Can a robot pass a university entrance exam? | Noriko Arai

205,170 views ・ 2017-09-13

TED


Dvaput kliknite na engleske titlove ispod za reprodukciju videozapisa.

Prevoditelj: Senzos Osijek Recezent: Sanda L
00:13
Today, I'm going to talk about AI and us.
0
13014
3660
Danas ću govoriti o nama i umjetnoj inteligenciji.
00:18
AI researchers have always said
1
18206
2143
Istraživači umjetne inteligencije uvijek govore
00:20
that we humans do not need to worry,
2
20373
2594
da se mi ljudi ne trebamo brinuti
00:22
because only menial jobs will be taken over by machines.
3
22991
3580
jer će strojevi preuzeti jedino teške, slabo plaćene poslove.
00:27
Is that really true?
4
27274
1603
Je li to zaista istina?
00:30
They have also said that AI will create new jobs,
5
30365
3827
Također govore da će roboti stvoriti nova radna mjesta,
00:34
so those who lose their jobs will find a new one.
6
34216
3411
tako da oni koji izgube posao, mogu pronaći novi.
00:38
Of course.
7
38264
1355
Naravno.
00:39
But the real question is:
8
39643
2172
No, pravo je pitanje koje se postavlja:
00:41
How many of those who may lose their jobs to AI
9
41839
4105
Koliko će onih koji zbog robota izgube posao
00:45
will be able to land a new one,
10
45968
2489
biti u mogućnosti ponovno se zaposliti,
00:48
especially when AI is smart enough to learn better than most of us?
11
48481
5850
pogotovo kad su roboti dovoljno pametni da mogu učiti bolje od većine nas?
00:55
Let me ask you a question:
12
55397
2185
Da vas pitam:
00:58
How many of you think
13
58666
1798
Koliko vas misli
01:00
that AI will pass the entrance examination of a top university by 2020?
14
60488
6094
da će do 2020. g. robot moći riješiti prijemni ispit vrhunskog sveučilišta?
01:07
Oh, so many. OK.
15
67836
2457
Tako mnogo. OK.
01:10
So some of you may say, "Of course, yes!"
16
70317
4358
Dakle, neki će od vas reći: “Naravno, da!”
01:15
Now singularity is the issue.
17
75369
2134
Problem je jedinstvenost.
01:18
And some others may say, "Maybe,
18
78590
3095
Neki drugi reći će: “Možda bi mogao,
01:21
because AI already won against a top Go player."
19
81709
4508
jednom je već pobijedio najboljeg igrača igrice Go.”
01:27
And others may say, "No, never. Uh-uh."
20
87213
3681
A ostali će reći: “Ne, nikad. A-a.”
01:32
That means we do not know the answer yet, right?
21
92195
3593
To znači da još ne znamo odgovor, zar ne?
01:36
So that was the reason why I started Todai Robot Project,
22
96268
4960
Dakle, to je razlog zbog kojega sam započela projekt Todai Robot,
01:41
making an AI which passes the entrance examination
23
101252
3872
tvoreći robota s umjetnom inteligencijom koji će moći položiti prijemni ispit
01:45
of the University of Tokyo,
24
105148
2589
na Sveučilištu u Tokiju,
01:47
the top university in Japan.
25
107761
2537
najboljem japanskom sveučilištu.
01:51
This is our Todai Robot.
26
111464
2548
Ovo je naš robot Todai.
01:56
And, of course, the brain of the robot is working in the remote server.
27
116131
5810
Naravno, njegov mozak radi preko udaljenog servera.
02:02
It is now writing a 600-word essay
28
122747
4222
Trenutno piše 600 riječi dugačak esej
02:06
on maritime trade in the 17th century.
29
126993
4119
o pomorskoj trgovini u 17. stoljeću.
02:11
How does that sound?
30
131136
1765
Kako vam to zvuči?
02:14
Why did I take the entrance exam as its benchmark?
31
134113
4104
Zašto sam uzela prijemni ispit kao mjerilo?
02:19
Because I thought we had to study the performance of AI
32
139098
4741
Zato što sam smatrala da moramo proučiti izvedbu umjetne inteligencije
02:23
in comparison to humans,
33
143863
2114
u usporedbi s ljudima,
02:26
especially on the skills and expertise
34
146001
2860
pogotovo u vještinama i znanju
02:28
which are believed to be acquired only by humans
35
148885
4088
za koje se smatra da ih mogu steći samo ljudi
02:32
and only through education.
