actually teaching you the concept that you came to the products for. And so we very much look at at both, which is one of the reasons that I love working and learning, is that it has this kind of overlap of like truly helping students and any really learner around the world, but also making sure that we do so in a way that's helping them grow as a learner and thinker.

[00:19:22] Alex Sarlin: Yeah, and the pedagogical effectiveness sort of comes in five flavors. At Google, you've sort of identified five learning science principles that everything is measured across the curiosity and active learning. Yeah. As you develop different features, are there sort of different mixes of the pedagogical principles you,

[00:19:36] Tal Oppenheimer: yeah.

It's both different features, but also for a single sort of student or learner. Yeah. It also really depends on kind of where in their learning journey are they? Right. If they're very much kind of first being introduced to a subject. You may not wanna jump into like the very deep active learning experience.

That's right. That's answering all the details. You sort of need to have that kind of overview and that introduction.

Yeah.

And then kind of as people progress, really thinking about how do you help them go deeper. And that's where, you know, depending on the subject, active learning, again, teach kind of the math example, needs a very different set of tools and capabilities than what an active learning experience might look like about how you bring actually a novel to life.

Yes. Or some of those areas and make it's, is. So good at so many of these aspects that you really have this opportunity to create a kind of multimodal learning experience that's not just multimodal, but also personalized. Yes. And I know you've gotta maybe see some of the Learn Your Way demos and those things.

Yeah. It's exciting and we look at kind of how we can bring, bring a lot of those pieces. Yeah.

[00:20:28] Alex Sarlin: Learn Your Way is one of the newest Google products and it's basically personalized textbooks coming outta the Google research team where people can decide, you tell the system their interests and the textbook adapts, even the illustrations adapt.

It's a really interesting product. Yeah. And looking at

[00:20:41] Tal Oppenheimer: kind of grade level and otherwise, um, to make sure that the content is. Underst and accessible for what you specifically

[00:20:46] Alex Sarlin: Yeah. It levels it to, there's a learning level and to interest. It's a really interesting model. You know, you mentioned the sort of iterative nature, and this is something I think is really interesting compared to traditional search, right?

Is that the first thing you ask an LLM in a, even in a single learning session, might be very different than the sixth, than seventh than eighth thing. It's a conversation. It's a conversation, yeah. And from UX perspective, that's a different thing to design for. Yes. You're designing for long term, you're designing for consistent experience for people coming in and out.

You run the uxr team for the team. That looks in terms of the interfaces that you build, how you're building interfaces, where it's not, you know, transactional, just give and take, it's settling down for a dialogue and a conversation and really trying to learn together. What does that look like? Yeah.

[00:21:26] Tal Oppenheimer: Um, it's fascinating because we also have very different variations of even that experience across our products, right?

What you might have in, in Notebook LM for how you design such an experience where you're grounded in a certain specific set of content,

right, can look

very different than what you maybe have in, in Gemini. Were. Sometimes it's grounded in content and sometimes it's much more open-ended, and so you need that, that flexibility.

I think in general it's a mix of pieces, right? One key element is as kind of an end learner, a lot of the input tends to be very text-based, which is very different than other products, which, you know, it used to be click-based, whether it's a touch screen or with your mouse, or even kind of image upload based.

Yeah. And right now you have the opportunity for all of those. But we do see that most people are typing or voice dictating or kind of speaking tends to be text heavy. And this means that there is. A of different steps that need to happen, right? There's everything from obviously understanding the user and things that the models have gotten very, very good at.

But also sometimes knowing when to disambiguate and knowing when to suggest that maybe a different form factor, maybe a better fit, maybe a quiz or a flashcard, which is, it's not a text-based response. It's actually a very interactive response. Maybe the better fit. And with all of these, one of the key kind of UX challenges that we have is also discoverability.

Yeah, to some extent. You have products that can do so much, but that also means it can be really hard to learn everything it can do. Very much so To some extent you can never quite learn truly everything it can do because it's also constantly getting better. Yeah. And so how do we help bring those, like bridge that gap and help folks understand what's possible, but still keep kind of any learner in control of the experience that they need and want?

Yeah. Across

the board. And you see that we kind of approach that differently in different products. So for example, notebook, lm. Starts with you first and foremost, upload content that's relevant and we see that resonating really strongly. Yeah. With learners, because usually you have something you're trying to learn.

Yeah.

And kind of starting with that is very helpful. And we also have, you know, integrations with Google Classroom that allows a teacher to put relevant content in so that you know, they can distribute that to their whole class so they can all have the relevant materials. Study, but you also have that, you know, studio panel on the, on the right hand side.

Exactly. That can help you create that, you know, video overview or that audio overview to get that breadth, mind

map, mind maps, flashcard, that you flash, you kind of start, go sooner. Yes.

Flashcards, uh, and quizzes just rolling out on mobile as well. Amazing. That flashcards in your pocket piece. And you see that actually that combination of both a chat interface and some of those kind of very visible building blocks also very helpful.

Yeah. Both for individuals who are learning on their own or in that classroom setting, or even as a student who's, you know, maybe creating that study guide to share with other students in their class.

Yeah. Uh,

so it's been very fun to get to actually work on learning across all of these products and both with our.

UX research team go really deep on what the, you know, foundational user needs are and how we, you know, bring together learning science to make sure our solutions are actually solving those things with how we actually, how that manifests in all these different products across the board. Right. YouTube has a very different set of Yes.

