TNW – Luis von Ahn – Harnessing Human Time and Energy for Society | The Next Web


– I wanna start by asking you guys a
question, how many of you have had to fill out a web form where you’ve been asked
to read a distorted sequence of characters like this one? Okay, good. Outstanding. How many of
you have found it really, really annoying? Excellent. Okay, I invented these. That
thing is called the captcha. And the reason it’s there is to make sure that
you, the entity, filling out the form, are actually a human and not some sort of
computer program that was written to submit the form millions of times. And the
reason it works is because humans, at least non-visually impaired humans, have
no trouble reading these distorted characters. Whereas, computer programs
can’t do it as well yet. So for example, in the case of Ticketmaster, the reason
you have to type these captchas is to prevent ticket scalpers from writing a
program that can buy millions of tickets two at a time. Now, captchas are used all
over the Internet. It turns out that approximately 200 millions of these are
typed everyday by people around the world. Now, when I first heard this number, I
was quite proud of myself. I thought, “Look at the impact that my work has had.”
But then I started feeling bad. See, here’s the thing. Not only do people find
them annoying, but also each time you type a captcha, you basically waste 10 seconds
of your time. And if you multiply 10 seconds by 200 million, you get that
humanity, as a whole, is wasting about 500,000 hours everyday typing these
annoying catches because of me. So I started feeling bad. And then I started
thinking, “Is there any way in which we can make good use of this time?” See, the
thing is, during those 10 seconds while you’re typing a captcha, your brain is
doing something amazing. Your brain is doing something that computers cannot yet
do. So the question is, “Can we get you to do something useful?” And the answer to
that was yes. And this is what we are doing now with captchas. What you may not
know is that… So nowadays with captcha, while you’re typing a captcha not only are
you authenticating yourself as a human, but you’re also helping us to digitize
books. So let me explain how that works. So there’s a lot of projects out there
trying to digitize books. Google, for example, has one. The Internet Archive has
one. Where the basic idea is you’re taking books and putting them online.
So the way these projects work is that you start with a physical book thing. You’ve
seen those things like a book. You start with one of these and then you scan it.
Now, scanning a book is like taking a digital photograph of every page of the
book. It gives you an image for every page of the book. The next step in the process
is that the computer needs to be able to decipher all of the words in these images.
That’s done using a technology called O.C.R. or optical character recognition.
The problem with O.C.R. is that it’s not perfect. Especially for older books
where the ink has faded, the computer cannot recognize a lot of the words. So
for things that were written more than 20, 30 years ago, the computer cannot
recognize about 20 to 30% of the actual words, but humans can. So what we’re doing
now is we’re taking all of the words that the computer cannot recognize in the book
digitization process and we’re getting people to read them for us while they type
captchas on the Internet. Okay, so these words that you’re typing, they’re actually
words that are coming directly from books that the computer cannot recognize. And
we’re using what people are typing all over the world to help us digitize books.
Now, the reason they are two words instead of one nowadays is that, you see, for one
of the words is a word for which the system already knows the answer, and then
the other word is a word that the system just got out of a book. It does not know
what it is because the computer cannot figure out what it was and it’s gonna give
both words to the user and it’s not gonna say which one is which. It’s just gonna
say, “Please type both.” And if the user types… For the one for which the system
knows the answer, if the user types that word correctly, then the system assumes
that the user is a human and it also gets some confidence that they also type the
other word correctly. And if you repeat this process about 10 different times with
10 different people and they all agree on what the new word is, then one word
becomes digitized with very high accuracy. Okay, this is the basic idea of how this
works. It turns out it’s working very well. So we’re digitizing approximately
100 million words a day using this, which is the equivalent of about two million
books a year that are all being corrected one word at a time by people typing
captchas on the Internet. Now, since we’re doing so many words,
funny things can happen. And this is especially true because, see, now we’re
giving people two randomly chosen English words right next to each other. And so
funny things can happen. So for example, we showed… So this is called a captcha.
