How to Build Duolingo’s Learning Engine with Science and AI | Bozena Pajak by Kazuki Nakayashiki

How to Build Duolingo’s Learning Engine with Science and AI | Bozena Pajak

Kazuki Nakayashiki

May 16, 2025 7:34 PM

Summary

Introduction to Bozena Pajak (00:08-00:28): Bozena Pajak, VP of Learning at Duolingo, shares her extensive background in language learning and cognitive science, including a PhD in linguistics from UC San Diego and her role in enhancing education technology.
Journey to Duolingo (01:06-06:30): Bozena discusses her early passion for languages in Poland, her academic journey, and the decision to join Duolingo in 2015, highlighting the alignment of her expertise with the company's needs at the time.
Initial Role and Contributions (12:30-15:37): Initially, Bozena focused on improving course content and conducting research to evaluate the effectiveness of language lessons. She describes how her academic experiences informed the development of Duolingo's language learning methodology.
Balancing Science and User Experience (16:14-21:14): The discussion touches on balancing research-backed learning methodologies with user preferences, emphasizing that while scientific principles guide product development, insights gained from user interactions are equally crucial.
Impact of AI on Language Learning (40:26-44:58): Bozena discusses how Duolingo has leveraged generative AI for content generation and creating conversational features, emphasizing the potential for personalized learning experiences.
Future Vision and Legacy (49:53-57:36): Bozena expresses her desire to continue contributing to Duolingo and advising other startups, emphasizing her commitment to education and the importance of ensuring access to learning opportunities for all.
Advice for Learners (54:34-55:46): Her main advice for learners is to maintain consistency in their practice, stating that daily engagement is more essential than the volume of content consumed.

Transcript

Hi, everyone. Welcome back to another episode of Glasp Talk. Today, we are excited to have Bozena Pajak with us. Bozena is a seasoned expert in language learning and cognitive science, currently serving as the VP of learning at Duolingo. With nearly a decade at Duolingo, she has held many roles, including director and a staff learning scientist, shaping the learning experiences for millions worldwide. Beyond Duolingo, Bozena holds a PhD in linguistics

from UC San Diego and is a mentor at Texas and serves on the Cognitive Sciences Advisory Board at UC San Diego, reflecting a deep commitment to advancing education and language research. And today, we will dive into her insights into the science of language learning and her vision for the future of education technology. And thank you for joining today. Thank you for having me. Very excited to be here. Thank you. So, first of all, you've had a fascinating journey from academia to leading

the learning team at Duolingo. And so, we are interested in what inspired your passion for language learning and linguistics, first of all. Well, what inspired my passion, I think it started really when I was little. And I grew up in Poland. In Poland, everyone tells kids that they need to study languages because that's really the way to improve your career prospects. In Poland, well, we speak Polish and basically nobody else outside of Poland, very few people want to learn Polish. And

so, in Poland, we need to learn other languages. So, that's something that I've heard a lot throughout my life and something I heard growing up. And so, when I was little, I just tried learning languages and I really liked it. And so, I even had this idea that every year I would start learning a new language. And I did that for some time. Of course, I didn't continue for very long. You can't, it's hard to add too many languages. But I tried that

and so I enjoyed learning languages when I was in Poland. And then I taught languages as well. And then when I ended up going to the US to do my PhD in linguistics, I ended up specializing in how people learn languages. This is something that I was doing myself. I did for many years. And so, I wanted to research this topic. I wanted to understand how this works in our minds. How do we learn and how can we make learning work better?