Ways that we can assist with learning on kind of video based experiences as well.

[00:24:02] Alex Sarlin: Yeah. Interesting model here about the YouTube TED-Ed, they're doing an interesting thing with YouTube for music education. You can actually embed interactives inside the video and be practicing.

[00:24:10] Tal Oppenheimer: Yeah. So like how we create active learning experiences on video, which is a very.

Engaging medium. Yes. But it's actually a very lean back medium.

[00:24:16] Alex Sarlin: Yes. I that it can

[00:24:18] Tal Oppenheimer: be very hard to actually fully retain from.

[00:24:19] Alex Sarlin: Yeah. So I wanna double click on two of the aspects of it. One is the multimodal, this idea. Yeah. You mentioned that, you know, text-based input text or voice-based input is the norm right now.

Mm-hmm. We have a sort of theory and EdTech insiders that things are gonna get more and more multimodal. Yes. Just as the internet. Became more and more multimodal and it's not that comfortable for many young people to use extended text. Right. To write a whole paragraph about what you want and give all the details.

It's actually not something a lot of people know how to do. Yeah. I'm curious how you see both the input and the output. We know that video overviews the output can be multimodal. Do you see more multimodal input?

[00:24:51] Tal Oppenheimer: Absolutely. I think we already see some of this, right? Especially for kind of mobile applications.

Yes. Whether it's a Gemini or notebook, lm. In the student or in the classroom. Right. Take a picture. You get assignments. They're still often on physical paper, right? You're still reading physical textbooks and you don't wanna type in that math problem. Right. That's very hard to do. And a photo's very fast.

Yes. And so we already see that in kind of photos. We also start to see that as a lot of classrooms have become more digital, especially after the pandemic and how they actually distribute some of their content. Yeah. We also see video content being kind of classwork and otherwise as well. And so we see kind of all of those start to be inputs, and I think one of the things that we can do to help is.

One, make it a lot easier Yeah. To bring those multimodal inputs in. Yeah. Um, but also realize that it's rarely, you're only using images, you're only using video, or you're only using text or voice. Right. It's actually most often that. Even in a particular journey.

Yep.

Maybe using all the above. And so how do we allow for that flexibility throughout the experience and definitely on the consumption side.

And actually even more so on the consumption side, rarely is one media the optimal for the entire conversation. Yeah. Right. Video overviews can be great for an overview. Yeah. But then you need podcast

[00:25:54] Alex Sarlin: videos and

[00:25:55] Tal Oppenheimer: amazing with generative, not even limited. The main clips we usually think of when we think of media, right?

Yeah. Video, images, text, voice. But you actually can use code generation. Yeah. Right? Like flashcards and quizzes. You have canvas in Gemini. That's true, right? That you can actually kind of have these experiences across the board. Yeah.

[00:26:11] Alex Sarlin: The output could be a quizzing machine or it could be an interactive, you know, interactive code based module where you actually Simulation.

Yeah. Or students.

[00:26:18] Tal Oppenheimer: Exactly. Or students can even. You know, ask to create something within Gemini, kind of through, through Canvas, and simulate that directly.

[00:26:24] Alex Sarlin: Very exciting and powerful. The other thing I wanted to double click on, and something that's come up a lot at this conference is multiplayer mode.

Mm-hmm. Is the idea of, you know, starting to get to more social learning. Yeah. I think we, we've already started to really go down that road of the paradigm of, you know, students go really deep with AI in this sort of back and forth conversation. I'm hearing a lot here about, you know. Working with each other, or teachers and students working together with AI as a facilitator?

I'm curious how Google thinks about it. I haven't seen a lot of products in that sense yet, but there are some.

[00:26:50] Tal Oppenheimer: Yeah. Yeah. It's generally an area. I think we feel very strongly, and I feel very strongly that learning is fundamentally a social experience. And our goal in general, how we approach a lot of these technologies is about how do we help make people be the best version themselves, the best learners, both in the mix, and we very much wanna actually make the human interactions that are part of learning.

Easier and more frequent. Yes. We very much believe that. Like you still need that human connection across the board for many reasons. Motivation, like true learning, like there's also a lot of social skills that develop there. Like many, many reasons.

Yeah.

But I think that there's more we can do in the product to actually enable that.

We start to see it with some of the abilities to create multimodal output, for example. And then you do have the, whether it's a teacher or a student, kind of creating those objects that then become. Learning products themselves for everyone else, the social reason, it can be used, the social piece, but there's a lot more that we can do to kind of make that even easier.

Right. I'd say things like our classroom integrations for some of our products start to allow that more specifically for the teacher student relationship in particular.

Yeah.

Trying to take the burden off of a teacher from having to make like a burden of this for every student. Like make one help distribute it.

They can still get the personalized experience, uh, as a recipient, but I think there's a lot more that we, we can do. It feels like. It's amazing how amazing the technology already is with how new it already is. I totally. And how much more there is to still Yes. Explore and like truly deliver to help our learners.

[00:28:07] Alex Sarlin: Yeah. I think there's a potential for sort of a jigsaw model where, you know, students are working on different parts of a problem, creating artifacts and then sharing with each other, and suddenly you have this and there's, you know, class created artifacts. There's

[00:28:17] Tal Oppenheimer: so many use cases from that actually across like all learning journeys, not just.