36
152997
2335
i to samo kroz obrazovanje.
02:35
To enter Todai, the University of Tokyo,
37
155782
4043
Za upis na Todai, Sveučilište u Tokiju,
02:39
you have to pass two different types of exams.
38
159849
4421
potrebno je proći dva tipa ispita.
02:44
The first one is a national standardized test
39
164294
3760
Prvi je test standardizirani
02:48
in multiple-choice style.
40
168078
2403
s više mogućih odgovora.
02:50
You have to take seven subjects
41
170505
2455
Potrebno je položiti sedam predmeta
02:52
and achieve a high score --
42
172984
1955
i ostvariti visok rezultat --
02:54
I would say like an 85 percent or more accuracy rate --
43
174963
4772
rekla bih, uz točnost od oko 85% ili više,
02:59
to be allowed to take the second stage written test
44
179759
4087
da bi se moglo pristupiti drugom, pisanom dijelu ispita,
03:03
prepared by Todai.
45
183870
2159
koji priprema Sveučilište.
03:06
So let me first explain how modern AI works,
46
186994
5317
Objasnit ću vam kako funkcionira današnja umjetna inteligencija
03:12
taking the "Jeopardy!" challenge as an example.
47
192335
3069
na primjeru kviza Jeopardy.
03:17
Here is a typical "Jeopardy!" question:
48
197539
3079
Ovo je tipično pitanje i kvizu:
03:20
"Mozart's last symphony shares its name with this planet."
49
200642
4461
“Posljednja Mozartova simfonija dijeli ime s ovim planetom.”
03:26
Interestingly, a "Jeopardy!" question always asks,
50
206195
4013
Zanimljivo, pitanja tog kviza
03:30
always ends with "this" something:
51
210232
3328
uvijek završavaju riječju “ovaj”,
03:33
"this" planet, "this" country,
52
213584
2827
ovaj planet, ova država,
03:36
"this" rock musician, and so on.
53
216435
2608
ovaj glazbenik itd.
03:39
In other words, "Jeopardy!" doesn't ask many different types of questions,
54
219067
4299
Drugim riječima, Jeopardy ne postavlja mnogo različitih tipova pitanja,
03:43
but a single type,
55
223390
1837
već uvijek isti tip,
03:45
which we call "factoid questions."
56
225251
2536
koji zovemo “pitanja o činjenicama”.
03:48
By the way, do you know the answer?
57
228975
2167
Kad već spominjemo, znate li odgovor?
03:53
If you do not know the answer and if you want to know the answer,
58
233980
4055
Kad ne biste znali odgovor, a htjeli biste ga znati,
03:58
what would you do?
59
238059
1287
što biste učinili?
04:00
You Google, right? Of course.
60
240160
3132
Potražili na Googleu, zar ne? Naravno.
04:03
Why not?
61
243316
1480
Zašto ne?
04:04
But you have to pick appropriate keywords
62
244820
3592
Ali za pretragu morate odabrati prikladne ključne riječi,
04:08
like "Mozart," "last" and "symphony" to search.
63
248436
4364
kao što su “Mozart”, “posljednja” i “simfonija”.
04:13
The machine basically does the same.
64
253462
2400
Stroj radi praktički isto to.
04:16
Then this Wikipedia page will be ranked top.
65
256457
4660
Onda vam se na vrhu pojavi ova stranica s Wikipedije.
04:21
Then the machine reads the page.
66
261840
1908
Onda robot pročita tu stranicu.
04:23
No, uh-uh.
67
263772
1171
Ne, a-a.
04:25
Unfortunately, none of the modern AIs,
68
265470
3462
Nažalost, nijedan moderni robot,
04:28
including Watson, Siri and Todai Robot,
69
268956
3968
ni Watson, ni Siri, ni Todai,
04:32
is able to read.
70
272948
1661
ne može čitati.
04:35
But they are very good at searching and optimizing.
71
275437
3800
Ali, poprilično su dobri u pretraživanju i prilagođavanju.