So we showed this word, the word “Christians.” There’s nothing wrong with
it, but if we show it along with another randomly chosen word right next to it,
then funny things can happen. So for example, we showed this, that’s kind of
bad. But it’s even worse because the particular website where this was
presented… So there’s millions of websites using the system. The particular
website where this was presented happened to be called “The Embassy of the Kingdom
of God.” Oops. Here’s another really bad one. It’s a U.S. politician,
JohnEdwards.com. So we keep on insulting people left and right everyday. And
there’s not much we can do because, you see, for one of the words, we don’t know
what it is so there’s not much filtering we can do. So it’s funny. Okay, now this
is my favorite number of this project. It’s 1.1 billion. This is the total number
of distinct people that have ever helped us digitize at least one word out of a
book through captchas. Okay, so this is about 15% of the world’s population has
helped us digitize books through these captchas. And this is the number that
inspires my work. I mean, so basically, what I try to do in my work is I try to
coordinate the efforts of millions of people to do something that is good, okay?
Now, another way of seeing this project is it’s kind of like recycling human
processing power. Okay, we took something that people anyways did, which is typing
catches, and then we try to reuse it for something else, which is to digitize
books, okay? So this is kind of like recycling human or mental processing
power, okay? And this idea is something that I use in many of the projects that I
worked on and I’m gonna talk about another project that is related to this idea. Now
many people have asked me, you know, where did this idea come from about recycling
human processing power. I’m not the only one who’s had it. But in my particular
case, the idea came from an idea that I had when I was a kid, which… I was very
obsessed with this idea. I thought it was an amazing genius idea.
It turns out… First of all, a lot of people have had this idea, and second, it
wasn’t that great. But the idea that I had when I was a kid was this one. So
everybody knows, if you, you know, whenever you exercise, for example, use a
stationary bicycle or something, you could be generating electricity or you could be
transforming the kinetic energy that you’re using when you’re exercising into
electricity. You could be doing that. So the idea that I had is I wanted to make a
gym. Actually, no, I was more ambitious than that. I wanted to make a chain of
gyms, where the idea is that it was free to go to the gym. The gym was completely
free, but all the exercise equipment was connected to the power grid and the gym
would sell power to the power grid all generated by the people who went to
exercise. So that was the idea that I wanted to do. I really liked it. And it
turns out, this is not a very good idea because it turns out that people are not
very good at generating power, so it’s better to just charge them. But this idea
is relatively similar to the idea of digitizing books with captchas. I mean,
we’re taking activity that anyways people do and we’re reusing the energy for
something else. The big difference is that this is actual, you know, electrical
energy versus, in the case of captchas, it’s mental processing energy. Okay, so
let me… With that in mind, let me talk about this new project that I’ve started
working on. It’s a company called Duolingo. It’s worth mentioning how the
project started. So it started about four years ago. I was in a very fortunate
position in my life. I has just sold a second company to Google. It was this book
digitization captcha company to Google and I didn’t have to work anymore. So I
decided to retire and it lasted for about a week then I got bored. Then I started
thinking, “What is it that I really wanna do with the rest of my life? What is it
that I really wanna do?” And I’ve always had a passion for education. I’m also
a professor in computer science so I always had a passion for education.
So what I knew I wanted to do is I wanted to work on something related to education.
Now, my views on education were or have been very shaped by where I am from. So I
am from Guatemala. This is a public service announcement, by the way. I am
from Guatemala. That’s where it is. And by the way, this is not the same as where
they keep the prisoners. That’s Guantanamo. That’s not where it is. Now,
if you don’t know about Guatemala, Guatemala is a very poor country. And a
lot of people talk about education as being something that brings equality to
different social classes, but I always saw it differently. I always saw education as
something that brings inequality to different social classes. Because what
happens, and particularly in Guatemala, but in many other similar countries is
that the people who are wealthy can pay for really good educations for themselves.
And in fact, they can, you know, end up going to Harvard because they just have
enough money that can pay their way through the whole thing and they can get
really good education. And because they have such a good education, they remain
wealthy. Whereas the people who don’t have very much money can barely learn how to
read and write. And because they can barely learn how to read and write, they
remain not having very much money. So I always saw this is a huge divider. So what
I wanted to do is I wanted to do something related to education that would give the
best access to education to everybody, regardless of whether you had money or
not. This is kind of what I wanted to do. Now, I decided to start with one type of
education that is something that is pretty big everywhere in the world and it is
learning another language. This is pretty big everywhere in the world except for the
United States, it turns out. So there is about 1.2 billion people in the world
learning a foreign language. It’s the total number of people. Now, this is a
funny market. It’s strange because it has the following property. The majority of
these people, about 800 million of these people, have satisfied three properties.