Clear cut one where you're trying to deliver something together.

Yeah.

But it's also very true, like studying almost always actually a social experience's. True in many, in many settings. And even homework, especially in higher education. Yeah. Where those problems are hard. Yeah. The study group and the library

[00:28:36] Alex Sarlin: doing the problem set together is like a very normal piece of higher ed and AI could play a really interesting role sort of facilitating Yes.

And supporting that kind of thing.

[00:28:43] Tal Oppenheimer: And we even have some, some earlier explorations and kind of Google research and otherwise around kind of how do we also help people learn to collaborate work better?

[00:28:49] Alex Sarlin: Yeah. Okay. It is incredibly interesting stuff. I mean, when you walk around the floor here, you just see all of the different places that Google is playing in AI and it is just so exciting.

The reading piece. Are you involved in that? Yeah. Tell us about the Google early reading, because that is a really interesting, and it sort of feels like an outlier in some ways 'cause it's so specific to a particular subject.

[00:29:09] Tal Oppenheimer: Yeah, so there's a couple elements that we looked at here. We both look at kind of early reading, so there's products like Read along that have been available and are also now integrated into classroom, which really focus on how we can help.

Early reading acquisitions and to a large extent, right, the ability to read and consume content unlocks so many more. Yes. Learning capabilities that are, are pretty key. And so, like we've really been, Google has actually been focused on that for quite some time on helping do that at scale in a way that's truly accessible, but also hopefully like fun and engaging.

Yeah. Yeah. Um, so some of the areas where generative AI can really help in terms of making sure that, you know. The story that one kid might be very motivated to read might be very different than the story another kid motivated to read. And actually you can still help them learn to read at, you know, their desired level across the board in those areas.

But we've also been starting to explore how we can actually help even kind of more senior levels where readings actually still. And critical thinking about things you read is just as important, if not more so than it used to be. Yeah. And it's still something for people to continue to learn and how can you make that easier and more fun and more accessible across the board.

And so we have kind of learn about, um, also has an experience where you can kind of now upload a PDF and get a bit of an experience on top of that to actually, that's interesting how you can engage with the content and starts to bring in some of those multimodal aspects. But because it's targeting a little bit of an older audience mm-hmm.

Allows you to actually specify, you know. Key areas you're looking to dig into, think like themes and a literature assignment you might get that you wanna kind of explore more deeply in the reading.

[00:30:29] Alex Sarlin: You know, we have a literacy problem in the US that is very deep. It's not just for early learners at all.

You have very, very high levels of functional literacy in the US and around the world. And it's interesting to hear, see that Google's going into it directly just because many of your tools are content agnostic. They're really about students upload their own content. It could be high school biology, it could be college, you know, astronomy.

It could be anything. But the reading space is an interesting one for the space. It's interesting for Google as well.

[00:30:55] Tal Oppenheimer: Yeah. We generally always at Google, try to go both broad to make sure that we can solve big problems, but we also know within that breadth we wanna be able to solve like the real problems that our learners have.

Yes. And so you need to, we often find that we explore 'em at both altitudes.

Yes.

And then think about kind of how do we bring them together so that folks can, don't have to try to. Between all these different corners and edges and you actually get what they need. But how do we make sure we're really solving that specific problem very deeply?

[00:31:17] Alex Sarlin: Especially a problem like reading that unlocks everything. Yes.

[00:31:19] Tal Oppenheimer: It unlocks so many other things. Thank you. Yeah.

[00:31:21] Alex Sarlin: So last question for you is working with schools. Yeah. And working with districts. Mm-hmm. So, you know, Google Classroom is already, and Chromebooks are absolutely ubiquitous in the US education system and around the world.

You're creating all of these sort of really interesting, you know, semi experimental or new features. Mm-hmm. But then it's all on this sort of classroom, you know, infrastructure. Mm-hmm. At least has a potential to, in this classroom infrastructure. I'm curious as a director of product, how you're thinking about how all the pieces come together and sort of, you have this amazing delivery mechanism Yeah.

Which very few people have. Mm-hmm. To actually get things in schools at scale. How do you think about it?

[00:31:51] Tal Oppenheimer: Yeah, so we generally think about both. We try to build all of our products in a way that they can help kind of any learner who's trying to access them directly, because we think it's very important to make sure all of these tools are accessible.

I think in a classroom, there's a couple elements here. First and foremost. Learn of the learning and academic situation, which means not only the students, but also the teachers and certain areas, depending on kind of the age, you may also, you know, wanna get parents' feedback or otherwise Yeah. Or departments of education.

Yes. And things. And I do think with how nascent the technology is and how much it can really make education much more accessible and effective across the world. It's really important to do so with the community that's most impacted by it. And so we do work very closely. In general, I see classroom as one particular way that we can help create experiences that are even more powerful if you have a teacher in the loop.

But we really think about this as just one way to access our products and a way that, you know, if you do have a teacher on classroom. They can set up that notebook from the start. Right? But actually any teacher can access kind of notebook, LM today and create that and share it with their, with their students across the board.

And we really, that's very intentional. Yeah. We wanna make sure that both of those things are accessible. And of course, if you're in the classroom setting, there are other, you know, management benefits that you have that can kind of help in that experience. And so we mostly see it as a yes and right. We wanna help.