04:40
It will recognize
72
280158
2023
Prepoznat će
04:42
that the keywords "Mozart," "last" and "symphony"
73
282866
2935
da se ključne riječi “Mozart”, “posljednja” i “simfonija”
04:45
are appearing heavily around here.
74
285825
2903
često pojavljuju na toj stranici.
04:49
So if it can find a word which is a planet
75
289790
4375
Ako pronađe riječ čije je značenje planet
04:54
and which is co-occurring with these keywords,
76
294189
3648
i koja se pojavljuje zajedno s tim ključnim riječima,
04:57
that must be the answer.
77
297861
1989
mora da je to onda odgovor.
05:00
This is how Watson finds the answer "Jupiter," in this case.
78
300762
5186
Tako Watson u ovom slučaju pronađe odgovor “Jupiter”.
05:08
Our Todai Robot works similarly, but a bit smarter
79
308433
4049
Todai robot radi na sličan način, ali je čak i pametniji
05:12
in answering history yes-no questions,
80
312506
3239
kad odgovara pitanja iz povijesti s mogućim odgovorima DA i NE.
05:16
like, "'Charlemagne repelled the Magyars.' Is this sentence true or false?"
81
316560
5663
Npr.: “Karlo Veliki protjerao je Mađare. Je li ova izjava istinita ili ne?”
05:23
Our robot starts producing a factoid question,
82
323181
4073
Naš robot počne to pretvarati u činjenično pitanje,
05:27
like: "Charlemagne repelled [this person type]" by itself.
83
327278
4899
poput: "Karlo Veliki protjerao je [ovaj narod]".
05:32
Then, "Avars" but not "Magyars" is ranked top.
84
332995
4732
Pri vrhu rezultata pojavi se riječ Avari, a ne Mađari.
05:38
This sentence is likely to be false.
85
338357
3049
Onda je velika vjerojatnost da je rečenica bila neistinita.
05:42
Our robot does not read, does not understand,
86
342772
4860
Naš robot ne može čitati niti razumjeti,
05:48
but it is statistically correct in many cases.
87
348335
4144
ali je statistički u mnogim slučajevima u pravu.
05:54
For the second stage written test,
88
354147
2508
U drugom dijelu ispita
05:56
it is required to write a 600-word essay like this one:
89
356679
5106
potrebno je napisati esej od 600 riječi prema ovom principu:
06:01
[Discuss the rise and fall of the maritime trade
90
361809
2278
[Objasni uspone i padove pomorske trgovine
06:04
in East and Southeast Asia in the 17th century ...]
91
364111
2422
u istočnoj i jugoistočnoj Aziji u 17. stoljeću ...]
06:06
and as I have shown earlier,
92
366557
1387
Kao što sam već pokazala,
06:07
our robot took the sentences from the textbooks and Wikipedia,
93
367968
4194
robot je preuzeo rečenice iz udžbenika i s Wikipedije,
06:12
combined them together,
94
372186
1961
udružio ih
06:14
and optimized it to produce an essay
95
374171
3619
i prilagodio ih tako da se može napisati esej,
06:17
without understanding a thing.
96
377814
2207
a da zapravo ništa nije razumio.
06:20
(Laughter)
97
380045
1737
(Smijeh)
06:21
But surprisingly, it wrote a better essay
98
381806
4895
Začudo, napisao je esej bolje
06:26
than most of the students.
99
386725
1561
nego većina studenata.
06:28
(Laughter)
100
388310
2391
(Smijeh)
06:30
How about mathematics?
101
390725
1529
A što je s matematikom?
06:33
A fully automatic math-solving machine
102
393354
3158
Stroj koji će potpuno samostalno rješavati matematičke zadatke
06:36
has been a dream
103
396536
1631
san je mnogih
06:38
since the birth of the word "artificial intelligence,"
104
398191
4679
od samih početaka razvitka umjetne inteligencije,
06:43
but it has stayed at the level of arithmetic for a long, long time.
105
403785
6007
ali dugo je vremena stagnirao na razini aritmetike.
06:51
Last year, we finally succeeded in developing a system
106
411530
5350
Prošle smo godine konačno uspjeli razviti sustav
06:56
which solved pre-university-level problems from end to end,
107
416904
5173
koji može od početka do kraja riješiti zadatke potrebne za upis na fakultet,
07:02
like this one.
108
422101
1262
poput ovoga.