And so, in my research, I did a lot of experiments, teaching people many languages or just parts, small parts of a language. And then trying to figure out in what condition people learn better so that we can improve our methodology. And so, that's something I got to learn a lot about during my PhD and after when I was doing research in the academic world. Wow, so fascinating. But was your research, like a PhD research and a postdoc research

related to what you do today at Duolingo? Or is it different or similar? It's very related. So, what I did in my research, I really learned a lot about how people learn languages, how we should be teaching. And then when I basically decided to leave academia and I got this Duolingo job, which itself I can tell more of the story how I got it. But the job essentially meant that I was using everything I learned before. All my experience

doing research, understanding how we learn, my experience learning languages on my own, my experience teaching languages, it all really started coming together in my job at Duolingo where I've been able to apply it and build a product that teaches languages. And I've been trying to apply all my knowledge on this topic. Wow, but you were the research associate and lecturer at the time, right, at Northwestern

University, and then you joined Duolingo in 2015. But Duolingo started in 2011 and 2012, around that time. And did you know Duolingo at the time? Or how, yeah, I'm curious how did you join Duolingo? Yeah, yeah. So, I like to say that it was written in the stars because it just all felt it's a pretty incredible experience. So, yes, I was working at Northwestern University. I was a researcher. I was a lecturer. And I really thought that I would have this

academic career, that I would stay at a university. But I was having doubts whether I, you know, whether I actually wanted this kind of job. I really liked research, but I felt like I really wanted something where I can feel the immediate impact of my work. And I wanted to do work that moved a little faster. Research, unfortunately, goes very slowly. It's very hard to move quickly because it takes time to run studies, to write papers. And so, for me, it was a little bit

frustrating. And so, I was thinking about what else I could do that would really use my expertise. And I had a lot of conversations with my husband about it at the time. And one day, we were talking about it and he asked me, well, if you were to work anywhere else outside of academia and whatever company you wanted, what company would that be? And Duolingo immediately came to my mind because I was using it. I knew about it. A friend recommended

it to me. And so, I was using it just to actually to practice the languages I knew, learn some new languages. So, I was already familiar with the product. And I thought, oh, yeah, that sounds like a company where I could actually use my expertise. And so, in that conversation, my husband and I just checked Duolingo's website and we saw what kinds of jobs they were hiring for. And then, there was that job basically written for me. And so, at the

time, so yes, that was 2015, Duolingo was looking for someone with expertise in learning and teaching. At the time, they didn't really have any experts with this kind of skill set. So, they wanted someone with a PhD in linguistics or a related field, someone with experience teaching languages. Basically, everything was right for all the requirements. And so, I decided that I just need to apply for that job. And so, I sent my application, I think, the same day. I heard back

the next day. And I was invited to go to Pittsburgh to the office. And then, at the end of my interview, Luis Bonin, the CEO, basically told me, well, we like you. We want to hire you. So, it was that day. Yeah. I mean, then we had more follow-up conversations. But yes, it was a fast process at the time. At the time, Duolingo was pretty small. It was still a small startup. I think we were maybe about 45 people when I started. So, things moved faster

in terms of hiring as well. But yeah, so I didn't even have that much time to think about it. I I decided that I need to take this opportunity. And I moved to Pittsburgh and no regrets. Wow. So, do you remember your first impression on the Dewey's and the team at the time? Or was it a duo at the time? Sorry, there's no question. Yeah, yeah. I mean, so the office was, you know, we switched offices as we grew. So,

back then, it was a pretty small office. I remember that everyone was very friendly. It just felt immediately like such a nice community. I remember that I spent a whole day there and we all had lunch together. That's when I learned that at Duolingo every day everyone had lunch together sitting at the same table and there were just great conversations and we still do this to this day at Duolingo. There is an hour where everyone

just has lunch and we want people to be able to have those conversations. So it felt like a great place, very friendly and very creative. People were very curious, very interested, constantly trying to solve some problems and come up with something new. So I immediately loved the whole company, the environment, the culture because it felt like there were a lot of smart people and everyone wanted to move really quickly to solve problems,

to build something useful and that was very appealing to me. Really interesting. You mentioned when you joined Duolingo, the team was around 45 people, 45 members, but there were other learning scientists as well or you were the only one? No, no, I was the only one. I was the only one. So Duolingo for those first few years didn't have any learning scientists. I see. And what is the team then? So there are many engineers mainly?

