TEDxCMU
Luis von Ahn: Massive-scale online collaboration
Filmed:
Readability: 3.8
1,740,008 views
After re-purposing CAPTCHA so each human-typed response helps digitize books, Luis von Ahn wondered how else to use small contributions by many on the Internet for greater good. In this talk, he shares how his ambitious new project, Duolingo, will help millions learn a new language while translating the web quickly and accurately -- all for free.
Luis von Ahn - Computer scientist
Luis von Ahn builds systems that combine humans and computers to solve large-scale problems that neither can solve alone. Full bio
Luis von Ahn builds systems that combine humans and computers to solve large-scale problems that neither can solve alone. Full bio
Double-click the English transcript below to play the video.
00:15
How many of you had to fill out some sort of web form
0
0
2000
00:17
where you've been asked to read a distorted sequence of characters like this?
1
2000
2000
00:19
How many of you found it really, really annoying?
2
4000
2000
00:21
Okay, outstanding. So I invented that.
3
6000
3000
00:24
(Laughter)
4
9000
2000
00:26
Or I was one of the people who did it.
5
11000
2000
00:28
That thing is called a CAPTCHA.
6
13000
2000
00:30
And the reason it is there is to make sure you, the entity filling out the form,
7
15000
2000
00:32
are actually a human and not some sort of computer program
8
17000
3000
00:35
that was written to submit the form millions and millions of times.
9
20000
2000
00:37
The reason it works is because humans,
10
22000
2000
00:39
at least non-visually-impaired humans,
11
24000
2000
00:41
have no trouble reading these distorted squiggly characters,
12
26000
2000
00:43
whereas computer programs simply can't do it as well yet.
13
28000
3000
00:46
So for example, in the case of Ticketmaster,
14
31000
2000
00:48
the reason you have to type these distorted characters
15
33000
2000
00:50
is to prevent scalpers from writing a program
16
35000
2000
00:52
that can buy millions of tickets, two at a time.
17
37000
2000
00:54
CAPTCHAs are used all over the Internet.
18
39000
2000
00:56
And since they're used so often,
19
41000
2000
00:58
a lot of times the precise sequence of random characters that is shown to the user
20
43000
2000
01:00
is not so fortunate.
21
45000
2000
01:02
So this is an example from the Yahoo registration page.
22
47000
3000
01:05
The random characters that happened to be shown to the user
23
50000
2000
01:07
were W, A, I, T, which, of course, spell a word.
24
52000
3000
01:10
But the best part is the message
25
55000
3000
01:13
that the Yahoo help desk got about 20 minutes later.
26
58000
3000
01:16
Text: "Help! I've been waiting for over 20 minutes, and nothing happens."
27
61000
3000
01:19
(Laughter)
28
64000
4000
01:23
This person thought they needed to wait.
29
68000
2000
01:25
This of course, is not as bad as this poor person.
30
70000
3000
01:28
(Laughter)
31
73000
2000
01:30
CAPTCHA Project is something that we did here at Carnegie Melllon over 10 years ago,
32
75000
3000
01:33
and it's been used everywhere.
33
78000
2000
01:35
Let me now tell you about a project that we did a few years later,
34
80000
2000
01:37
which is sort of the next evolution of CAPTCHA.
35
82000
3000
01:40
This is a project that we call reCAPTCHA,
36
85000
2000
01:42
which is something that we started here at Carnegie Mellon,
37
87000
2000
01:44
then we turned it into a startup company.
38
89000
2000
01:46
And then about a year and a half ago,
39
91000
2000
01:48
Google actually acquired this company.
40
93000
2000
01:50
So let me tell you what this project started.
41
95000
2000
01:52
So this project started from the following realization:
42
97000
3000
01:55
It turns out that approximately 200 million CAPTCHAs
43
100000
2000
01:57
are typed everyday by people around the world.
44
102000
3000
02:00
When I first heard this, I was quite proud of myself.
45
105000
2000
02:02
I thought, look at the impact that my research has had.
46
107000
2000
02:04
But then I started feeling bad.
47
109000
2000
02:06
See here's the thing, each time you type a CAPTCHA,
48
111000
2000
02:08
essentially you waste 10 seconds of your time.
