TEDxYouth@Sydney
Oscar Schwartz: Can a computer write poetry?
Filmed:
Readability: 4.3
875,724 views
If you read a poem and feel moved by it, but then find out it was actually written by a computer, would you feel differently about the experience? Would you think that the computer had expressed itself and been creative, or would you feel like you had fallen for a cheap trick? In this talk, writer Oscar Schwartz examines why we react so strongly to the idea of a computer writing poetry -- and how this reaction helps us understand what it means to be human.
Oscar Schwartz - Writer and poet
Oscar Schwartz's research and writing concerns the influence of digital technology on culture and human interaction. Full bio
Oscar Schwartz's research and writing concerns the influence of digital technology on culture and human interaction. Full bio
Double-click the English transcript below to play the video.
00:12
I have a question.
0
881
1230
00:15
Can a computer write poetry?
1
3422
1943
00:18
This is a provocative question.
2
6959
2077
00:21
You think about it for a minute,
3
9715
1718
00:23
and you suddenly have a bunch
of other questions like:
of other questions like:
4
11457
2590
00:26
What is a computer?
5
14769
1381
00:28
What is poetry?
6
16710
1575
00:30
What is creativity?
7
18707
1689
00:33
But these are questions
8
21650
1172
00:34
that people spend their entire
lifetime trying to answer,
lifetime trying to answer,
9
22846
3070
00:37
not in a single TED Talk.
10
25940
2224
00:40
So we're going to have to try
a different approach.
a different approach.
11
28188
2445
00:42
So up here, we have two poems.
12
30657
2143
00:45
One of them is written by a human,
13
33839
2276
00:48
and the other one's written by a computer.
14
36139
2102
00:50
I'm going to ask you to tell me
which one's which.
which one's which.
15
38754
2410
00:53
Have a go:
16
41858
1156
00:55
Poem 1: Little Fly / Thy summer's play, /
My thoughtless hand / Has brush'd away.
My thoughtless hand / Has brush'd away.
17
43038
4056
00:59
Am I not / A fly like thee? /
Or art not thou / A man like me?
Or art not thou / A man like me?
18
47118
3394
01:02
Poem 2: We can feel / Activist
through your life's / morning /
through your life's / morning /
19
50536
3299
01:05
Pauses to see, pope I hate the / Non
all the night to start a / great otherwise (...)
all the night to start a / great otherwise (...)
20
53859
4247
01:10
Alright, time's up.
21
58130
1359
01:11
Hands up if you think Poem 1
was written by a human.
was written by a human.
22
59513
4096
01:17
OK, most of you.
23
65547
1490
01:19
Hands up if you think Poem 2
was written by a human.
was written by a human.
24
67061
3023
01:23
Very brave of you,
25
71172
1190
01:24
because the first one was written
by the human poet William Blake.
by the human poet William Blake.
26
72855
4285
01:29
The second one was written by an algorithm
27
77784
2949
01:32
that took all the language
from my Facebook feed on one day
from my Facebook feed on one day
28
80757
3692
01:36
and then regenerated it algorithmically,
29
84473
2763
01:39
according to methods that I'll describe
a little bit later on.
a little bit later on.
30
87260
3590
01:43
So let's try another test.
31
91218
2404
01:46
Again, you haven't got ages to read this,
32
94398
2093
01:48
so just trust your gut.
33
96515
1612
01:50
Poem 1: A lion roars and a dog barks.
It is interesting / and fascinating
It is interesting / and fascinating
34
98151
4045
01:54
that a bird will fly and not / roar
or bark. Enthralling stories about animals
or bark. Enthralling stories about animals
35
102220
4303
01:58
are in my dreams and I will sing them all
if I / am not exhausted or weary.
if I / am not exhausted or weary.
36
106547
4060
02:02
Poem 2: Oh! kangaroos, sequins, chocolate
sodas! / You are really beautiful!
sodas! / You are really beautiful!
37
110631
3985
02:06
Pearls, / harmonicas, jujubes, aspirins!
All / the stuff they've always talked about (...)
All / the stuff they've always talked about (...)
38
114640
4358
02:11
Alright, time's up.
39
119022
1158
02:12
So if you think the first poem
was written by a human,
was written by a human,
40
120204
3137
02:15
put your hand up.
41
123365
1215
02:17
OK.
42
125687
1154
02:18
And if you think the second poem
was written by a human,
was written by a human,
43
126865
2675
02:21
put your hand up.
44
129564
1155
02:23
We have, more or less, a 50/50 split here.
45
131779
3810
02:28
It was much harder.