We know that in. Institution setting, there's a different set of needs and we wanna make sure we can meet those needs directly, both for teachers, for administrators, and for students across the board. But we also wanna make sure that access doesn't change.

[00:33:14] Alex Sarlin: Right. That even if you're not a Google Classroom school, or if you're a learner working outside the traditional system, you can still access.

Yeah. And

[00:33:20] Tal Oppenheimer: we feel very strongly that like everyone's a learner or everyone can be a learner regardless of their age. Right. Academic learning is one that we're very deeply focused on because of just how critical and fundamental it's always, but especially at this particular. The technology. Yes. But we really try to build it in a way that whether you're learning for work or you're learning for life, or just curiosity, that these products can still, still really help you.

[00:33:39] Alex Sarlin: Yeah. And that individual sort of, I don't wanna call it a consumer use case, but the individualized use case. Yeah. Or it could be a learner of any age trying to come in and learn something new, either for work purposes or curiosity versus the institutional use case. I feel like this is another theme I've seen a lot at the conference here because Yeah.

You know, this sort of learner. There's institutional first products. Mm-hmm. Like classroom and then individual first products. Yeah. But you're obviously deciding to not choose to allow things to work. In both contexts

[00:34:06] Tal Oppenheimer: we think of, yeah, we think about them and how can they scale across both because in, even in an institution setting, you do have a learner in, in, in the experience, and you wanna make sure that it's great for them.

We also wanna make sure that it's great for every other player in that, in that situation, because we really do wanna make it easier for everyone. You need to think about the actual, the teacher and the administrator in those situations across the board. And if it's a, you know, last mile only consideration, you end up with different products that are still, can be a little bit rough on the edges.

And so we do work very kind of closely also with our, our classroom team.

[00:34:35] Alex Sarlin: Yeah. Amazing. Thank you so much. This is really, thank you

[00:34:38] Tal Oppenheimer: for being here.

[00:34:39] Alex Sarlin: Oh no, I'm, I'm honored to be here. This is Tal Oppenheimer, she's the director of product management for the Google Learn team. The LearnX Team, is that right? Yes.

I working

[00:34:46] Tal Oppenheimer: across learning and Google Labs.

[00:34:48] Alex Sarlin: Learning.

We're here at the Google Deep Mind AI and Learning Forum with Julia Wilkowski. She's the learning science team lead. A very important role when you're doing pedagogical product like Google is. Welcome to the podcast.

[00:35:03] Julia Wilkowski: Great. Thanks Alex. It's great to be here.

[00:35:05] Alex Sarlin: It's great to chat with you. I think you have the most amazing job at EdTech.

Literally. Uh, we, I would agree.

[00:35:11] Julia Wilkowski: Not every day is sunshine and roses, but to be at this. Amazing precipice of learning and big technology is a really fascinating place to be. It

[00:35:20] Alex Sarlin: really is. So tell us a little bit about what it means to be a learning science team lead. You're informing the base model, you're informing all of these features, you're informing how features get into Google surfaces, like YouTube.

What does it look like on a daily basis?

[00:35:32] Julia Wilkowski: Yeah, so we try to represent the voice of a teacher. How trying to help build great learning experiences. When someone comes to Google search looking for help with their homework, someone might come to Gemini looking for a deeper conversation. Someone might come to YouTube looking to learn a new skill, and so we wanna try to make sure that we have.

Great learning experiences in all of those different surfaces. So we have people embedded on those product teams who work throughout the product development lifecycle from ideation, brainstorming, early evaluations. I'd say the bulk of the work that we do is really evaluating the quality of those outputs.

Yep. So we've tried to align on five learning science principles, which distill the body of learning science research, and product teams use these to guide their product development and also evaluate how they work. So these five principles are managing cognitive load, inspiring, active learning, deepening, metacognition, inspiring and motivating through curiosity and adapting to the learner.

[00:36:26] Alex Sarlin: Yeah, and you know, learning science is 80 years of all these journals. Sometimes it's combative. It's a very elegant approach to get down to five core principles, and it allows you to work across many teams and they can sort of. Pick and pick. Well, for what we're doing right now, curiosity is gonna be the key or what we're doing here.

Maybe it's cognitive load. Tell us about how that works.

[00:36:45] Julia Wilkowski: Yeah, so again, we work with product teams and the product leads to really prioritize which learning science principles are gonna be most relevant for that product. So for example, if you think about Google search, we really prioritize managing cognitive load and maybe some motivation and curiosity with links to, you know, why this is important in the real world.

Are a little less important in search, but if you come to Gemini and you're actually willing to engage in a longer conversation, we have the opportunity to help you learn, actively interact with that content. Maybe we can even get you to reflect on how that learning experience was and start to build those metacognitive skills.

Yeah,

[00:37:22] Alex Sarlin: it's incredibly exciting to be able to bake these learning principles in across many services, and one of the things that we were really odd and inspired, attic insiders. You know you were doing Learn LM designing a model specifically for the learning use case, and then Learn LM really became baked into the Core Gemini model across all platforms.

Tell us about how that came about, because that wasn't the most obvious thing that was gonna happen with Learn LM from the outside and when it happened, it just felt like a real dedication to learning.

[00:37:48] Julia Wilkowski: Yeah. I think Google's really trying to build a great core model that can work across many different use cases.