49
113000
3000
02:11
And if you multiply that by 200 million,
50
116000
2000
02:13
you get that humanity as a whole is wasting about 500,000 hours every day
51
118000
3000
02:16
typing these annoying CAPTCHAs.
52
121000
2000
02:18
So then I started feeling bad.
53
123000
2000
02:20
(Laughter)
54
125000
2000
02:22
And then I started thinking, well, of course, we can't just get rid of CAPTCHAs,
55
127000
3000
02:25
because the security of the Web sort of depends on them.
56
130000
2000
02:27
But then I started thinking, is there any way we can use this effort
57
132000
3000
02:30
for something that is good for humanity?
58
135000
2000
02:32
So see, here's the thing.
59
137000
2000
02:34
While you're typing a CAPTCHA, during those 10 seconds,
60
139000
2000
02:36
your brain is doing something amazing.
61
141000
2000
02:38
Your brain is doing something that computers cannot yet do.
62
143000
2000
02:40
So can we get you to do useful work for those 10 seconds?
63
145000
3000
02:43
Another way of putting it is,
64
148000
2000
02:45
is there some humongous problem that we cannot yet get computers to solve,
65
150000
2000
02:47
yet we can split into tiny 10-second chunks
66
152000
3000
02:50
such that each time somebody solves a CAPTCHA
67
155000
2000
02:52
they solve a little bit of this problem?
68
157000
2000
02:54
And the answer to that is "yes," and this is what we're doing now.
69
159000
2000
02:56
So what you may not know is that nowadays while you're typing a CAPTCHA,
70
161000
3000
02:59
not only are you authenticating yourself as a human,
71
164000
2000
03:01
but in addition you're actually helping us to digitize books.
72
166000
2000
03:03
So let me explain how this works.
73
168000
2000
03:05
So there's a lot of projects out there trying to digitize books.
74
170000
2000
03:07
Google has one. The Internet Archive has one.
75
172000
3000
03:10
Amazon, now with the Kindle, is trying to digitize books.
76
175000
2000
03:12
Basically the way this works
77
177000
2000
03:14
is you start with an old book.
78
179000
2000
03:16
You've seen those things, right? Like a book?
79
181000
2000
03:18
(Laughter)
80
183000
2000
03:20
So you start with a book, and then you scan it.
81
185000
2000
03:22
Now scanning a book
82
187000
2000
03:24
is like taking a digital photograph of every page of the book.
83
189000
2000
03:26
It gives you an image for every page of the book.
84
191000
2000
03:28
This is an image with text for every page of the book.
85
193000
2000
03:30
The next step in the process
86
195000
2000
03:32
is that the computer needs to be able to decipher all of the words in this image.
87
197000
3000
03:35
That's using a technology called OCR,
88
200000
2000
03:37
for optical character recognition,
89
202000
2000
03:39
which takes a picture of text
90
204000
2000
03:41
and tries to figure out what text is in there.
91
206000
2000
03:43
Now the problem is that OCR is not perfect.
92
208000
2000
03:45
Especially for older books
93
210000
2000
03:47
where the ink has faded and the pages have turned yellow,
94
212000
3000
03:50
OCR cannot recognize a lot of the words.
95
215000
2000
03:52
For example, for things that were written more than 50 years ago,
96
217000
2000
03:54
the computer cannot recognize about 30 percent of the words.
97
219000
3000
03:57
So what we're doing now
98
222000
2000
03:59
is we're taking all of the words that the computer cannot recognize
99
224000
2000
04:01
and we're getting people to read them for us
100
226000
2000
04:03
while they're typing a CAPTCHA on the Internet.
101
228000
2000
04:05
So the next time you type a CAPTCHA, these words that you're typing
102
230000
3000
04:08
are actually words that are coming from books that are being digitized
103
233000
3000
04:11
that the computer could not recognize.
104
236000
2000
04:13
And now the reason we have two words nowadays instead of one
105
238000
2000
04:15
is because, you see, one of the words
106
240000
2000
04:17
is a word that the system just got out of a book,
107
242000
2000
04:19
it didn't know what it was, and it's going to present it to you.
108
244000
3000
04:22
But since it doesn't know the answer for it, it cannot grade it for you.