First of all, they’re learning English. 800 million of them are learning English.
Second, the main reason they are learning English is to get a job or to get a better
job. Basically, it’s for job purposes. And third, they are of low socioeconomic
classes. Okay, so basically the majority of people who were learning another
language are relatively poor people who are trying to learn a language so that
they can get a better job. Okay, that’s the majority of people learning a
language. At the same time, usually the ways there are to learn a language require
people to have quite a bit of money, especially with software. So for example,
in the U.S., there is a big language learning company called Rosetta Stone
where they sell their product for about between $500 and $1000, which is very
expensive. So it’s a funny thing. It’s this ironic thing that most of the
solutions there are to learn a language are targeting the minority of the people
who were trying to learn a language. The majority just cannot afford, you know,
even a $50 solution. So what I wanted to do was I wanted to make a way to learn a
language that was just completely free. That’s what I wanted to do. But of course,
at the same time, you have to figure out a way to make money off of it, to kind of
finance it. So the question became when I started working on Duolingo was, “How do
we make a completely free way to learn a language, but at the same time, finance
it?” And then this idea of recycling human processing power came back again. So the
question was, “How can we take all of the people who are learning a language and
somehow use their mental processing power for something else that we can monetize
much like with the gym?” And it turns out we can do this. And the key idea here is
that if you already learned the language, many of the exercises that you have to do
while you’re learning a language are about translation because, you know, imagine
you’re learning some other language, they get you to translate little songs or
something like that. Well, if we got you to translate something that has never been
translated before while you’re learning a language, then we could make money off of
that. And so the way Duolingo ended up working or the way it works is as follows.
So we basically teach you a language. Now, after we teach you a concept, so for
example, the concept would be food words. We’ve taught you everything about food,
how to order food in a restaurant, etcetera, after we taught you a concept,
we say, “Hey, if you want to practice what you just learned about food words with
something from the real world, then here’s a document that is related to food.” It
could be like a menu or something that you can help us translate in order to
practice. And those translations are the things that
we sell. So for example, C.N.N. is one of our clients, where the idea is that C.N.N.
writes all of their news in English. Then they send us their news articles in
English and then we give them to our users. In this particular case, it is
Spanish-speaking users who are learning English. We give them the articles. And to
practice, what they do is they translate these English articles into Spanish. And
then once many of our users have translated the same article, we send it
back to C.N.N. and then they pay us for the translation. So this is how the basic
idea works. In some sense, C.N.N. is sponsoring free language education. So
that’s the basic idea of Duolingo. So what it is, with Duolingo, you can learn
another language, while simultaneously translating the web. So let me play you
this one and a half minute video that explains the concept hopefully a little
better than me. So hopefully this works. – [recording] It’s a big world out there.
Billions of us are trying to live, love, prosper, and make sense of our brief
time on this planet. Since the dawn of humanity, we’ve been passing information from one person
to another through a common language. Unfortunately, you can’t communicate with
others without knowing or learning their language first. A similar issue has
manifested on the web where text can be penned in dozens of languages, each of
which demands a reader’s fluency. We have developed an elegant solution to both
problems, a way for you to learn a language for free, while at the same time,
helping to translate text from the web, enabling a wealth of language shackled
information to be liberated for all of humanity. It’s called Duolingo. Here’s how
it works. Let’s say you’re a native English speaker who wants to learn
Spanish. We start by giving you a sentence from a Spanish website and asking you to
translate it. Wait, back up, how can you translate a
language you don’t know? First, Duolingo only gives you sentences that fit your
language level. Beginners get the really simple sentences from the web and advanced
users get the more complex ones. This way, everybody becomes a valuable translator.