And that's where Learn LM fits in. We have contributed training data, we've contributed evaluations. And again, evaluations are really the heart of what our team does, is thinking about how might someone use the product? What is the quality of their response, what are the expectations that we'd wanna see in a response?

How a teacher want the response to look.

[00:38:14] Tal Oppenheimer: Yeah.

[00:38:15] Julia Wilkowski: And that's where we come in and say, here are the bunch of education use cases. Here's how the responses should be. We were really glad that the company has really decided to prioritize education and see it that the models really benefit. Everybody when they are trained for learning.

[00:38:33] Alex Sarlin: Yeah. And how do you use human feedback in that loop? Or did you use it to train the benchmarks? Do you still use human feedback to sort of evaluate the pedagogical effectiveness of the outputs?

[00:38:42] Julia Wilkowski: Yeah, we use humans. We shouldn't say we use humans. We work with humans to evaluate the quality of our models.

Again, through every stage of the process, through the beginning design phases. I talked about product development, but we also work with the core modeling teams to figure out what are the best. Practices and how the model should behave. And then we work with humans to evaluate the quality of responses.

You might have seen these in our Learn LM technical reports. You'll see that humans are generating conversations, humans are evaluating conversations. We try to automate some of that as well, but there's always a human in the loop. To make sure that we are doing the right thing pedagogically.

[00:39:18] Alex Sarlin: Yeah, and as we all struggle with, in the learning science field, there's different ways to evaluate outputs.

The lagging indicator is always outcomes, right? Yes. Did it work to actually achieve the learning outcome? That's a tricky thing to measure at the best of times. It's especially tricky when you have millions and millions of users doing all sorts of different things. I'm curious how you even begin to approach the outcomes-based research.

[00:39:38] Julia Wilkowski: Yes, you're right. That would be the holy grail to be able to say, okay, someone who used this tool or this product in this way over X amount of time. Increased their learning outcomes because we don't have ready access to that data. Although we do have a number of studies planned and in progress, we have to look at shorter time horizon evaluations as well.

So I would say the next shortest time horizon studies is we do some fast efficacy studies where we might give someone a pretest. To use a tool, give them a posttest and see whether or not they were able to increase their pretest scores. Again, it's a proxy. It doesn't measure necessarily long-term learning or transfers we would like to see.

We can also follow up with them over time to see if they were able to retain that information. Yeah, so I'd say that's the next best thing that we're working on, and then we do a number of other foundational studies and look at proxy metrics. Like evaluating against our learning science principles. Yep.

They're not a be all end all, but they are a reasonable proxy for what we think good learning looks like.

[00:40:36] Alex Sarlin: One of the newest products that came out from Google is this Learn Your Way product. Mm-hmm. From Google Research. And I saw the exactly that kind of fast efficacy research where exactly, where it was like people who did a personalized textbook versus a standard model or different kind of learning, uh, had, you know, a significant increase both.

Right after. And you know, I think a week later as you measured something like that, really interesting to be able to do that fast turnaround research and say, look, this is going in the right direction. It's working. And you do that for all the new products that come out in the Google Learn next team.

[00:41:03] Julia Wilkowski: We try to, yeah.

Yes. And you know, sometimes we need to get it out in the world and see how people are going to use it and see what use cases they have. So what we try to do as many studies as we can.

[00:41:13] Alex Sarlin: Yeah. So let's talk about multimodal. It's something we're really passionate about at Tech Insiders. Google is obviously out front.

I noticed in the Learn Your Way paper that you had to develop an educational illustration model just to be able to do education illustrations in line. That was really exciting. But obviously Google is really, really doubling down on video and audio to some extent as well. Yeah. The fact that Learn LM is baked into Gemini and Gemini is the fuel behind all a lot of these generative models means that potentially you could have learning principles baked into the multimodal outputs.

[00:41:43] Julia Wilkowski: Yes.

[00:41:43] Alex Sarlin: That's very exciting for somebody like me, and I'm sure for somebody like you, I'd love to hear how you think about it.

[00:41:48] Julia Wilkowski: Yeah. I'm not sure that yet those models are using the Learn Science principles to create the videos, although I'm gonna go check with the teams and see. How we might be able to work on that.

Yeah,

that be great. But

I'm also really excited about the potential for multimodal inputs as well as multimodal outputs. Right? In a future world, could instead of a teacher grading 50 essays, say, Hey, each one of you needs to have a conversation with this AI tool, and that will help assess your knowledge and your ability to explain things.

Right? So I think there's the voice. I think there's. It may even grade your presentation. Yeah. It might help you assess your physical skills.

[00:42:25] Alex Sarlin: Yeah,

[00:42:26] Julia Wilkowski: so I'm really excited about the input opportunities as well as the output opportunities, and

[00:42:30] Alex Sarlin: that becomes performance tasks or authentic assessments. It just gets a lot closer to the dream of education where it's not that everything is a proxy, it's a multiple choice test.

Exactly. The test, whether you can do this, you can actually do it. That's where we're all going. Thank you so much. Julia Wilkowski is the learning science team lead at Google.

[00:42:47] Julia Wilkowski: Alex,

[00:42:49] Alex Sarlin: we're here in London at the Google Deep Mind event with Maureen Heymans, the engineering lead for the LearnX team at Google. Always great to see you, Maureen.