109
247000
3000
04:25
So what we do is we give you another word,
110
250000
2000
04:27
one for which the system does know the answer.
111
252000
2000
04:29
We don't tell you which one's which, and we say, please type both.
112
254000
2000
04:31
And if you type the correct word
113
256000
2000
04:33
for the one for which the system already knows the answer,
114
258000
2000
04:35
it assumes you are human,
115
260000
2000
04:37
and it also gets some confidence that you typed the other word correctly.
116
262000
2000
04:39
And if we repeat this process to like 10 different people
117
264000
3000
04:42
and all of them agree on what the new word is,
118
267000
2000
04:44
then we get one more word digitized accurately.
119
269000
2000
04:46
So this is how the system works.
120
271000
2000
04:48
And basically, since we released it about three or four years ago,
121
273000
3000
04:51
a lot of websites have started switching
122
276000
2000
04:53
from the old CAPTCHA where people wasted their time
123
278000
2000
04:55
to the new CAPTCHA where people are helping to digitize books.
124
280000
2000
04:57
So for example, Ticketmaster.
125
282000
2000
04:59
So every time you buy tickets on Ticketmaster, you help to digitize a book.
126
284000
3000
05:02
Facebook: Every time you add a friend or poke somebody,
127
287000
2000
05:04
you help to digitize a book.
128
289000
2000
05:06
Twitter and about 350,000 other sites are all using reCAPTCHA.
129
291000
3000
05:09
And in fact, the number of sites that are using reCAPTCHA is so high
130
294000
2000
05:11
that the number of words that we're digitizing per day is really, really large.
131
296000
3000
05:14
It's about 100 million a day,
132
299000
2000
05:16
which is the equivalent of about two and a half million books a year.
133
301000
4000
05:20
And this is all being done one word at a time
134
305000
2000
05:22
by just people typing CAPTCHAs on the Internet.
135
307000
2000
05:24
(Applause)
136
309000
8000
05:32
Now of course,
137
317000
2000
05:34
since we're doing so many words per day,
138
319000
2000
05:36
funny things can happen.
139
321000
2000
05:38
And this is especially true because now we're giving people
140
323000
2000
05:40
two randomly chosen English words next to each other.
141
325000
2000
05:42
So funny things can happen.
142
327000
2000
05:44
For example, we presented this word.
143
329000
2000
05:46
It's the word "Christians"; there's nothing wrong with it.
144
331000
2000
05:48
But if you present it along with another randomly chosen word,
145
333000
3000
05:51
bad things can happen.
146
336000
2000
05:53
So we get this. (Text: bad christians)
147
338000
2000
05:55
But it's even worse, because the particular website where we showed this
148
340000
3000
05:58
actually happened to be called The Embassy of the Kingdom of God.
149
343000
3000
06:01
(Laughter)
150
346000
2000
06:03
Oops.
151
348000
2000
06:05
(Laughter)
152
350000
3000
06:08
Here's another really bad one.
153
353000
2000
06:10
JohnEdwards.com
154
355000
2000
06:12
(Text: Damn liberal)
155
357000
3000
06:15
(Laughter)
156
360000
2000
06:17
So we keep on insulting people left and right everyday.
157
362000
3000
06:20
Now, of course, we're not just insulting people.
158
365000
2000
06:22
See here's the thing, since we're presenting two randomly chosen words,
159
367000
3000
06:25
interesting things can happen.
160
370000
2000
06:27
So this actually has given rise
161
372000
2000
06:29
to a really big Internet meme
162
374000
3000
06:32
that tens of thousands of people have participated in,
163
377000
2000
06:34
which is called CAPTCHA art.
164
379000
2000
06:36
I'm sure some of you have heard about it.
165
381000
2000
06:38
Here's how it works.
166
383000
2000
06:40
Imagine you're using the Internet and you see a CAPTCHA
167
385000
2000
06:42
that you think is somewhat peculiar,
168
387000
2000
06:44
like this CAPTCHA. (Text: invisible toaster)
169
389000
2000
06:46
Then what you're supposed to do is you take a screen shot of it.
170
391000
2000
06:48
Then of course, you fill out the CAPTCHA
171
393000
2000
06:50
because you help us digitize a book.