46
136157
1436
02:29
The answer is,
47
137617
1712
02:31
the first poem was generated
by an algorithm called Racter,
by an algorithm called Racter,
48
139353
3483
02:34
that was created back in the 1970s,
49
142860
3002
02:37
and the second poem was written
by a guy called Frank O'Hara,
by a guy called Frank O'Hara,
50
145886
3189
02:41
who happens to be
one of my favorite human poets.
one of my favorite human poets.
51
149099
2668
02:44
(Laughter)
52
152631
3058
02:48
So what we've just done now
is a Turing test for poetry.
is a Turing test for poetry.
53
156046
3228
02:52
The Turing test was first proposed
by this guy, Alan Turing, in 1950,
by this guy, Alan Turing, in 1950,
54
160018
4547
02:56
in order to answer the question,
55
164589
1564
02:58
can computers think?
56
166177
1637
03:00
Alan Turing believed that if
a computer was able
a computer was able
57
168245
2770
03:03
to have a to have a text-based
conversation with a human,
conversation with a human,
58
171039
3078
03:06
with such proficiency
such that the human couldn't tell
such that the human couldn't tell
59
174141
2770
03:08
whether they are talking
to a computer or a human,
to a computer or a human,
60
176935
2966
03:11
then the computer can be said
to have intelligence.
to have intelligence.
61
179925
2856
03:15
So in 2013, my friend
Benjamin Laird and I,
Benjamin Laird and I,
62
183270
3295
03:18
we created a Turing test
for poetry online.
for poetry online.
63
186589
2988
03:21
It's called bot or not,
64
189601
1277
03:22
and you can go and play it for yourselves.
65
190902
2044
03:24
But basically, it's the game
we just played.
we just played.
66
192970
2251
03:27
You're presented with a poem,
67
195245
1528
03:28
you don't know whether it was written
by a human or a computer
by a human or a computer
68
196797
3028
03:31
and you have to guess.
69
199849
1166
03:33
So thousands and thousands
of people have taken this test online,
of people have taken this test online,
70
201039
3191
03:36
so we have results.
71
204254
1449
03:37
And what are the results?
72
205727
1428
03:39
Well, Turing said that if a computer
could fool a human
could fool a human
73
207704
2879
03:42
30 percent of the time
that it was a human,
that it was a human,
74
210607
3019
03:45
then it passes the Turing test
for intelligence.
for intelligence.
75
213650
2397
03:48
We have poems on the bot or not database
76
216625
2438
03:51
that have fooled 65 percent
of human readers into thinking
of human readers into thinking
77
219087
2979
03:54
it was written by a human.
78
222090
1395
03:55
So, I think we have an answer
to our question.
to our question.
79
223959
2817
03:59
According to the logic of the Turing test,
80
227546
2348
04:01
can a computer write poetry?
81
229918
1928
04:03
Well, yes, absolutely it can.
82
231870
2351
04:07
But if you're feeling
a little bit uncomfortable
a little bit uncomfortable
83
235782
2346
04:10
with this answer, that's OK.
84
238152
1927
04:12
If you're having a bunch
of gut reactions to it,
of gut reactions to it,
85
240103
2316
04:14
that's also OK because
this isn't the end of the story.
this isn't the end of the story.
86
242443
3205
04:18
Let's play our third and final test.
87
246594
2324
04:22
Again, you're going to have to read
88
250000
1750
04:23
and tell me which you think is human.
89
251774
1909
04:25
Poem 1: Reg flags the reason
for pretty flags. / And ribbons.
for pretty flags. / And ribbons.
90
253707
3718
04:29
Ribbons of flags / And wearing material /
Reasons for wearing material. (...)
Reasons for wearing material. (...)
91
257449
4321
04:33
Poem 2: A wounded deer leaps
highest, / I've heard the daffodil
highest, / I've heard the daffodil
92
261794
3918
04:37
I've heard the flag to-day /
I've heard the hunter tell; /
I've heard the hunter tell; /
93
265736
3446
04:41
'Tis but the ecstasy of death, /
And then the brake is almost done (...)
And then the brake is almost done (...)
94
269206
3702
04:44
OK, time is up.
95
272932
1599
04:46
So hands up if you think Poem 1
was written by a human.
was written by a human.
96
274555
3837
04:51
Hands up if you think Poem 2
was written by a human.
was written by a human.
97
279973
3038
04:55
Whoa, that's a lot more people.
98
283035
2331
04:58
So you'd be surprised to find that Poem 1
99
286327
2968
05:01
was written by the very
human poet Gertrude Stein.
human poet Gertrude Stein.
100
289319
3993
05:06
And Poem 2 was generated
by an algorithm called RKCP.
by an algorithm called RKCP.