This amazing. Yeah,

[00:42:59] Maureen Heymans: so fun to talk to you always. Yeah.

[00:43:00] Alex Sarlin: I feel just so excited by all the ideas I'm hearing here. There's so many interesting people from across the space, lots of diverse perspectives, teachers, organization, leaders, technologists. What has stood out to you so far about the event?

[00:43:13] Maureen Heymans: Yeah, it's been, I mean, so many things, so it's hard to pick, but I think what's been exciting is to see how many people are talking about encouraging curiosity and like really nurturing passions.

And I just had finished a book from Rebecca Wind Drop about disengaged teen, and that really resonated with me because she's talking about how we can. Encourage my students to become explorer. Yeah. And I think for me it's always been kind of how do you make learning so much more engaging so that you know, students follow the passion, they understand the meaning of learning, and they can end up becoming people finding solution to big problem in society.

Right. And so I think seeing that. Of how AI can actually be leveraged to make, you know, learning so much more engaging and intangible, but also really encouraging the next generation to be explorers.

[00:44:04] Alex Sarlin: A hundred percent. And Rebecca Winthrop just spoke at the conference and we interviewed her just a couple weeks ago.

So we'll link to the episode with her on from the Anti Insiders Podcast. Incredibly interesting. Yeah. That idea of explorers creating people who are metacognitive about their own learning. They don't have to sort of follow a script. They're not just trying to achieve and get good grades. It's sort of been the dream of what education should be, and it feels like we're starting to see ai with AI coming in.

We're starting to collectively imagine what that really looks like. What kind of education system would enable. More and more explorers. And I'm curious how you see the Google Suite, these amazing Google suite of AI tools is helping bring about that type of, uh, system. Yeah,

[00:44:42] Maureen Heymans: it's a good question and I always go back of like how I see how we can evolve learning at Google around those three evolution.

And I think all of them are really helping building this next generation of explore. The first one is really how do we make learning more engaging? Right? You know, those students, not every student is motivated to be explor, and so how can you make it more engaging? And so. Some ways, uh, to make it more multi-model, right?

Yes. If in addition to text, you start seeing visuals and you start being able to interact with the store with voice, and, but we know that, you know, students learn better, multiple type of content. And so I think that's the exciting first step of making learning much more engaging. And you see tools like you notebook a lamb where you can.

You know, upload your study guide and it's going to create a podcast. It's going to create some video overviews and we just found recently, you know, flashcard and study guides. So it's giving you many different ways to consume information and really build this deep understanding and also spark curiosity.

'cause you can find connection there, you know, a notebook and have mind map that shows you connection. So, so I think that's like one first step, how you make learning much more engaging via, you know, more visuals. More different type of content. And then the next one is how you make learning much more active and interactive.

Right. How do you start learning by doing, you know, first practicing, because we know that applying your knowledge help with learning, but also getting one step beyond of like really applying what you learn through your world application. Right. And so, you know, recently we launched test prep in a few of a product like in Gemini, in search, in Notebook where again.

Based on your own content, we are able to help you practice. Right? And then based on what you understood or your misconception will advise what you can practice more. I mean, like what you can revise, you know, contents to fill those gap and then practice again and go through that loop. So that's another way to make it much more interactive.

And then the last one, which I know everybody's been talking about, is how do we make it more personal? Right. I mean, we all know that if you can have a personal tutor for every student, you can really drive much more learning outcome. And so again, that's a space where if you can ground it on your content.

And so that's one thing we did with test prep. You can upload your own study guide. Right? You know, students wants to prepare for what they're gonna be quiz on. And so if it can be grounded on the learning context, it's much better. Yeah. But then also, how can we, again, adapt the level? Based on your understanding, how we can understand your progress and really adapt the learning so that, you know, we help you achieve your goal and we help you become lifelong learners.

Yeah,

[00:47:20] Alex Sarlin: and test prep is a very specific and really interesting part of the education system where people are highly motivated. It's exactly very specific outcome time bound. So it's a place where AI can sort of. Show its full capabilities. Exactly. Of all the different pieces of personalization. I've also noticed some reading products here this year that are really interesting.

I'd love to, you know, I asked the director of product management about this because so many of Google's products, I think is a huge strength of Google. Allow people to upload their own content. They're sort of. Content agnostic. Learn about, you know, learn your way. Yeah. Gemini itself, notebook. But you're starting to move into some reading products, which I think is a really interesting extension of the Google Education brand.

I'd love to hear you talk about that, if it's something you

[00:47:59] Maureen Heymans: Yeah, yeah, yeah. And of course, you know, I will admit that Google has so many nice exploration in this space that you might be thinking of some reading product that might, I think of like, read along. Yeah. Yeah. Read along. It's always been a product that became really popular in India, where it's really helping student learn and again, making it more engaging and visual.

But in general, right, we know that there's so many opportunity to help a student read and write, but the question is how to do it, right? Right. Yes. And because of course it's big concern about cognitive overload and you know, the writing is dead and those type of thing, and so. Yes, we've been thinking of how we can help with the reading, the critical thinking, building that critical thinking, but also helping build the skillset for writing and really, you know, having that struggle of thinking about the topic and really write down your thought.

And so still, like early on, you know, read Along is one of those first product, but really starting to think of how we can really help students build a skillset in the world of ai. Because this is difficult to add some challenges somehow you do as a writing, but how do you do it in a way that, you know, you first write down your own thought, right?