Yeah, there were mostly engineers. There were some product designers. There was one person who, actually a couple of people who were more like machine learning engineers. There was one product manager. Yeah, but mostly engineers. Okay. So meaning you needed to set up how you work and then how you collect data and to analyze data and set up and what... I'm curious what you did at the beginning and yeah.

Yeah, good question. Actually, it took me a long time to really figure out what I should be doing because the company existed already for a few years and so they had their own way of doing things and they didn't have anyone in my kind of role, a learning scientist. So they hired me because they felt like they should have someone with this kind of expertise. But when I arrived, they didn't quite know what to do with me. And so I had to figure out where I could

actually be useful. And so I did different things. At the beginning, I just tried to learn a lot of different things about what was happening. Some of the work was just improving our courses. And so initially I was even going into some of the courses and actually rewriting the content and trying to run some experiments with tweaking the content and changing how we were teaching certain things and testing it through the framework that the company was already using of A-B testing.

So I didn't have to set that up. That was already something we had. But I was able to set up my own experiments that were more about the content of our courses, the sequence in which we were teaching certain things. I also did studies that were more kind of academic style, trying to evaluate the effectiveness of our courses. And so actually recruiting people from our learners, from Duolingo users, and asking them, for example, to take a test in a language,

take like a proficiency test to see how much they've learned. And so that also gave us some insights about how well our courses were working. And that is something we continue doing to this day, actually both things, with experiments, with modifying, improving the course content, and evaluating the effectiveness of our courses. Now we have a whole team of learning scientists who do just that. They just run research studies. And so initially there was a lot of that,

and also collaborating with others on just coming up with new things that we wanted to build. So working with the product designers, with engineers, thinking about, okay, what should our lessons look like? What kinds of exercises should we have? What other types of experiences should we be building to make sure that we teach effectively? And just generally, I was there to to provide guidance on what is it that, how do people learn languages? What are the different

pieces that are important? And what are the things that we already covered that we do well? And what are the gaps? And so I was there to really provide a framework for what are the things that we should be focusing on to teach better. Since you have academic background, and I was curious to know, like, how did you balance like a scientific rigor with the practical aspect of building a widely used language learning product? I mean, do you, I mean, when in choice, when you have choice,

oh, should we, did you follow your science, how to say, backed path? Or, like, if the metrics is correct, if the user preferred that way, I mean, scientifically, it's wrong, but user prefer this way. And to choose that way, I was curious how to balance that. I'm not sure there has ever been like a tension of this type, exactly. I would say it's always been both, trying to apply the knowledge I have about the research, what we know about learning, about teaching,

but then also looking at our experiments. And a lot of what we are building at Duolingo still, and we're trying to apply research, but it's not always clear how to apply it because the research generally is done in different settings. So maybe there are research studies done in, well, like the types of experiments I did, where I had students come into the lab and they, and they learned, you know, they didn't learn in an app. They just had some lessons that they

went through. Or research is done in the classroom and people evaluate certain methods, looking at how students learn in a classroom setting. What we're building at Duolingo, it's different. It's an app. And it's an app that people can quit at any point. So we also can't force people to go through the experience. Like, you know, when I was setting up my experiments, well, people came in and they had to finish it. Or at least, you know, they were paid for it. So

they finished it. Then in the classroom, it's kind of hard to quit. You're there. So at Duolingo, we really had to, and still have to, innovate in just how we even apply research findings and be very creative about it. So we might know, for example, that something works, but how do we actually, how do we actually implement it? So there are many, just to give you some concrete examples, what really works the best for us is try to apply research that's more

about the general principles around learning. Like we know, for example, that people learn well when they practice through recall or retrieval. So instead of reviewing things by just kind of looking at, let's say, your notes or a textbook, you try to bring out things from your memory. Try to remember on your own, kind of like test yourself. Like I know, okay, well, how do I say cat in Spanish? And I try to remember it on my own. Oh, gato. So that's really the most effective

way to practice, to review. That's just known from research. It's been very well studied. And so we need to come up with our own way of implementing this kind of recall in the app. And so we need to come up with exercises that actually help learners bring things out of their memory. And so we might give you exercises where you need to, we give you a sentence, let's say in English, when you're learning Spanish, and we