07:05
This is the original problem written in Japanese,
109
425648
4002
Ovo je izvorni zadatak napisan na japanskom.
07:09
and we had to teach it 2,000 mathematical axioms
110
429674
4397
Morali smo ga naučiti 2000 aksioma
07:14
and 8,000 Japanese words
111
434095
2774
i 8000 japanskih riječi
07:16
to make it accept the problems written in natural language.
112
436893
4558
kako bi mogao prepoznati zadatke napisane običnim jezikom.
07:22
And it is now translating the original problems
113
442234
3542
Sada može prevesti izvorne zadatke
07:25
into machine-readable formulas.
114
445800
3139
u formule koje računalo može pročitati.
07:30
Weird, but it is now ready to solve it, I think.
115
450578
6099
Mislim da je sada spreman riješiti ih.
07:36
Go and solve it.
116
456701
1411
Hajde, riješi ih.
07:38
Yes! It is now executing symbolic computation.
117
458818
4284
To! Sada računa i pretvara simbole.
07:44
Even more weird,
118
464030
1580
Vrlo neobično,
07:45
but probably this is the most fun part for the machine.
119
465634
4825
ali ovo je vjerojatno najzabavniji dio.
07:50
(Laughter)
120
470483
2351
(Smijeh)
07:52
Now it outputs a perfect answer,
121
472858
2815
Sada je izbacio savršeno točan odgovor,
07:55
though its proof is impossible to read, even for mathematicians.
122
475697
4707
iako bi izvod i dokaz i matematičarima bilo teško pročitati.
08:02
Anyway, last year our robot was among the top one percent
123
482773
6961
Kako god, prošle je godine naš robot bio među najboljih 1%
08:10
in the second stage written exam in mathematics.
124
490199
3633
u drugom dijelu pisanog ispita iz matematike.
08:14
(Applause)
125
494652
3210
(Pljesak)
08:18
Thank you.
126
498412
1311
Hvala.
08:19
So, did it enter Todai?
127
499747
2471
I, je li upisao fakultet?
08:22
No, not as I expected.
128
502981
3058
Ne, iako sam očekivala da hoće.
08:26
Why?
129
506783
1399
Zašto?
08:28
Because it doesn't understand any meaning.
130
508206
2639
Jer ne razumije nikakva značenja.
08:32
Let me show you a typical error it made in the English test.
131
512308
4079
Pokazat ću vam tipičnu pogrešku koju je radio na ispitu iz engleskog.
08:36
[Nate: We're almost at the bookstore. Just a few more minutes.
132
516411
2977
[Nate: Uskoro smo u knjižari. Još samo nekoliko minuta.]
08:39
Sunil: Wait. ______ . Nate: Thank you! That always happens ...]
133
519412
3039
[Sunil: Čekaj. ______ . Nate: Hvala! To se uvijek događa ...]
08:42
Two people are talking.
134
522475
1151
Dvoje ljudi razgovara.
08:43
For us, who can understand the situation --
135
523650
2054
Za nas koji možemo razumjeti situaciju --
08:45
[1. "We walked for a long time." 2. "We're almost there."
136
525704
2773
[1. "Dugo već hodamo." 2. "Uskoro smo tamo."
08:48
3. "Your shoes look expensive." 4. "Your shoelace is untied."]
137
528501
3032
3. "Cipele ti izgledaju skupo." 4. "Odvezala ti se vezica."]
08:51
it is obvious number four is the correct answer, right?
138
531557
2873
očito je da je točan odgovor pod brojem 4, zar ne?
08:54
But Todai Robot chose number two,
139
534454
2238
No, robot je izabrao broj 2,
08:56
even after learning 15 billion English sentences
140
536716
5360
čak i nakon što je naučio 15 milijardi rečenica
09:02
using deep learning technologies.
141
542100
2728
koristeći napredne tehnologije učenja.
09:07
OK, so now you might understand what I said:
142
547600
4172
U redu, sad možete razumjeti moju izjavu
09:12
modern AIs do not read,
143
552399
2648
kako roboti ne čitaju,
09:15
do not understand.
144
555071
1413
ne razumiju.
09:17
They only disguise as if they do.
145
557516
3169
Samo se pretvaraju da to mogu.
09:24
This is the distribution graph
146
564867
2981
Ovo je graf distribucije
09:27
of half a million students who took the same exam as Todai Robot.