And then you can start interacting with the AI to get feedback, to brainstorm, to expand your thinking. Yep. And then do your own writing and then get. Feedback and get into that loop so that you don't dedicate everything to ai, but you really use it as an extension.

[00:49:26] Alex Sarlin: Yeah, Google always tackles like really large problems.

I think that's something I admire about it and you know, learning English and is the number one thing people are learning in the world is over 2 billion people learning English at any given time right now. And then of course, learning to read is sort of this universal barrier. To future learning. So it's really interesting to see Google sort of leaning into that space and saying, this is gonna unlock so many other things.

Yeah. And especially in an age of AI reading and writing and critical thinking, you know, becomes sort of a feedback loop. You have to be able to do it to use AI for anything else. Yes. Uh, it's really interesting. So let's talk about the video pieces a little bit more. Mm-hmm. Because I, I did something that just sticks with me a lot.

I feel like Google's been way ahead with the VO model, with just the video overviews, which now can come in different flavors of overview. I'm curious how you see Google and YouTube, of course, how you see Google's video strategy as regards to learning, evolving over the last few years. How are they thinking about, um, you know, what role video can play in learning?

[00:50:20] Maureen Heymans: Yes. I mean it's going back to making learning much more engaging, right? Yeah. We see that, you know, if you can have a combination of visual and text and you know, it's really makes learning much more real and engaging and helps with understanding. And so, you know, of course we have an amazing community of creators on YouTube and so we wanna leverage them.

And I always think, you know, those are really amazing teachers that managed to make learning so engaging. And so we should leverage them, but we can both use AI to again, be an extension of those creators, right? They can make the video content much more engaging and interactive. And so we launched some of those features where, you know, as you watch a video you can pose and ask question, but of course we can, you know.

Content creation, especially learning content creation takes a lot of time. Yes. Right. And so if we could help creators kind of make it easier for them via VO and other creation tool to make it easier for them to produce high quality content, now you can really reach a large amount of students. That's one way to really, you know.

What creators can do. The other way I can imagine is also like, again, you can create video and visuals for any content. Yes, any concept, right? I mean, of course you will have lots of amazing content on video platforms, but then it might not cover the four, you know, scale of topics in different languages and personalized to you, right?

So again, you could imagine creating those on demand content based on your context, based on your interest. You know your language and really have those personalized video experience that it's been a concept just for you and also based on your misconception and all of that. Right. So I think Notebook alone is the first glimpse at this because you can upload your content and it's going to create some video of a view.

[00:52:03] Alex Sarlin: Yeah.

[00:52:03] Maureen Heymans: But I can imagine this expanding much further. Yeah.

[00:52:06] Alex Sarlin: Then that combination of sort of tapping the massive YouTube video library and combined with generating new video at the point of need is a pretty killer combination in terms of.

[00:52:15] Maureen Heymans: Exactly. Yeah. Yeah. And that's really, again, you know, we're not trying to replace those amazing human creation.

Right. And I think it's a great example of how do you bring the best of human, I mean, especially in this case, creators that are so good at producing high quality, engaging content, and the best of AI to not only save them time, but also. Bring that content to the next level by making it more interactive, more personal.

Yeah.

[00:52:38] Alex Sarlin: One really interesting theme of the day it feels like to me, is almost like intellectual humility. It feels like it's bringing together these incredible people from all over the world and sort of admitting that nobody has. The full suite of answers about what AI education should look like or the best practices for it.

Instead, we're sort of forming them together and trying to sort of figure out the best next step. It's really impressive for an enormous company like Google that's had so much success to sort of come with that humble approach, and I think the Learn team has always exhibited that. I'm curious how that sort of informs your work, that idea of, you know, we wanna know the answers, we're heading that way.

We're gonna test everything we do, but we're not gonna sort of. Just drop a whole bunch of solutions and say, this is what AI should look like. Yes,

[00:53:18] Maureen Heymans: definitely. I mean, optimistic. So in general, I feel like long term we'll find all the solution to problems, but clearly we need to also be aware all the challenges, right?

Mm-hmm. And so I think, and, and we know at Google we are not going to be able to do it at all. That's where we need to work with experts and we have in-house. Experts, which is great. You know, folks that are pedagogy expert, that used to be teachers, and you know, I've been reading a lot of book about learning, but I'm not educated as much as those people.

But then, you know, having external expert, that's what they do, is really think about those challenges. Think about how you are going to deploy those tools at c you know, directly in them. Education system globally, I think is so important because I think I'm optimistic of the future, but there will be a lot of hurdles along the way.

Yeah. And we need to partner with everyone, you know, teacher first, but then of course leaders in the space. And so I think it's amazing to hear everybody. It is an excitement for ai, but at the same time, you know, it's important to. Be aware of the challenges, right? Yeah. Because when you're too optimistic, you miss the big problems.

And so you know, those discussion, you know, with so many people bringing the perspective from different side of the world, they've been super interesting to open up those challenges. So we identify the right solutions.

[00:54:36] Alex Sarlin: People just representing so many different education systems and perspectives and approaches.

It's really been inspiring. The last question I have for you is something that I've also heard as a theme, and I know you think a lot about, you know, for the future of ai, which is getting to sort of multiplayer collaborative mm-hmm. Social learning. Mm-hmm. Yep. It's something that. You know, I think is the dream when we think about what AI should do in a classroom, it should help people learn from each other.