Tell you okay say this sentence in Spanish, but you kind of have to formulate it on your own, but that exact Exercise the way we implement it on Duolingo, you know, it doesn't exactly Exist elsewhere. We had to we have to figure out how to apply it in a way that that would be intuitive to people to do And so that's that's what's useful for us when we apply those more general principles more general Findings about learning. It's not it's much harder for us to apply some research findings that are specifically about

Let's say in a classroom. Okay, this specific activity maybe works. Well Because then we well, we don't know Well, is it is it about the environment in which? People are learning is about the teacher. It's not clear how exactly to apply something This specific So So I would say we're We always Just taking the research that Exists and then combining it with the insights we have from implementing it in our app

Seeing how learners react to it and learning a lot Through that and figuring out how to motivate people to actually do The kinds of exercises that are well supported by research. I See but in that sense, do you do you read academic papers to nowadays and it also in that sense You know I think a lot of researchers at universities or or research institutions want to collaborate with during or to apply The research results to see if it works in the real world or not. And does that happen? Oh

Yeah. Yeah. I mean, so yes, we I Try to read research pages my whole team We try to stay up to date on what the research is so we go to conferences and We try to make sure that we just know what the developments are so that we we can apply any new findings We also try to stay in touch with the research community in different ways For example, we invite Researchers Professors to to our colloquium series so they come and give a talk and and we can have meetings with them

so that's also a great way for us to to see what research is Getting done and then interact a little bit with with people to see what ideas they have what? Give them some insights about how it's working on our side how it What happens when you try to apply those findings? People reach out to us to collaborate and we we collaborate sometimes so we even offer some research grants to on specific projects so so so researchers have done this

And they they often study People learning on Duolingo We can't unfortunately support So many collaborations, so we collaborate some but we definitely get a lot more requests for collaboration than we can support But how do you choose so Any criteria? So usually what we choose to do is through research grants so we might have like a specific Project in mind. For example, we want we want to Evaluate how well Duolingo

teaches let's say speaking skills and so we might Then create a call for proposals like within this topic The researchers can propose their ideas how they would assess this and then we we evaluate those proposals. It's basically like other Research grant applications. We we look at the team. We look at the The research plan they propose and then then we pick the strongest one. So that's That's generally what we have done

I see And if you can choose any institute or if you can choose any so project or research So what would you like to collaborate or research? Do you have any like? Oh, yeah, did you get my question? Um Yeah, like anything I would collaborate on Ah, there are so many Interesting topics. Yeah, I don't know if I have like a um a favorite one, I mean something that's um that's definitely on my mind a lot is how how well um

Learning on Duolingo applies to the real world something we think a lot about we want people to learn a language on Duolingo And then they'll actually use that language somewhere in some other context and so that's um That's something that uh, that's very interesting to us like a study that um that evaluates um How Duolingo learners maybe use their languages, uh in in different real life contexts Interesting, but I guess you already did some kind of research around that, right?

Not yet On on this topic yeah on the topic We haven't done that much a little bit a little bit See, okay. Maybe our audience will reach out to you. Hey, yeah, we would like to collaborate with you. So yeah, I hope I'm curious like, you know What was the like core metrics or KPI before you joined Duolingo? What were they pursuing? And how it's changed after you joined um I so before I joined and when I joined, um, mainly we were it was it was very simple

We were really just paying attention to um to Daily active users. That was really the main the main metric um, and then uh after I joined things, I mean we generally expanded the the types of metrics that we we've been looking at uh, so definitely a big one has been Looking much more at retention so, you know how how many people come back the next day and A week later two weeks later. Um, and that that is something that um