172
395000
2000
06:52
But then, first you take a screen shot,
173
397000
2000
06:54
and then you draw something that is related to it.
174
399000
2000
06:56
(Laughter)
175
401000
2000
06:58
That's how it works.
176
403000
3000
07:01
There are tens of thousands of these.
177
406000
3000
07:04
Some of them are very cute. (Text: clenched it)
178
409000
2000
07:06
(Laughter)
179
411000
2000
07:08
Some of them are funnier.
180
413000
2000
07:10
(Text: stoned founders)
181
415000
3000
07:13
(Laughter)
182
418000
3000
07:16
And some of them,
183
421000
2000
07:18
like paleontological shvisle,
184
423000
3000
07:21
they contain Snoop Dogg.
185
426000
2000
07:23
(Laughter)
186
428000
3000
07:26
Okay, so this is my favorite number of reCAPTCHA.
187
431000
2000
07:28
So this is the favorite thing that I like about this whole project.
188
433000
3000
07:31
This is the number of distinct people
189
436000
2000
07:33
that have helped us digitize at least one word out of a book through reCAPTCHA:
190
438000
3000
07:36
750 million,
191
441000
2000
07:38
which is a little over 10 percent of the world's population,
192
443000
2000
07:40
has helped us digitize human knowledge.
193
445000
2000
07:42
And it is numbers like these that motivate my research agenda.
194
447000
3000
07:45
So the question that motivates my research is the following:
195
450000
3000
07:48
If you look at humanity's large-scale achievements,
196
453000
2000
07:50
these really big things
197
455000
2000
07:52
that humanity has gotten together and done historically --
198
457000
3000
07:55
like for example, building the pyramids of Egypt
199
460000
2000
07:57
or the Panama Canal
200
462000
2000
07:59
or putting a man on the Moon --
201
464000
2000
08:01
there is a curious fact about them,
202
466000
2000
08:03
and it is that they were all done with about the same number off people.
203
468000
2000
08:05
It's weird; they were all done with about 100,000 people.
204
470000
3000
08:08
And the reason for that is because, before the Internet,
205
473000
3000
08:11
coordinating more than 100,000 people,
206
476000
2000
08:13
let alone paying them, was essentially impossible.
207
478000
3000
08:16
But now with the Internet, I've just shown you a project
208
481000
2000
08:18
where we've gotten 750 million people
209
483000
2000
08:20
to help us digitize human knowledge.
210
485000
2000
08:22
So the question that motivates my research is,
211
487000
2000
08:24
if we can put a man on the Moon with 100,000,
212
489000
3000
08:27
what can we do with 100 million?
213
492000
2000
08:29
So based on this question,
214
494000
2000
08:31
we've had a lot of different projects that we've been working on.
215
496000
2000
08:33
Let me tell you about one that I'm most excited about.
216
498000
3000
08:36
This is something that we've been semi-quietly working on
217
501000
2000
08:38
for the last year and a half or so.
218
503000
2000
08:40
It hasn't yet been launched. It's called Duolingo.
219
505000
2000
08:42
Since it hasn't been launched, shhhhh!
220
507000
2000
08:44
(Laughter)
221
509000
2000
08:46
Yeah, I can trust you'll do that.
222
511000
2000
08:48
So this is the project. Here's how it started.
223
513000
2000
08:50
It started with me posing a question to my graduate student,
224
515000
2000
08:52
Severin Hacker.
225
517000
2000
08:54
Okay, that's Severin Hacker.
226
519000
2000
08:56
So I posed the question to my graduate student.
227
521000
2000
08:58
By the way, you did hear me correctly;
228
523000
2000
09:00
his last name is Hacker.
229
525000
2000
09:02
So I posed this question to him:
230
527000
2000
09:04
How can we get 100 million people
231
529000
2000
09:06
translating the Web into every major language for free?
232
531000
3000
09:09
Okay, so there's a lot of things to say about this question.
233
534000
2000
09:11
First of all, translating the Web.
234
536000
2000
09:13
So right now the Web is partitioned into multiple languages.
235
538000
3000
09:16
A large fraction of it is in English.
236
541000
2000
09:18
If you don't know any English, you can't access it.