147
567872
5777
rezultata pola milijuna studenata koji su pisali isti ispit kao i robot.
09:34
Now our Todai Robot is among the top 20 percent,
148
574558
5165
Robot je u najboljih 20 posto
09:40
and it was capable to pass
149
580986
2415
i uspio je proći ispit
09:43
more than 60 percent of the universities in Japan --
150
583425
3941
na više od 60 posto sveučilišta u Japanu,
09:47
but not Todai.
151
587390
1377
ali ne u Tokiju.
09:50
But see how it is beyond the volume zone
152
590116
4025
No, vidimo da se nalazi iznad područja
09:54
of to-be white-collar workers.
153
594165
2864
gdje su budući zaposlenici na “finim uredskim poslovima”.
10:00
You might think I was delighted.
154
600060
2858
Mogli biste pomisliti da sam bila presretna.
10:03
After all, my robot was surpassing students everywhere.
155
603939
3971
Naposlijetku, moj je robot nadmašio velik broj studenata.
10:09
Instead, I was alarmed.
156
609022
2691
Nasuprot, bila sam uznemirena.
10:13
How on earth could this unintelligent machine outperform students --
157
613086
5607
Kako je uopće moguće da ovakav neinteligentni stroj nadmaši studente,
10:18
our children?
158
618717
1292
našu djecu?
10:20
Right?
159
620033
1153
Zar ne?
10:22
I decided to investigate what was going on in the human world.
160
622101
4402
Odlučila sam istražiti što se to događa s ljudima.
10:28
I took hundreds of sentences from high school textbooks
161
628542
4729
Uzela sam stotine rečenica iz srednjoškolskih udžbenika
10:33
and made easy multiple-choice quizzes,
162
633859
3313
i sastavila jednostavne ispite na zaokruživanje
10:37
and asked thousands of high school students to answer.
163
637196
4143
te dala tisućama srednjoškolaca da riješe.
10:42
Here is an example:
164
642690
1176
Evo jedan primjer:
10:43
[Buddhism spread to ... , Christianity to ... and Oceania,
165
643890
2818
[Budizam se proširio na... kršćanstvo na ... i Oceaniju,
10:46
and Islam to ...]
166
646732
1151
a islam na...]
10:47
Of course, the original problems are written in Japanese,
167
647907
2740
Naravno, pitanja su bila napisana na japanskom,
10:50
their mother tongue.
168
650671
1155
njihovom materinjem jeziku.
10:51
[ ______ has spread to Oceania.
169
651850
1515
[____ se proširilo na Oceaniju.
10:53
1. Hinduism 2. Christianity 3. Islam 4. Buddhism ]
170
653389
2417
1. hinduizam 2. kršćanstvo 3. islam 4. budizam]
10:55
Obviously, Christianity is the answer, isn't it?
171
655830
2299
Očito je da je kršćanstvo točan odgovor, zar ne?
10:58
It's written!
172
658153
1214
Pa napisano je!
11:01
And Todai Robot chose the correct answer, too.
173
661482
4026
I robot je izabrao točan odgovor,
11:06
But one-third of junior high school students
174
666758
4879
ali trećina učenika prve godine srednje škole
11:11
failed to answer this question.
175
671661
2612
nije znala odgovoriti na pitanje.
11:16
Do you think it is only the case in Japan?
176
676456
3159
Mislite li da se to događa samo u Japanu?
11:19
I do not think so,
177
679639
1976
Ne bih baš rekla
11:21
because Japan is always ranked among the top in OECD PISA tests,
178
681639
6371
jer je Japan uvijek bio među najboljima na OECD PISA testovima,
11:28
measuring 15-year-old students' performance in mathematics,
179
688034
3927
koji ocjenjuju petnaestogodišnje učenike u matematici,
11:31
science and reading
180
691985
1964
znanosti i čitanju,
11:33
every three years.
181
693973
1636
svake tri godine.
11:39
We have been believing
182
699390
2053
Vjerovali smo
11:41
that everybody can learn
183
701467
2043
da svatko može učiti
11:43
and learn well,
184
703534
1905
i dobro naučiti
11:45
as long as we provide good learning materials
185
705463
3697
ako pružimo dobre materijale
11:49
free on the web
186
709184
1455
besplatno na internetu,
11:50
so that they can access through the internet.