That really tells us a lot about just the quality of the experience um is it uh Is it good? Is it intuitive? Also, is it Just not fresh too frustrating not too difficult. So whenever we um, we make changes um to improve How we how we teach we often see, um, see big gains in retention, um, because learners can actually um Progress more easily through our courses. They don't get stuck. Um, which can get frustrating if you're learning a language and then

Uh, it becomes suddenly too hard and so those are the kinds of things we've been trying to to improve um We also started adding a lot of other other kinds of metrics that we track like for example Time spent learning we want to make sure that people Spend time in the app, but they actually mostly spend time On the educational pieces of it. So actually in in lessons learning things We also later added um time spent learning well, so you can be learning

but there's also Really the most effective way of going through the content which is really going down the duolingo path instead of maybe doing some other Quests or going back and review the material so we want to also optimize Time spent learning. Well, how how much time people are spending really on the next? Next thing that's that's on their path We also look um at um, for example different, um Subscores for example

Time spent speaking time spent listening. We want to make sure that people are spending enough time on the different Parts of language that we're trying to teach them um, we look at content difficulty so we want to make sure that What we serve to our learners Is at the appropriate level of challenge and duolingo personalizes the experience and so Uh in general, that's something we pay a lot of attention to we want to make sure that

What we give you as a learner is is right. It's at the right level so if you're struggling we'll give you a little bit easier exercises if you're Answering everything correctly. We'll give you slightly harder exercises and overall we track The difficulty of the content and on average we want to make sure that uh, we are pushing people to do um more difficult things with with the changes that we're making. So those are

just some examples. I see. Interesting. And then, by the way, I like the mascot duo. And he or she sent me like a funny he sent me funny notification. It's time to Spanish or Banish and it's something like that. And that's so funny. And do you think that mascot duo influence impacted on the metrics? Or I am curious how much that mascot impacted on the metrics and retention and activation? Yeah, that's a that's a good question. I mean, we think that

the duo has a big impact. Initially, we didn't we didn't really know how much of an impact he could have. It was just a little design. It was an owl because education. Owl is kind of a symbol of wisdom education. So we we picked an owl, but we didn't really know how much it would develop. But gradually we we started leaning more into it. Duo appearing in different places in the app. And people just really reacted very

positively to duo. And so we we started doing much, much more with him. And so on social media, on TikTok, and then duo sending different notifications. So at some point, Duolingo became famous for passive aggressive notifications like oh, you know, these these reminders don't seem to be working. So we'll stop sending them for now. And people really liked that it felt it felt funny to them. Passive aggressive, but people actually

came back after those types of notifications. So, so we started playing much more with what else duo can do. And now, yes, now we have well duo has been on red carpet in Hollywood. He's been in many different places. And yes, we're just continuing to expand what he does. But duo was dead, right? Back in February, I saw the tweet and post duo was dead. But is he alive? Did he come back? So it turns out that he he faked his own death.

So if you go to, yeah, if you go to Pittsburgh, I can meet you. Yeah, yeah. But was do exist? Yeah. But did do exist when you joined? Yeah, yeah, do exist. It's just, it was a much smaller part of the app. But Duolingo always had, or since very early on, it had a little icon of duo, the owl, and he was there in the lessons, but it wasn't nearly as prominent as it is now, especially with the whole the whole kind of costume of a duo

that started later. I see. And as a research perspective, I was, I'm curious, you know, how much that kind of mascot impact on the people's learning, or language learning? It does. Did someone do this research? That's a that's a great idea for research. Actually, I don't think what we haven't done this kind of study, and I'm not sure anyone else has. But yeah, that's, that's interesting. I mean, we do know that people, people really like duo. And so

that's, that's part of why they like the app, and they come back, and they, they want to engage with with the exercises, because duo is there and cheering them up. So he's very supportive. And I mean, sometimes he's, he's a little mean. That's a lot of different emotions. So so we know that people, people really enjoy that. And that's partly at least why they're coming back, because it's it's part of this whole fun

experience. But we haven't studied it in a rigorous way. Maybe we should apply to during a grant program, I think. Yeah, maybe. So, yeah, then, what, you know, during during the 10, almost 10 years, a decade experience, and did you find any surprising findings from your research or product development? During go, I mean, you expect, I mean, let's say you expect something, but it turns out different results or something like that.