237
543000
2000
09:20
But there's large fractions in other different languages,
238
545000
2000
09:22
and if you don't know those languages, you can't access it.
239
547000
3000
09:25
So I would like to translate all of the Web, or at least most of the Web,
240
550000
3000
09:28
into every major language.
241
553000
2000
09:30
So that's what I would like to do.
242
555000
2000
09:32
Now some of you may say, why can't we use computers to translate?
243
557000
3000
09:35
Why can't we use machine translation?
244
560000
2000
09:37
Machine translation nowadays is starting to translate some sentences here and there.
245
562000
2000
09:39
Why can't we use it to translate the whole Web?
246
564000
2000
09:41
Well the problem with that is that it's not yet good enough
247
566000
2000
09:43
and it probably won't be for the next 15 to 20 years.
248
568000
2000
09:45
It makes a lot of mistakes.
249
570000
2000
09:47
Even when it doesn't make a mistake,
250
572000
2000
09:49
since it makes so many mistakes, you don't know whether to trust it or not.
251
574000
3000
09:52
So let me show you an example
252
577000
2000
09:54
of something that was translated with a machine.
253
579000
2000
09:56
Actually it was a forum post.
254
581000
2000
09:58
It was somebody who was trying to ask a question about JavaScript.
255
583000
3000
10:01
It was translated from Japanese into English.
256
586000
3000
10:04
So I'll just let you read.
257
589000
2000
10:06
This person starts apologizing
258
591000
2000
10:08
for the fact that it's translated with a computer.
259
593000
2000
10:10
So the next sentence is is going to be the preamble to the question.
260
595000
3000
10:13
So he's just explaining something.
261
598000
2000
10:15
Remember, it's a question about JavaScript.
262
600000
3000
10:19
(Text: At often, the goat-time install a error is vomit.)
263
604000
4000
10:23
(Laughter)
264
608000
4000
10:27
Then comes the first part of the question.
265
612000
3000
10:30
(Text: How many times like the wind, a pole, and the dragon?)
266
615000
4000
10:34
(Laughter)
267
619000
2000
10:36
Then comes my favorite part of the question.
268
621000
3000
10:39
(Text: This insult to father's stones?)
269
624000
3000
10:42
(Laughter)
270
627000
2000
10:44
And then comes the ending, which is my favorite part of the whole thing.
271
629000
3000
10:47
(Text: Please apologize for your stupidity. There are a many thank you.)
272
632000
4000
10:51
(Laughter)
273
636000
2000
10:53
Okay, so computer translation, not yet good enough.
274
638000
2000
10:55
So back to the question.
275
640000
2000
10:57
So we need people to translate the whole Web.
276
642000
3000
11:00
So now the next question you may have is,
277
645000
2000
11:02
well why can't we just pay people to do this?
278
647000
2000
11:04
We could pay professional language translators to translate the whole Web.
279
649000
3000
11:07
We could do that.
280
652000
2000
11:09
Unfortunately, it would be extremely expensive.
281
654000
2000
11:11
For example, translating a tiny, tiny fraction of the whole Web, Wikipedia,
282
656000
3000
11:14
into one other language, Spanish.
283
659000
3000
11:17
Wikipedia exists in Spanish,
284
662000
2000
11:19
but it's very small compared to the size of English.
285
664000
2000
11:21
It's about 20 percent of the size of English.
286
666000
2000
11:23
If we wanted to translate the other 80 percent into Spanish,
287
668000
3000
11:26
it would cost at least 50 million dollars --
288
671000
2000
11:28
and this is at even the most exploited, outsourcing country out there.
289
673000
3000
11:31
So it would be very expensive.
290
676000
2000
11:33
So what we want to do is we want to get 100 million people
291
678000
2000
11:35
translating the Web into every major language
292
680000
2000
11:37
for free.
293
682000
2000
11:39
Now if this is what you want to do,
294
684000
2000
11:41
you pretty quickly realize you're going to run into two pretty big hurdles,
295
686000
2000
11:43
two big obstacles.
296
688000
2000
11:45
The first one is a lack of bilinguals.
297
690000
3000
11:48
So I don't even know
298
693000
2000
11:50
if there exists 100 million people out there using the Web
299
695000
3000
11:53
who are bilingual enough to help us translate.