And second, if you’re really lost, you can always see possible translations for words
you don’t know. Afterwards, Duolingo helps you understand and analyze the words you
hovered over through educational examples. You can also vote on the quality of other
student’s translations, which helps you learn by seeing how others translated the
same sentence. And because you create valuable translations while you learn, we
return the favor by offering Duolingo completely free of charge, no ads, no
hidden fees, no subscriptions, just free. To put the potential benefit of Duolingo
into perspective, think about this. If one million people would use Duolingo to
learn, the entirety of English Wikipedia could be translated to Spanish in just 80
hours. Duolingo. Learn a language while translating the web. – Okay, so this is the video that we
launched with… Thank you. Thank you. So this is the video that we launched with.
When we launched, we had no idea if this was going to work. We put it out there.
Our biggest question was whether people were actually going to use our system to
learn a language. We really had no idea this was going to happen. But fortunately,
it’s been very well adopted. So one of the first few things that happened right after
we launched this, within about a month of launching, actually PC Magazine gave us
our Editor’s Choice for Language Learning. They gave us the same rating as Rosetta
Stone, which is $500 piece of software, except ours is totally free. Now to me,
the most amazing thing about this at the time was that PC Magazine still exists.
I did not realize that. Now, so in terms of adoption, the traffic has been really
great. This is a bit outdated actually. So by
now, we have about 12 and a half million active users learning a language in
Duolingo. In fact, there are now more people learning a language in Duolingo
than in the entire U.S. public school system. So all the people in all middle
schools and high schools in the U.S., there are less people learning a
language there than in Duolingo. In fact, by now, Duolingo is the most popular way
to learn languages on line. So we have a website. We also have apps. It’s also on
the iPhone and on Android. It is the top education app in the world in Android and
in iOS also. This is of any kind of education. And also, Apple just chose
Duolingo as of maybe four months ago as the iPhone App of the Year because
they like it. This one is particularly good for me. This is the search volume on
Google between the query Rosetta Stone, which was the leading language learning
software up until a few months ago. That’s yellow. And then the red is Duolingo,
which we’ve surpassed now. What’s amazing to me about this is Rosetta Stone spends
about $150 million a year on advertising. Whereas with Duolingo, we have spent zero on advertising. It’s purely through word
of month. And of course, the things that people say are very positive. So for
example, this person said, “In the past two days I’ve learned more from Duolingo
than Rosetta Stone in over a month.” Or this other person, who was my mother. Now,
one of the reasons people like Duolingo so much and one of the reasons it’s growing
so much is because we realized something. You know, when we were starting, we
realized that one of the hardest things about learning a language by yourself or
really learning anything by yourself is keeping yourself motivated. It’s just very
hard to keep yourself motivated because learning anything that’s worthwhile
learning requires a lot of effort so it’s hard to keep yourself motivated. So what
we did is we actually made Duolingo really turned into something that feels like
you’re playing a game while you’re learning a language. So for example, this
is actually the old version of Duolingo. It’s just looks different now, but it
works very similarly. Where the idea is that learning a language
in Duolingo, we split it into different, we call them skills. So these are
different units. So for example, food is a skill or plurals is a skill. The idea is
that you learn a language by learning each one of these skills. At the beginning,
there’s only one skill, which is called Basics. And then when you finish that
skill, you unlock new skills. And basically, the whole idea is that you’re
unlocking more and more skills, it really feels like a game. Now, we also have
this… for each skill, we have this pretty sophisticated kind of meter that
measures how well you know the skill. This is the strength bar for each skill. The
idea of this is that every time you do something correctly relating to that
skill, your strength goes up. And every time you do something incorrectly related
to that skill, your strength goes down. And also, if you don’t use that skill or
you don’t use Duolingo for a while, your strength just naturally goes down. This is
to model the fact that you’re forgetting things. Now, this is actually a thing that
keeps people coming back quite a bit because they want to keep all of their
skills golden. So this is something that gets people to come back quite a bit. Now,
in case you don’t come back, we also have all kinds of psychological tricks to get
you to come back. So for example, we send you a message with this guy. This is our
mascot. Actually, this is our mascot before we redesigned it. But we send you
this guy. It turns out we tried a lot of things to see what got you to come back
more, and it turns out guilt really helped. We, in fact, even A/B tested the
amount of tears at the bottom that there would be and this guy was a winner to get
you come back. Now, while you’re actually learning a skill… This is what it looks
like while you’re learning a skill. So basically, we teach you all the different
words and things. So we teach you, for example in this case, we teach you the
word for boy with pictures of boy. You also can speak to the app to practice your
pronunciation. So this is basically how you learn. We give you tiny exercises that
we call Challenges. So in each skill, you have to get all
these challenges and you have to get them correctly. And if you ever answer