187
710663
3069
tako da su lako dostupni.
11:53
But such wonderful materials may benefit only those who can read well,
188
713756
5859
No, svi ti dobri materijali mogu koristiti samo onima koji mogu dobro čitati,
12:00
and the percentage of those who can read well
189
720534
3935
a postotak onih koji to mogu
12:04
may be much less than we expected.
190
724493
3378
mogao bi biti mnogo manji nego što smo očekivali.
12:10
How we humans will coexist with AI
191
730040
4241
Kako će ljudi koegzistirati s umjetnom inteligencijom,
12:14
is something we have to think about carefully,
192
734305
3522
pitanje je koje trebamo dobro razmotriti
12:17
based on solid evidence.
193
737851
2137
i bazirati na čvrstim dokazima.
12:21
At the same time, we have to think in a hurry
194
741063
3977
Istovremeno, moramo razmišljati u žurbi
12:25
because time is running out.
195
745064
2402
jer vremena je sve manje.
12:28
Thank you.
196
748106
1162
Hvala vam.
12:29
(Applause)
197
749626
3933
(Pljesak)
12:34
Chris Anderson: Noriko, thank you.
198
754211
2080
Chris Anderson: Noriko, hvala Vam.
12:36
Noriko Arai: Thank you.
199
756315
1765
Noriko Arai: Hvala Vama.
12:38
CA: In your talk, you so beautifully give us a sense of how AIs think,
200
758104
5304
Chris: Tijekom govora odlično ste nam prikazali kako roboti razmišljaju,
12:43
what they can do amazingly
201
763432
1564
što rade izvrsno,
12:45
and what they can't do.
202
765020
1695
a što ne mogu raditi.
12:46
But -- do I read you right,
203
766739
1494
Ali, jesam li dobro shvatio
12:48
that you think we really need quite an urgent revolution in education
204
768257
5270
da mislite kako zaista trebamo hitne promjene u obrazovanju
12:53
to help kids do the things that humans can do better than AIs?
205
773551
4155
kako bi djecu bolje naučili raditi stvari koje ljudi već rade bolje od robota?
12:57
NA: Yes, yes, yes.
206
777730
1328
Noriko: Da, da, da.
12:59
Because we humans can understand the meaning.
207
779082
4035
Jer mi ljudi možemo razumjeti značenja.
13:03
That is something which is very, very lacking in AI.
208
783141
4906
To je nešto što jako nedostaje u umjetnoj inteligenciji.
13:08
But most of the students just pack the knowledge
209
788071
4368
Ali, većina studenata samo “spremi” znanje,
13:12
without understanding the meaning of the knowledge,
210
792463
3903
a da ne razumije značenje gradiva,
13:16
so that is not knowledge, that is just memorizing,
211
796390
2809
dakle, to uopće nije pravo znanje, već samo memoriranje,
13:19
and AI can do the same thing.
212
799223
2450
a to mogu i roboti.
13:21
So we have to think about a new type of education.
213
801697
3631
Dakle, moramo smisliti novi oblik edukacije.
13:25
CA: A shift from knowledge, rote knowledge, to meaning.
214
805352
3289
Chris: Zamijeniti učenje napamet razumijevanjem.
13:28
NA: Mm-hmm.
215
808665
1151
Noriko: Da.
13:29
CA: Well, there's a challenge for the educators. Thank you so much.
216
809840
3240
Chris: Onda imamo izazov za profesore. Hvala Vam puno.
13:33
NA: Thank you very much. Thank you.
217
813104
1698
Noriko: Hvala Vama. Hvala.
13:34
(Applause)
218
814826
1185
(Pljesak)
O ovoj web stranici

Ova stranica će vas upoznati s YouTube videozapisima koji su korisni za učenje engleskog jezika. Vidjet ćete lekcije engleskog koje vode vrhunski profesori iz cijelog svijeta. Dvaput kliknite na engleske titlove prikazane na svakoj video stranici da biste reproducirali video s tog mjesta. Titlovi se pomiču sinkronizirano s reprodukcijom videozapisa. Ako imate bilo kakvih komentara ili zahtjeva, obratite nam se putem ovog obrasca za kontakt.

https://forms.gle/WvT1wiN1qDtmnspy7