300
698000
2000
11:55
That's a big problem.
301
700000
2000
11:57
The other problem you're going to run into is a lack of motivation.
302
702000
2000
11:59
How are we going to motivate people
303
704000
2000
12:01
to actually translate the Web for free?
304
706000
2000
12:03
Normally, you have to pay people to do this.
305
708000
3000
12:06
So how are we going to motivate them to do it for free?
306
711000
2000
12:08
Now when we were starting to think about this, we were blocked by these two things.
307
713000
3000
12:11
But then we realized, there's actually a way
308
716000
2000
12:13
to solve both these problems with the same solution.
309
718000
2000
12:15
There's a way to kill two birds with one stone.
310
720000
2000
12:17
And that is to transform language translation
311
722000
3000
12:20
into something that millions of people want to do,
312
725000
3000
12:23
and that also helps with the problem of lack of bilinguals,
313
728000
3000
12:26
and that is language education.
314
731000
3000
12:29
So it turns out that today,
315
734000
2000
12:31
there are over 1.2 billion people learning a foreign language.
316
736000
3000
12:34
People really, really want to learn a foreign language.
317
739000
2000
12:36
And it's not just because they're being forced to do so in school.
318
741000
3000
12:39
For example, in the United States alone,
319
744000
2000
12:41
there are over five million people who have paid over $500
320
746000
2000
12:43
for software to learn a new language.
321
748000
2000
12:45
So people really, really want to learn a new language.
322
750000
2000
12:47
So what we've been working on for the last year and a half is a new website --
323
752000
3000
12:50
it's called Duolingo --
324
755000
2000
12:52
where the basic idea is people learn a new language for free
325
757000
3000
12:55
while simultaneously translating the Web.
326
760000
2000
12:57
And so basically they're learning by doing.
327
762000
2000
12:59
So the way this works
328
764000
2000
13:01
is whenever you're a just a beginner, we give you very, very simple sentences.
329
766000
3000
13:04
There's, of course, a lot of very simple sentences on the Web.
330
769000
2000
13:06
We give you very, very simple sentences
331
771000
2000
13:08
along with what each word means.
332
773000
2000
13:10
And as you translate them, and as you see how other people translate them,
333
775000
3000
13:13
you start learning the language.
334
778000
2000
13:15
And as you get more and more advanced,
335
780000
2000
13:17
we give you more and more complex sentences to translate.
336
782000
2000
13:19
But at all times, you're learning by doing.
337
784000
2000
13:21
Now the crazy thing about this method
338
786000
2000
13:23
is that it actually really works.
339
788000
2000
13:25
First of all, people are really, really learning a language.
340
790000
2000
13:27
We're mostly done building it, and now we're testing it.
341
792000
2000
13:29
People really can learn a language with it.
342
794000
2000
13:31
And they learn it about as well as the leading language learning software.
343
796000
3000
13:34
So people really do learn a language.
344
799000
2000
13:36
And not only do they learn it as well,
345
801000
2000
13:38
but actually it's way more interesting.
346
803000
2000
13:40
Because you see with Duolingo, people are actually learning with real content.
347
805000
3000
13:43
As opposed to learning with made-up sentences,
348
808000
2000
13:45
people are learning with real content, which is inherently interesting.
349
810000
3000
13:48
So people really do learn a language.
350
813000
2000
13:50
But perhaps more surprisingly,
351
815000
2000
13:52
the translations that we get from people using the site,
352
817000
3000
13:55
even though they're just beginners,
353
820000
2000
13:57
the translations that we get are as accurate as those of professional language translators,
354
822000
3000
14:00
which is very surprising.
355
825000
2000
14:02
So let me show you one example.
356
827000
2000
14:04
This is a sentence that was translated from German into English.
357
829000
2000
14:06
The top is the German.
358
831000
2000
14:08
The middle is an English translation
359
833000
2000
14:10
that was done by somebody who was a professional English translator
360
835000
2000
14:12
who we paid 20 cents a word for this translation.
361
837000
2000
14:14
And the bottom is a translation by users of Duolingo,
362
839000
3000
14:17
none of whom knew any German
363
842000
2000
14:19
before they started using the site.