101
294100
5038
05:11
Now before we go on, let me describe
very quickly and simply,
very quickly and simply,
102
299162
3319
05:14
how RKCP works.
103
302505
1781
05:16
So RKCP is an algorithm
designed by Ray Kurzweil,
designed by Ray Kurzweil,
104
304873
3850
05:20
who's a director of engineering at Google
105
308747
2222
05:22
and a firm believer
in artificial intelligence.
in artificial intelligence.
106
310993
2360
05:25
So, you give RKCP a source text,
107
313822
3991
05:29
it analyzes the source text in order
to find out how it uses language,
to find out how it uses language,
108
317837
4469
05:34
and then it regenerates language
109
322330
1948
05:36
that emulates that first text.
110
324302
2528
05:38
So in the poem we just saw before,
111
326854
2113
05:40
Poem 2, the one that you all
thought was human,
thought was human,
112
328991
2625
05:43
it was fed a bunch of poems
113
331640
1550
05:45
by a poet called Emily Dickinson
114
333214
2035
05:47
it looked at the way she used language,
115
335273
2189
05:49
learned the model,
116
337486
1165
05:50
and then it regenerated a model
according to that same structure.
according to that same structure.
117
338675
4258
05:56
But the important thing to know about RKCP
118
344732
2178
05:58
is that it doesn't know the meaning
of the words it's using.
of the words it's using.
119
346934
2838
06:02
The language is just raw material,
120
350359
2276
06:04
it could be Chinese,
it could be in Swedish,
it could be in Swedish,
121
352659
2160
06:06
it could be the collected language
from your Facebook feed for one day.
from your Facebook feed for one day.
122
354843
4179
06:11
It's just raw material.
123
359046
1652
06:13
And nevertheless, it's able
to create a poem
to create a poem
124
361380
2697
06:16
that seems more human
than Gertrude Stein's poem,
than Gertrude Stein's poem,
125
364101
3327
06:19
and Gertrude Stein is a human.
126
367452
2153
06:22
So what we've done here is,
more or less, a reverse Turing test.
more or less, a reverse Turing test.
127
370846
4072
06:27
So Gertrude Stein, who's a human,
is able to write a poem
is able to write a poem
128
375940
5179
06:33
that fools a majority
of human judges into thinking
of human judges into thinking
129
381143
3738
06:36
that it was written by a computer.
130
384905
1826
06:39
Therefore, according to the logic
of the reverse Turing test,
of the reverse Turing test,
131
387176
4141
06:43
Gertrude Stein is a computer.
132
391341
1916
06:45
(Laughter)
133
393281
1462
06:47
Feeling confused?
134
395358
1294
06:49
I think that's fair enough.
135
397193
1515
06:51
So far we've had humans
that write like humans,
that write like humans,
136
399546
4116
06:55
we have computers that write
like computers,
like computers,
137
403686
3111
06:58
we have computers that write like humans,
138
406821
3055
07:01
but we also have,
perhaps most confusingly,
perhaps most confusingly,
139
409900
3632
07:05
humans that write like computers.
140
413556
2375
07:08
So what do we take from all of this?
141
416938
1766
07:11
Do we take that William Blake
is somehow more of a human
is somehow more of a human
142
419611
3157
07:14
than Gertrude Stein?
143
422792
1249
07:16
Or that Gertrude Stein is more
of a computer than William Blake?
of a computer than William Blake?
144
424065
3046
07:19
(Laughter)
145
427135
1552
07:20
These are questions
I've been asking myself
I've been asking myself
146
428711
2323
07:23
for around two years now,
147
431058
1465
07:24
and I don't have any answers.
148
432547
2309
07:26
But what I do have are a bunch of insights
149
434880
2330
07:29
about our relationship with technology.
150
437234
2534
07:32
So my first insight is that,
for some reason,
for some reason,
151
440999
3609
07:36
we associate poetry with being human.
152
444632
3111
07:40
So that when we ask,
"Can a computer write poetry?"
"Can a computer write poetry?"
153
448197
3715
07:43
we're also asking,
154
451936
1193
07:45
"What does it mean to be human
155
453153
1798
07:46
and how do we put boundaries
around this category?
around this category?
156
454975
3172
07:50
How do we say who or what
can be part of this category?"
can be part of this category?"
157
458171
3658
07:54
This is an essentially
philosophical question, I believe,
philosophical question, I believe,
158
462376
3351
07:57
and it can't be answered
with a yes or no test,
with a yes or no test,
159
465751
2229
08:00
like the Turing test.