something wrong, you lose a life. So there, you have three lives. And for each
lesson, if you lose all lives, you have to start over. Okay, so this is quite
gamified. We got these hearts from the game Zelda, of course. So that’s the basic
idea of how it works. Now, if you’re able to pass the whole lesson without losing
all your lives, then you get a bunch of points. And then after that, that is the
point on the website. So we have website and apps. On the website, if you pass a
lesson, that is the point where we tell you, “Hey, you just learned about food or
about something. Do you wanna help us translate something that is related to
what you just learned?” And on the website… on the apps, there’s no
translation yet. But on the website, that’s where you can go and help us
translate something. And through these translations, is how we monetize
Duolingo. Now in fact, the translations are one of the reasons why people like
Duolingo, at least the web version, is because they can be learning with real
world content. So for example, BuzzFeed and C.N.N. are some of our clients. Now,
imagine learning English by reading BuzzFeed articles. You would learn a very
funny English, but it’s super fun to do that. And this is one of the reasons why
people like Duolingo so much. Now, when we started working on Duolingo, we actually
didn’t know anything how to teach languages. Neither me nor my co-founder
knew anything about teaching languages. We’re both computer scientists. So we
figured we probably should learn something about how to teach languages if we’re
gonna work on a language learning website. So what we did is we read a bunch of books
on about to teach languages and then we also went and talked to experts that are
experts on how to learn languages throughout the Untied States. We went and
talked to a few of them. And then what we quickly discovered after talking to a lot
of them is that all the stuff there is about how to best learn a language, we
started really feeling like it was similar to all the stuff about diets.
There’s hundreds of different diets, each of which contradict each other. So
sometimes, you know, today it’s okay to eat eggs. Tomorrow, no, that’s not okay.
You should never eat eggs. Today it’s okay to drink milk. Tomorrow, milk is the
unhealthiest thing in the world. And all these diets contradict each other. It was
very similar, the stuff that we’re seeing about how to learn a language. I mean,
some people say, “You should never use your native language. You should never
forget about your native language” Or some other people say, “No, I forget about the
native language” So they’re all these contradictory theories. And so we didn’t
know exactly what to do. So what we did is we kind of took what made sense for us and
we started with an original curriculum on how to teach a language with Duolingo. But
then, we found ourselves in a very fortunate position because what happened
is we started… so we started with an original curriculum for Duolingo, but then
we realized that we had so many users that we could start figuring out the answers to
a lot of these questions that we had at the beginning that nobody could answer
people get conflicting answers for. So for example, what we do now, and this is… we
do this continuously. So what we do now is we use all of our users, all of our 12 and
a half million active users, as an experiment. We’re just trying to figure
out, with these 12 and a half million users, what is the best way to teach a
language. And the types of things that we do are things like, for example, we wanna
figure out how early we should teach you plurals. What we do is that for the next
50,000 people that sign up, it takes about 12 hours for 50,000 people to sign
up to Duolingo because we get about 100,000 users a day. So for the next
50,000 people that sign up, for half of them, we’re gonna teach them plurals a
little later than the other half and then we’re gonna track them for the next days
or weeks or months to see which of these two conditions does better. So do the
people who learn plurals earlier, do they stick around with the site more? Do
they use the site more often? Do they learn better? And after we track them
for a few days or a few weeks, we can usually figure out which of these two is
better. And once we figure that out, we start teaching it that way to all our
users. So this is what we’re doing to try to improve language education. And we’ve
been able to improve it significantly since we started by just doing this type
of A/B testing on language education. And it’s because of this that… so
somebody did an external study to try to figure out how well you can learn with
Duolingo. And what they found is that… So this is the number. If you use Duolingo
for 34 hours, you learn the equivalent of one semester of college in the language,
which usually takes a lot longer than 34 hours. In some sense, learning on Duolingo
is extremely efficient. And the reason we’re able to do this is because we can
actually learn from our users at a very fast pace. If you think about it, in the
offline world, being able to this improvement of your curriculum, the cycle
of improvement takes years for each improvement. Whereas here, we can reduce
that cycle of improvement to a few days or a few weeks. So it allows us to go faster
as we’re improving to have it teached. Now, let me just talk about a couple of
quick things. Actually, I’m running out of time so let me talk about one thing. This
is the one that I’m most excited about, which is something that we’re going to
launch in about three weeks, and it’s this. So, a lot of the people that contact
us about Duolingo say the following. They say, “Okay, thank you. You’ve taught me a
language or you’re helping me learn a language. I wasn’t able to afford language
education before. Thank you. But now I have the following problem. I’ve learned
English with you guys, but I need to prove that I know English because my job or the
job that I’m applying for requires that I have an English certification.”