364
844000
2000
14:21
You can see, it's pretty much perfect.
365
846000
2000
14:23
Now of course, we play a trick here
366
848000
2000
14:25
to make the translations as good as professional language translators.
367
850000
2000
14:27
We combine the translations of multiple beginners
368
852000
3000
14:30
to get the quality of a single professional translator.
369
855000
3000
14:33
Now even though we're combining the translations,
370
858000
5000
14:38
the site actually can translate pretty fast.
371
863000
2000
14:40
So let me show you,
372
865000
2000
14:42
this is our estimates of how fast we could translate Wikipedia
373
867000
2000
14:44
from English into Spanish.
374
869000
2000
14:46
Remember, this is 50 million dollars-worth of value.
375
871000
3000
14:49
So if we wanted to translate Wikipedia into Spanish,
376
874000
2000
14:51
we could do it in five weeks with 100,000 active users.
377
876000
3000
14:54
And we could do it in about 80 hours with a million active users.
378
879000
3000
14:57
Since all the projects that my group has worked on so far have gotten millions of users,
379
882000
3000
15:00
we're hopeful that we'll be able to translate
380
885000
2000
15:02
extremely fast with this project.
381
887000
2000
15:04
Now the thing that I'm most excited about with Duolingo
382
889000
3000
15:07
is I think this provides a fair business model for language education.
383
892000
3000
15:10
So here's the thing:
384
895000
2000
15:12
The current business model for language education
385
897000
2000
15:14
is the student pays,
386
899000
2000
15:16
and in particular, the student pays Rosetta Stone 500 dollars.
387
901000
2000
15:18
(Laughter)
388
903000
2000
15:20
That's the current business model.
389
905000
2000
15:22
The problem with this business model
390
907000
2000
15:24
is that 95 percent of the world's population doesn't have 500 dollars.
391
909000
3000
15:27
So it's extremely unfair towards the poor.
392
912000
3000
15:30
This is totally biased towards the rich.
393
915000
2000
15:32
Now see, in Duolingo,
394
917000
2000
15:34
because while you learn
395
919000
2000
15:36
you're actually creating value, you're translating stuff --
396
921000
3000
15:39
which for example, we could charge somebody for translations.
397
924000
3000
15:42
So this is how we could monetize this.
398
927000
2000
15:44
Since people are creating value while they're learning,
399
929000
2000
15:46
they don't have to pay their money, they pay with their time.
400
931000
3000
15:49
But the magical thing here is that they're paying with their time,
401
934000
3000
15:52
but that is time that would have had to have been spent anyways
402
937000
2000
15:54
learning the language.
403
939000
2000
15:56
So the nice thing about Duolingo is I think it provides a fair business model --
404
941000
3000
15:59
one that doesn't discriminate against poor people.
405
944000
2000
16:01
So here's the site. Thank you.
406
946000
2000
16:03
(Applause)
407
948000
8000
16:11
So here's the site.
408
956000
2000
16:13
We haven't yet launched,
409
958000
2000
16:15
but if you go there, you can sign up to be part of our private beta,
410
960000
3000
16:18
which is probably going to start in about three or four weeks.
411
963000
2000
16:20
We haven't yet launched this Duolingo.
412
965000
2000
16:22
By the way, I'm the one talking here,
413
967000
2000
16:24
but actually Duolingo is the work of a really awesome team, some of whom are here.
414
969000
3000
16:27
So thank you.
415
972000
2000
16:29
(Applause)
416
974000
4000
ABOUT THE SPEAKER
Luis von Ahn - Computer scientistLuis von Ahn builds systems that combine humans and computers to solve large-scale problems that neither can solve alone.
Why you should listen
Louis von Ahn is an associate professor of Computer Science at Carnegie Mellon University, and he's at the forefront of the crowdsourcing craze. His work takes advantage of the evergrowing Web-connected population to acheive collaboration in unprecedented numbers. His projects aim to leverage the crowd for human good. His company reCAPTCHA, sold to Google in 2009, digitizes human knowledge (books), one word at a time. His new project is Duolingo, which aims to get 100 million people translating the Web in every major language.
Luis von Ahn | Speaker | TED.com