Yeah, I would say, for me, the most surprising things have been just around learner motivation, because we've had a lot of ideas and even knowing that certain things should be effective, and we should implement this. And then maybe we would implement some of those things. But learners weren't engaging with it. And so our engagement metrics, like the active users or retention, sometimes were down. And, and we often would

realize, oh, well, we implemented something that was just too difficult. And to me, that was, that was surprising at the beginning, that it mattered so much. That people may maybe retrospect, that's not so surprising. People don't like to do hard things. But that's, that's definitely that that was something that surprised me at the beginning that, oh, people really didn't want to do something that's a little bit

harder. Some people do, but the majority didn't, and maybe they didn't want to come back as often when when we made lessons more difficult. And when we included more exercises, for example, with the recall, I mentioned, where we want people to actually recall information on their own. And those types of exercises are hard, they are kind of, they can be unpleasant, because you need to think very hard. And, and some of those

experiments with these types of exercises, well, they would just hurt our engagement metrics. So yeah, that was surprising to me at the beginning. Now, I feel like it's, I've learned over time what what to expect. But initially, that was definitely a learning. Interesting. And there's a daily streak, the numbers and users can see, and we were talking before starting the conversation. But what, do you know what's the longest daily streak that user

has? I mean, it's been started in 2011, 2012 means it's been 12 or 13 years, right? Does someone have, I don't know, over 10,000 days or more? I don't know. Yeah, that's a good question. I don't. Yeah, I don't actually know what's the longest streak, but people definitely have, you know, at least over 1000, maybe even a couple of 1000. So some people have been on it for a long time. But yeah, I don't, I don't know what the longest streak is.

And I think you should ask the CEO, the founder, right? So what's your streak? He uses Duolingo every day. So I don't know if he... Maybe he doesn't like any day. Yeah, that's, I would not be surprised. He uses it for multiple languages. And now also, when in math, music, chess, other subjects that we started adding. Did you see Luis speaking 10 languages or so? Because if he learns, you know, every day, some languages must be able to

speak many languages or can't. He does, he does. And he's been, he's been focusing a lot on, actually, for many years on French. And he only studied French on Duolingo. And that's, that's actually pretty amazing. He, he said he traveled to France, he was able to communicate. So he's, he's able to watch movies in French. So yeah, that's, that's a language that he's been focusing on a lot. But he studied others, often testing things like

Japanese, as we're trying to improve that course. We're trying to learn Portuguese or Swedish. He's, I think, German, he's trying German these days. So he tries to test different courses. Amazing. And since 2022, ChachiPT came out, and I mean, AI, Rational Engaged Models came out, and I'm curious how AI impacted on, like in education or language learning. Did it change completely, or did it improve how people learn? I'm curious, yeah, the impact of AI.

Well, definitely impact has already been huge, and generally on education. I mean, still, I think we're all still figuring out how to really use it in the best possible ways in general for education, like what does it mean for classrooms and so on. But I can tell you that at Duolingo, we've been really trying to take full advantage of generative AI, and it's been transformative. So for example, we immediately started

using generative AI in two different ways to improve our product. So one way was to help us generate content. So before, all the content was written manually, and it was taking a long time to build new content, to improve content, and we're constantly trying to improve our courses. We want to refresh the content. We want to build more advanced content. And that was just taking a long time, and we have a lot of courses

that we just haven't been able to improve in many years. And since generative AI came out, we've been setting up just new processes to now generate content instead of writing it manually. And then, of course, we have a lot of checks to make sure that the content follows our guidelines and is of good quality, but all of it is done through AI. It's just so much faster. So it's allowing us to really create