Exactly. It should help teachers and students build and maintain relationships. Students maintain relationships, you know, parents connect to, you know, just strengthen the connections between all the different stakeholders in the educational ecosystem. How is Google beginning to think about this? I think we're starting to see an evolution of all of the AI tools to start to become more social.

Social. Yes. What does that look like for you?

[00:55:21] Maureen Heymans: I mean, of course, as you mentioned, the first piece is to collaborate with experts. But when it comes to, and collaborate with teachers and, you know, make sure teacher are in the loop, you know, AI should never replace teachers. They're so important, but we can be, you know, a predictivity boost for them so they have even more time for human connections.

So I think that's the first piece. It's like enabling teachers to have. Tools that can save them a lot of time. So they have more time for human connection. Yeah. In the classroom. But then I think there's also a lot of power in connection among students.

Yes.

And of course, AI has a little bit of risk of being a lonely tool, right?

Yes. Then it's available 24 7. It's not such a mental, it's great, but you know, it's like 1 0 1 relationship is ai and so I think, you know, big opportunity again to make learning more social. Because not only it's engaging, you know, like students crave for social connections, right. Often when you, they try to use those tools to use it for shortcuts is because they wanna spend time with their friends.

Right, exactly. So, so I think if we can make it more social so people are learning from each other, they, it's great. But I think it's also, I mean, know we discuss a lot about 21st century skillset and clearly. Collaboration is such an important one.

Yes.

And so if you can build tools that encourage collaboration, like hearing from, you know, others' perspective and building Sion together, this is so key.

I mean, it's so always what I tell my kids when you know, they might complete about, they. Group project. I was like, I need everything. And I'm like, you know, that's part of what you'll experience at work, right? It's like, you know, whether you're part of a team or leading a team, you know, you'll just want to empower everyone to participate and there'll be some people, everybody's bringing their own skillset.

And so I think encouraging kids to have to build those collaboration tool through ai. I think is going to be so important. I won't say that we have a full solution there. I think it's, again, something that is really exciting and I think we need to think a lot about it. Not the easiest part of how to use ai, but I think is going to be critical.

[00:57:20] Alex Sarlin: Yeah. It feels like the next chapter. Yes, exactly. There was web. Yeah. Maybe that's the false

[00:57:24] Maureen Heymans: evolution. Right, exactly. Like, you know, after making it personal, how do you make it social? Exactly. And collaborative.

[00:57:30] Alex Sarlin: Exactly. And the ability to share, you know, the Google cam, you know gem canvas. You can share articles.

Yeah, you can share notebook.

[00:57:36] Maureen Heymans: Yeah. And

[00:57:37] Alex Sarlin: share whole notebooks. Yes. I mean, there's already some implicit, you know, I think

[00:57:40] Maureen Heymans: sharing is like the P zero zero of like social, right? It's like, hey, you create great resources now. You should be able to share it. But then. Now, how do you make it much more interactive, right?

Yes. How can you, you know, I was chatting with someone yesterday, it was like, how can we have notebook lamb? You know, people can comment on the notebook lamb. Yeah, that's interesting. And then, you know, go to the next level of like learning by doing together. Right? Yeah. I think that would be super powerful.

I'm really

[00:58:06] Alex Sarlin: excited about that moment. Rebecca Winthrop did an exercise in the last session where she had us all sort of remember our own education and what worked for us. And I realized as I was doing it that I was thinking social thing after social thing after exactly. I thought like. Five social things before I thought of any class or any subject and I was like, wait a second, there's, I feel like there's something in there.

Yeah. Human

[00:58:24] Maureen Heymans: connection is what's special about the the bathroom. Exactly.

[00:58:27] Alex Sarlin: Bringing people together and just the relationships you build. You know, the studies about college students are much more likely to graduate if they feel like they have a professor who cared about them. You know, it's just like some really interesting sort of social currents in education.

Yeah. I feel like it's the next generation

[00:58:40] Maureen Heymans: and, and the biggest solution people found is true by bringing. People with different skillset. Right? Yeah. And so I think if you, that power of collaboration, 'cause everybody brings the unique value to the table is so important to solve some of the biggest,

[00:58:54] Alex Sarlin: yeah.

It also unlocks the power of Neurodivergence, I think, in the classroom. Yeah. So everybody is bringing their strengths to the same. Problem solving things together. Working together. It just feels like a, a vision of education. I'm starting to get very, very excited about. Exactly. And I think Google, Google's gonna be on the forefront of it as it has been.

Thank you. Yeah. So now

[00:59:11] Maureen Heymans: we'll add it as my false evolution of learning.

[00:59:14] Alex Sarlin: That's next. Maureen Heymans is the engineering lead of the Google LearnX Team. Thank you so much for being here with us. Thank you for inviting it. Always

[00:59:21] Maureen Heymans: a pleasure to talk to you. Always a

[00:59:22] Alex Sarlin: pleasure. Thanks for listening to this episode of EdTech Insiders.

If you like the podcast, remember to rate it and share it with others in the ed. Tech community for those who want even more. EdTech Insider, subscribe to the Free EdTech Insiders Newsletter on Substack.

Alex Sarlin

Host

Ben Kornell

Host