Design Notes is a show from Google Design about creative work and what it teaches us. Each episode we'll talk with people from unique creative fields to discover what inspires and unites us in our practice.
Welcome to Design Notes. This is a very special episode that was taped at SPAN 2017 in Pittsburgh. I'm Aaron Lamur. I'll be your guest host for a few conversations from that conference. This one is with Luis von Ahn who is the founder of Duolingo. He grew up in Guatemala. His mother deciding to buy him a Commodore 64 instead of the classic 8-bit Nintendo ended up having a profound impact on his career direction.
Unfortunately, I didn't have very many games and I didn't really have very many people that I knew that also had games kind of to share.
Eventually, his love for games and programming took him to Pittsburgh for graduate school where one of the first things that he created was the CAPTCHA. If you've been on the internet, you've seen a CAPTCHA. It's the slanty text that you have to type in.
It's a test that a computer can grade but it itself cannot pass.
For his next company, he took on gaming as a way to learn, specifically as a way to learn languages with Duolingo.
This is very related to the fact that I'm from Guatemala. I wanted to do something that would have immediate impact on people. If you're in a non-English-speaking country and you know English, usually you can make between 20 and 100% higher salary.
Design Notes is made by Google Design. We have more episodes from SPAN 2017 in Pittsburgh in the feed. Welcome Luis von Ahn.
I did. When I look back at the different projects you've done, I see interests in games and I see interest in human computational power. Going back to your childhood, what were the things that got you into that kind of stuff?
When I was eight, all my friends were getting Nintendos. I wanted a Nintendo. I asked my mom for a Nintendo.
This is the original Nintendo.
The original Nintendo, the NES. Yeah. But my mom did not want to get me a Nintendo and instead she got me this computer. It was at the time it was a Commodore 64 and she said, "This is what you get." And I don't think she realized those things were not very easy to use and I didn't have anybody to learn from. So I, you know, I kind of figured out how to use it, but it took me a while.
Do you think your mom was trying to push you down a specific path with the computer or she just had like an anti-Mario kind of?
No, no, no. I think she really thought, yeah, she probably talked to somebody and they said, "Look, you can play games with a computer, too, but you may learn something." That's probably what happened. But so I figured out how to use it.
But unfortunately, I didn't have very many games and I didn't really have very many people that I knew that also had games kind of to share.
So I started trying to figure out how to make my own. And that's kind of how I started getting into games and programming, etc. Eventually, I convinced my mom to buy me a Nintendo and that happened.
Were Nintendo and computers widespread in Guatemala at the time, or was that a rare thing to see?
Nintendos were widespread. Computers were not, particularly for — I mean, this was in the late 80s — an 8 to 10 year old kid did not have a computer.
When I think back to programming during that period, one of the things that I realize is a bias that you don't notice as an American is that most programming is written in English.
And not just the programming, but the instructional manuals, the way to learn are written in English. So, did you speak English at that point in time?
Yeah, I did. I mean, I was fortunate that I went to an American school, so I spoke — I mean, probably my English was not as good as a native speaker's English, but it was kind of good enough to read that stuff.
What was it like learning this sort of other language that's neither English nor Spanish while you were also a novice English speaker?
I don't think that was the biggest difficulty. I mean, the biggest difficulty was I literally had nobody to learn from. So I sometimes spent hours and hours and hours trying to figure things out. And sometimes it was just there was a typo in the instruction manual or something and that just cost me 10 hours or something. So it was tedious.
And what brought you to Pittsburgh originally from Guatemala?
I decided to come to college to the US. That was not in Pittsburgh. I went to college at Duke in North Carolina. And then I finished my undergrad and I wanted to get a PhD in computer science. And I mainly wanted to get a PhD because I didn't want to get a job and all these people were getting jobs. At the time the big thing was to get a job at Microsoft. All these people were getting jobs at Microsoft or either that or kind of Wall Street. I didn't particularly want to do that. So I decided I wanted to go to grad school and then I applied to a bunch of grad schools. I got into a few and CMU was one of them and it was kind of ranked number one in computer science and then I came and visited and I really liked my PhD adviser — well, I mean the guy who eventually became my PhD adviser. I just liked this one guy and I decided to come to Pittsburgh.
And was it during that period that you started working on CAPTCHA?
Yeah, I was very fortunate. I showed up at CMU. This was in August of the year 2000. And maybe two months later or not even two months later, like a month and a half later, I was lucky to be in the audience. This guy who eventually ended up working for Google but at the time was the chief scientist at Yahoo came and gave a talk at Carnegie Mellon and he said here are 10 problems that we don't know how to solve at Yahoo. And at the time Yahoo was kind of the biggest, you know, internet company. And you know I went home and I tried to think about all 10 problems. Nine of them I had no idea what to do with. And then there was one, the problem was: look, Yahoo was offering free email accounts at the time. It was sort of revolutionary that Yahoo and Hotmail were offering free email accounts. And they had this problem that some people were writing programs to obtain millions of email accounts from Yahoo. The problem was how do we stop them? I went home, I thought about it, then I discussed it with my PhD adviser and together we came up with this idea of a CAPTCHA, which is these distorted characters that you see all over the internet, you know, that you have to type whenever you're getting an account or whatever. We came up with the idea, we implemented the first prototype and within a couple of months it was live on Yahoo, which I now in retrospect realize how fast that was for such a large company and it's because they just had a really huge problem. It's kind of incredible to think of that in 2000.
I feel like everyone you meet at a conference now has some side project that they're trying to turn into something. But the tools to build something like that must have been kind of all hand-done. I mean there's not like APIs and stuff.
Yeah. There was nothing at the time. There was nothing. I mean we had to build even — even writing a program to distort images.