a lot more content. So for example, a month ago or so, we released 148 new courses, which we created in the past year. And it was thanks to AI. Before, it would have taken us multiple decades to create this amount of content. And so that's something that's helping us a lot, so we can move much faster in our improvements and just adding more courses. And then another thing that where generative AI is helping, it's allowing us to build

just new functionality, new features that we just couldn't offer before. So for language learning, something that's very important is actually speaking, having a conversation. And before, the technology wasn't there to really have this kind of conversation. And now, it's there. And so we started developing immediately a feature where you can have a conversation with a bot. So maybe you've seen a video called

With Lily, where you talk with this teenager with purple hair. She's kind of unimpressed by anything you do. And so we've created this feature. We continue improving it. But it's already been great to give our learners practice having a conversation at different levels of proficiency. So we try to make it appropriate for you, regardless of where you are in your journey. But it's particularly useful for people who are

a little bit more intermediate, more advanced, to actually practice talking. So that's something that generative AI has enabled us to do. And so I feel like, in general, when we think about education, generative AI has a lot of potential. It's just a matter of finding the right ways to use it. Like for languages, like I said, for Duolingo, it's been great. For content generation, it's been great for all those conversational features. I think generative AI

is just going to be really great for personalization. So that's something that I think will be the future, where everything will be more and more personalized, because we'll be able to adjust how we teach much more to each learner. Because right now, or before, having a teacher with a large classroom, it's much harder to personalize the instruction for every single person. But with this kind of technology,

I think we'll be able to create content, to create lessons for each person much more easily. And so that's something I'm very excited about. Yeah, totally. I totally agree with the AI input, the personalization, so that people can learn better. Because learning by doing is a better way to learn something effectively. That's what I know, what I understand. And I'm not a scientist, but that's what I know. With AI, people can speak

really, and so that they can practice more. And I think it's a better way to apply what they learn to actual situations. I think it's a similar environment. And another question is, in 2016, you published, which countries study which languages, and what can we learn from it? And I love the blog. It's fascinating to see how history, immigration, geography, and culture influence which languages different countries choose to learn. But do you think AI

will change this? I mean, the language people choose to learn, eventually? Because with AI, eventually, do we really need to learn language? That's the ultimate question, I think. Yeah, that's a good question. Maybe it will change it. But what we see is that, at least when we look at people who learn languages on Duolingo, people learn for two reasons. One big bucket is people just learning as a hobby. They are just trying to do something productive with their time.

And maybe they live in the US and they have some heritage. Maybe their grandparents are Italian, so they're just trying to learn Italian. Maybe they want to speak a little bit, but a lot of it is just trying to connect with some cultures better and to spend your time productively. Just do something that's enriching. And that's just not going to change. People in those buckets, they don't really need to learn a language. It's just a way

for them to improve themselves. And so we think that that's not going to go away. People will still want that to really connect in the same way. And then another big bucket is people learning English. I was in that situation growing up in Poland, having to learn English. That was one of the languages that I knew was very important. I think both of you learn English. For many people who live in countries where English is not spoken,

English is a language where it's important for career development. It's just a language that you want to actually speak or at least understand. And so even if translation works really well, that might not be enough for English. Would we be doing this podcast through some kind of translation? It would be much harder. There's always a lag with translation. You can't actually that easily use the language, have a conversation, do business,

interact with people. And with English, it has become this global language. And it seems like that's also not going away anytime soon. So generally, I might change it a little bit. But I think a lot of this need to learn languages will stay the same. Yeah, totally. Yes. Yeah, I totally agree with that. And so about the future. So do you have your future? What's the future vision and what's next for you? And I'm curious, would you keep

working at Duolingo maybe, but do you have some future vision for your career or life? I'm definitely happy to continue working at Duolingo. I feel like, especially now with AI, things are changing so quickly. It's all very exciting. I think there's a lot of potential to really do what we're trying to do, provide education for people. So I think I would like to be part of that. And so I definitely hope I'll continue doing that.