Unfortunately, the English certification is very expensive, so I’m now stuck. So,
you know, we got enough of these e-mails that we started looking into providing an
English certification to people. And what we found is this crazy thing. First of
all, we started looking at the certification market. If you ever have
to deal with this, it’s insane. There are about 30 million people in the world every
year that have to go through the hassle of taking an English certification exam.
Where the basic idea is these are standardized tests that you have to take.
There’s a few of them throughout the world. There’s maybe three or four of
them that are widely accepted throughout the world. They’re kind of monopolies in
their own areas. It costs about $300 to take a standardized test to prove that you
know English. By the way, the reason you take this is because many jobs,
international corporations require them and also universities or schools require
them, sometimes for graduation requirement. Okay, you have to pay about
$300 to take one of these. You also have to make appointment weeks in advance to go
to a testing center, to go take the test. A big problem of this is the testing
centers are only in the big cities throughout the world. So if you don’t live
in a big city, you have to travel to the testing center, pay $300, which by the
way, for many of the people taking this, this is a month’s salary. So $300, in many
places like my country, it’s a month’s salary. So you have to pay a month’s
salary to take a test and travel somewhere. So it’s this crazy thing. So
what we wanna do is we wanna have a language certification test that is not
$300 because it does not cost $300 to administer a simple test. And also, one
that you don’t have to travel for it. You can just take it from your own home and
your own mobile device. So what we’re doing is we’re actually launching this
thing in about three weeks called the Duolingo Test Center, which is an app
where you’re gonna be able to take a standardized test from your app. And now,
the hard part about this, what’s hard about this, the reason there are these
standardized testing centers is because you have to prevent cheating, you have to
make sure that the person who is taking the test is actually the right person, and
also that there’s nobody else helping them. Now, it’s hard if you’re taking a
test with your mobile device to make sure that you’re not cheating. So what we’re
doing is we’re actually… The way we’re gonna do it is whenever you start taking
the test, we’re actually gonna turn on the camera and we’re also gonna turn on the
audio to record everything that there is and we’re gonna have all kinds of tricks
about it. When you start taking the test, we’re gonna tell you, “Please show us your
whole room.” Also, this is funny. You have to take the test while you’re
standing in a corner so that there’s nobody behind you.
And by the way, when the camera is on, we’re actually gonna have a real person
looking at whether you’re doing all the right things. So one of the things we tell
you is, “Please stand in a corner so that there’s nobody behind helping you.” And
then there’s a person watching you take the whole test to make sure that you’re
not cheating. So we’re gonna do that. Unfortunately, the test is not gonna be
free because, well, we have to pay this person to actually watch you, but the test
is going to be $20 as opposed to $300 and you’re gonna be able to take it from a
phone. So I’m pretty excited about that. But that’s it. That’s all I wanted to say.
Thank you. [applause]

2 thoughts on “TNW – Luis von Ahn – Harnessing Human Time and Energy for Society | The Next Web

  1. Awesome.  Que inspirante, este filosofia nueva de negocios y el poder de la technologia.  (If my spanish is bad, I learned it in college 😉  I would just suggest one "minor" tweak to the business model.  Since people are providing a service that is being monetized, wouldn't it be great to let them earn credits toward a FREE final exam?   Entonces será perfecto!

Leave a Reply

Your email address will not be published. Required fields are marked *