Yeah. I mean today that is so easy because almost every programming language comes with just image manipulation tools. At the time it didn't. So you had to basically get the image, turn it into a matrix and then you apply the actual transformations etc. Like it was all handmade.
Yeah, it was a different time.
I'm assuming that people listening to this are familiar with the CAPTCHA because it would be difficult to have an email account right now if you have not filled out a CAPTCHA. But at its core, you're basically creating a little tiny game that a computer is terrible at and a person is pretty good at.
Yeah. And there's one more thing: a computer should be able to grade it, which is an interesting thing because there's this paradoxical thing that a computer should not be able to pass it. It's a test that a computer can grade but it itself cannot pass.
So, as you were developing it and you're starting to see people use it, how did you approach the idea of a computer grading the results?
I mean, this was part of the thing. A CAPTCHA is a test that can distinguish humans from computers. Turns out that computers are worse than humans at reading these distorted characters, but we couldn't have a human on the other side kind of having to grade these. It had to be graded by a computer because it's done millions of times a day. Very similar to this idea of a Turing test, which was this idea from the 1950s. The idea of a Turing test is a test that can tell humans from computers apart. But in the Turing test, there was a human judge trying to figure out if it was talking to a human or a computer. But in the case of a CAPTCHA, it's a computer judge that gives you a test and it's trying to figure out if it's a human or a computer.
Do you think CAPTCHAs are getting harder now?
Yeah, they are. What's happening ultimately is that computers are getting smarter and it is getting harder and harder to find things that humans are much better at than computers. The thing about a CAPTCHA is it has to be doable within 30 seconds. It should be graded by a computer and it should be that computers cannot do it almost at all whereas humans can very easily do it. It's getting harder and harder to find such things. And the way CAPTCHAs have evolved is basically the distortions have gotten harder and harder and that's happening. And at some point the whole concept of a CAPTCHA won't exist. At some point computers will be able to do most everything that humans can and that will be it.
So when you started Duolingo, what kinds of ideas about this like human-computer interaction did you take with you to Duolingo? Like we kind of broke down what a CAPTCHA is doing. What is Duolingo doing at that very core level?
Well, Duolingo was — it was started about six years ago. The idea was to teach people stuff. I actually didn't know what we wanted to teach. Turns out we ended up teaching languages, but at the time I just wanted to do something that would teach people stuff. I thought that learning on the internet would become a really huge thing. A lot of things would be learned on the internet. I also — this is very related to the fact that I'm from Guatemala — I wanted to do something that would have immediate impact on people. I really love math for example, but teaching math doesn't have immediate impact because you usually learn math in order to learn something else, to eventually become an engineer or something like that. Whereas languages — and this is why we ended up deciding on languages — for many countries in the world, particularly knowledge of English, if you're in a non-English-speaking country and you know English usually you can make between 20 and 100% higher salary just by the fact that you know English. I guess if you're also in an English-speaking country you need to know English.
But basically learning English is something that has immediate monetary impact to improve your life.
Starting with the idea that the language part is almost arbitrary and it's a training regimen. When the big wave came to online education, it was kind of skeuomorphic. It was kind of like how can we put a university in your laptop?
Yeah. Yeah. That was different.
And what you're doing is a little bit more — it reminds me more of like when I was playing sports as a kid like trying to learn basketball and the coach was like every day we're doing layup lines. You got to learn the layup. What kinds of inspirations did you take for the way that you wanted to teach people?
Yeah, in this case it really was games and this is very different. Duolingo started at around the same time as all these companies started that were trying to bring online education to fruition. Companies like Coursera — their approach has been, okay, we're going to take whatever is happening in the real world which is a lecture, we're going to record it, we're going to put it on video and that's online education. We had a very different approach. Our inspiration really at the end was games. We very early on realized that if we wanted to teach something to somebody over the internet, the biggest problem — learning anything by yourself — is staying motivated. It's really quite difficult to stay motivated. It's very similar to going to the gym. Everybody wants to do it.
But when you start showing up, man, it takes a lot of time and it's kind of painful and you don't see results immediately. So, you give up very easily. And this is why you see for example these online courses — the completion rate for these online courses is like 2%. So we knew that would be a problem. So this is why we went and looked into games. Games have this very nice property that a good fraction of the people can spend months if not years playing a single game. So we designed Duolingo to be as close to a game as possible. Of course we wanted to be educational and the results are much better. I mean, of the people that sign up to Duolingo, about a quarter, so about 25% become active users, kind of regularly active users for months. So, that's much better than 2%.
What kinds of longer-term outcomes and effects have you found from people who learn in a game as opposed to say like a college classroom?
So, we know about Duolingo how well it works. It's very good at taking you from zero knowledge of a language to a level that's called independent. So, it's not quite yet fluent. Basically, you will make a lot of mistakes at the end of — if you use Duolingo for a while, you'll make a lot of mistakes, but you'll be able to kind of navigate the world. You may not be able to have a conversation about 18th century philosophy, but you'll probably be able to order food at a restaurant and maybe even go out on a date.
You'll still have a thick accent and sound kind of funny. But our philosophy with Duolingo is that's where we want to get you. Afterwards, if you're really interested, you should probably just practice a lot actually speaking the language. Our philosophy is to get you from zero to this independent level. It's a little different — the outcomes in a classroom. Turns out learning a language, most classrooms are very ineffective. I believe that a very large fraction of the US population was supposed to have taken a foreign language for 3 or 4 years and a very large fraction of them can't say more than like taco. Now, what I think is better than a language classroom and also better than Duolingo is a one-on-one personal tutor. Problem with that is it's very expensive.