But I'm also at this stage in my career where I'm excited to give back and to really help others. That's why I started advising other startups just to pass on some of the learnings I've had and to help others. So that's also something I hope I'll continue doing more of because it's also very, very satisfying for me to see others succeed. Thank you, yeah, I love that. And yeah, in that sense, you are the mentor at Texas

and also advisory board at UC San Diego, right? What, as a learning scientist, what do you do, let's say, for Texas as a mentor? Do you advise startups from the learning perspective? Yeah, from the learning perspective or just advising from anything. I've been at Duolingo for a long time, so I have a lot of experience generally thinking about product development, how to build products that are intuitive, motivating for users.

And so that's something I also just advise the startups on, just kind of generally giving them this perspective on building intuitive products. And do you publish those learnings through website or newsletter, devlog or X? So do you do like media? So to tell your lessons to the people? Sorry, can you repeat the question? Just do you publish your learning through website, newsletter? Wow. Well, I try to post on LinkedIn a lot of my learning,

so that's what I try to do. I actually write some things and post there. Maybe one day I'll write in some other way. Maybe I'll write a book one day. Oh yeah, I would like to read it. And you're just personal learning, so you have been learning a lot of things, not only language, but also business, product development as a mentor. Where do you keep those notes or learnings? Do you write on like paper book, paper note?

Or do you use Apple? No, on my computer. On my computer. So I just use Google Drive and I just use some electronic notes. Interesting. Google notes. But this is gonna be, yeah, huge notes, right? So if you keep adding, right? Yeah, if I keep adding, yes. Okay. Is that how you learn something new? I mean, besides language? Adding notes to somewhere and review like once a week or so. I'm not sure that's how I learn.

That's, I think, how I process information. And that's how I, often when I write and I rewrite and restructure, that helps me think and helps me clarify my own thinking about things. It's not exactly learning new things, but I think it's helping me just maybe getting deeper insights about some topic. Thank you. So I wanna ask you this question about advice and to someone who started learning something, not only language, but something new,

and what's your advice to them from the learning scientist standpoint and also like a Duolingo master perspective? Well- What can be done better? Yeah. How can we- Definitely, if you're learning, if you're trying to learn a language, I would say the most important thing is to make sure you stay consistent. You actually do it every day or almost every day. That's the most important thing. What exactly you do every day is less important.

As long as you do something very regularly and it doesn't even have to be that much time, after many days, you will actually learn a lot. So that's the most important thing. And then there are, of course it matters exactly what you do, how much time you spend on different things. But if you're not consistent, it's just not, you're not going to learn anything. So I would say that's my main advice. Do you use time spent learning?

Yeah, for your learning as well? Sorry, sorry. Yeah, do you use time spent learning well for your learning as well? Yeah, for my own learning. Yeah, I try to, definitely. Okay, thanks. Yeah, compounding. Oh, yes, compounding. I think we're over time. Oh, yeah. So the last question, sorry. And so about the legacy. So since Glasp is a platform where people share what they're reading, learning, and we want to ask this question,

and what impact or legacy you want to leave behind for future generations? I think what I care about is something I've talked about earlier. I'm really passionate about educating people. I want to make sure people really have opportunities to learn new things. I grew up in a country where education gave me a lot of new opportunities. And so I hope that everyone can have access to those kinds of opportunities

and they can learn and they can really realize their potential. So that's what I'm hoping I'm doing with my work at Duolingo. I'm really helping create those opportunities so that people can succeed regardless of the background they come from, the country they're from. If they have access to good education, they'll be able to do a lot. Beautiful, yeah. And thank you. And thank you so much for joining today. Yeah, thank you.

Thank you very much. This was a lot of fun.

Share This Audio Transcript & Summary 📚

How to Build Duolingo’s Learning Engine with Science and AI | Bozena Pajak

00:00

57:40