160
468004
1327
08:01
I also believe that Alan Turing
understood this,
understood this,
161
469805
3045
08:04
and that when he devised
his test back in 1950,
his test back in 1950,
162
472874
3305
08:08
he was doing it
as a philosophical provocation.
as a philosophical provocation.
163
476203
2802
08:13
So my second insight is that,
when we take the Turing test for poetry,
when we take the Turing test for poetry,
164
481124
5541
08:18
we're not really testing
the capacity of the computers
the capacity of the computers
165
486689
3460
08:22
because poetry-generating algorithms,
166
490173
2893
08:25
they're pretty simple and have existed,
more or less, since the 1950s.
more or less, since the 1950s.
167
493090
4563
08:31
What we are doing with the Turing
test for poetry, rather,
test for poetry, rather,
168
499055
3118
08:34
is collecting opinions about what
constitutes humanness.
constitutes humanness.
169
502197
4615
08:40
So, what I've figured out,
170
508313
2729
08:43
we've seen this when earlier today,
171
511066
2972
08:46
we say that William Blake
is more of a human
is more of a human
172
514062
2478
08:48
than Gertrude Stein.
173
516564
1565
08:50
Of course, this doesn't mean
that William Blake
that William Blake
174
518153
2462
08:52
was actually more human
175
520639
1828
08:54
or that Gertrude Stein
was more of a computer.
was more of a computer.
176
522491
2327
08:57
It simply means that the category
of the human is unstable.
of the human is unstable.
177
525533
4714
09:03
This has led me to understand
178
531450
2074
09:05
that the human is not a cold, hard fact.
179
533548
2763
09:08
Rather, it is something
that's constructed with our opinions
that's constructed with our opinions
180
536832
3132
09:11
and something that changes over time.
181
539988
2855
09:16
So my final insight is that
the computer, more or less,
the computer, more or less,
182
544671
4479
09:21
works like a mirror
that reflects any idea of a human
that reflects any idea of a human
183
549174
4006
09:25
that we show it.
184
553204
1375
09:26
We show it Emily Dickinson,
185
554958
1884
09:28
it gives Emily Dickinson back to us.
186
556866
2321
09:31
We show it William Blake,
187
559768
1834
09:33
that's what it reflects back to us.
188
561626
2285
09:35
We show it Gertrude Stein,
189
563935
1839
09:37
what we get back is Gertrude Stein.
190
565798
2470
09:41
More than any other bit of technology,
191
569083
2368
09:43
the computer is a mirror that reflects
any idea of the human we teach it.
any idea of the human we teach it.
192
571475
5165
09:50
So I'm sure a lot of you have been hearing
193
578061
2287
09:52
a lot about artificial
intelligence recently.
intelligence recently.
194
580372
2862
09:56
And much of the conversation is,
195
584694
2830
10:00
can we build it?
196
588292
1189
10:02
Can we build an intelligent computer?
197
590383
3135
10:05
Can we build a creative computer?
198
593542
2763
10:08
What we seem to be asking over and over
199
596329
2113
10:10
is can we build a human-like computer?
200
598466
2724
10:13
But what we've seen just now
201
601961
1556
10:15
is that the human
is not a scientific fact,
is not a scientific fact,
202
603541
3088
10:18
that it's an ever-shifting,
concatenating idea
concatenating idea
203
606653
3530
10:22
and one that changes over time.
204
610207
2531
10:24
So that when we begin
to grapple with the ideas
to grapple with the ideas
205
612762
3152
10:27
of artificial intelligence in the future,
206
615938
2386
10:30
we shouldn't only be asking ourselves,
207
618348
1905
10:32
"Can we build it?"
208
620277
1368
10:33
But we should also be asking ourselves,
209
621669
1894
10:35
"What idea of the human
do we want to have reflected back to us?"
do we want to have reflected back to us?"
210
623587
3713
10:39
This is an essentially philosophical idea,
211
627820
2693
10:42
and it's one that can't be answered
with software alone,
with software alone,
212
630537
2997
10:45
but I think requires a moment
of species-wide, existential reflection.
of species-wide, existential reflection.
213
633558
4977
10:51
Thank you.
214
639040
1153
10:52
(Applause)
215
640217
2695
ABOUT THE SPEAKER
Oscar Schwartz - Writer and poetOscar Schwartz's research and writing concerns the influence of digital technology on culture and human interaction.
Why you should listen
Oscar Schwartz is an Australian writer and poet undertaking a PhD that asks whether a computer can write poetry. His research led to the development of a Turing test for poetry, which is available on a website he cofounded called bot or not.
Oscar Schwartz | Speaker | TED.com