Can a computer write poetry?
-
0:01 - 0:02I have a question.
-
0:03 - 0:05Can a computer write poetry?
-
0:07 - 0:09This is a provocative question.
-
0:10 - 0:11You think about it for a minute,
-
0:11 - 0:14and you suddenly have a bunch
of other questions like: -
0:15 - 0:16What is a computer?
-
0:17 - 0:18What is poetry?
-
0:19 - 0:20What is creativity?
-
0:22 - 0:23But these are questions
-
0:23 - 0:26that people spend their entire
lifetime trying to answer, -
0:26 - 0:28not in a single TED Talk.
-
0:28 - 0:31So we're going to have to try
a different approach. -
0:31 - 0:33So up here, we have two poems.
-
0:34 - 0:36One of them is written by a human,
-
0:36 - 0:38and the other one's written by a computer.
-
0:39 - 0:41I'm going to ask you to tell me
which one's which. -
0:42 - 0:43Have a go:
-
0:43 - 0:47Poem 1: Little Fly / Thy summer's play, /
My thoughtless hand / Has brush'd away. -
0:47 - 0:51Am I not / A fly like thee? /
Or art not thou / A man like me? -
0:51 - 0:54Poem 2: We can feel / Activist
through your life's / morning / -
0:54 - 0:58Pauses to see, pope I hate the / Non
all the night to start a / great otherwise (...) -
0:58 - 0:59Alright, time's up.
-
1:00 - 1:04Hands up if you think Poem 1
was written by a human. -
1:06 - 1:07OK, most of you.
-
1:07 - 1:10Hands up if you think Poem 2
was written by a human. -
1:11 - 1:12Very brave of you,
-
1:13 - 1:17because the first one was written
by the human poet William Blake. -
1:18 - 1:21The second one was written by an algorithm
-
1:21 - 1:24that took all the language
from my Facebook feed on one day -
1:24 - 1:27and then regenerated it algorithmically,
-
1:27 - 1:31according to methods that I'll describe
a little bit later on. -
1:31 - 1:34So let's try another test.
-
1:34 - 1:36Again, you haven't got ages to read this,
-
1:37 - 1:38so just trust your gut.
-
1:38 - 1:42Poem 1: A lion roars and a dog barks.
It is interesting / and fascinating -
1:42 - 1:47that a bird will fly and not / roar
or bark. Enthralling stories about animals -
1:47 - 1:51are in my dreams and I will sing them all
if I / am not exhausted or weary. -
1:51 - 1:55Poem 2: Oh! kangaroos, sequins, chocolate
sodas! / You are really beautiful! -
1:55 - 1:59Pearls, / harmonicas, jujubes, aspirins!
All / the stuff they've always talked about (...) -
1:59 - 2:00Alright, time's up.
-
2:00 - 2:03So if you think the first poem
was written by a human, -
2:03 - 2:05put your hand up.
-
2:06 - 2:07OK.
-
2:07 - 2:10And if you think the second poem
was written by a human, -
2:10 - 2:11put your hand up.
-
2:12 - 2:16We have, more or less, a 50/50 split here.
-
2:16 - 2:18It was much harder.
-
2:18 - 2:19The answer is,
-
2:19 - 2:23the first poem was generated
by an algorithm called Racter, -
2:23 - 2:26that was created back in the 1970s,
-
2:26 - 2:29and the second poem was written
by a guy called Frank O'Hara, -
2:29 - 2:32who happens to be
one of my favorite human poets. -
2:33 - 2:36(Laughter)
-
2:36 - 2:39So what we've just done now
is a Turing test for poetry. -
2:40 - 2:45The Turing test was first proposed
by this guy, Alan Turing, in 1950, -
2:45 - 2:46in order to answer the question,
-
2:46 - 2:48can computers think?
-
2:48 - 2:51Alan Turing believed that if
a computer was able -
2:51 - 2:54to have a to have a text-based
conversation with a human, -
2:54 - 2:57with such proficiency
such that the human couldn't tell -
2:57 - 3:00whether they are talking
to a computer or a human, -
3:00 - 3:03then the computer can be said
to have intelligence. -
3:03 - 3:07So in 2013, my friend
Benjamin Laird and I, -
3:07 - 3:10we created a Turing test
for poetry online. -
3:10 - 3:11It's called bot or not,
-
3:11 - 3:13and you can go and play it for yourselves.
-
3:13 - 3:15But basically, it's the game
we just played. -
3:15 - 3:17You're presented with a poem,
-
3:17 - 3:20you don't know whether it was written
by a human or a computer -
3:20 - 3:21and you have to guess.
-
3:21 - 3:24So thousands and thousands
of people have taken this test online, -
3:24 - 3:26so we have results.
-
3:26 - 3:27And what are the results?
-
3:28 - 3:31Well, Turing said that if a computer
could fool a human -
3:31 - 3:3430 percent of the time
that it was a human, -
3:34 - 3:36then it passes the Turing test
for intelligence. -
3:37 - 3:39We have poems on the bot or not database
-
3:39 - 3:42that have fooled 65 percent
of human readers into thinking -
3:42 - 3:43it was written by a human.
-
3:44 - 3:47So, I think we have an answer
to our question. -
3:48 - 3:50According to the logic of the Turing test,
-
3:50 - 3:52can a computer write poetry?
-
3:52 - 3:54Well, yes, absolutely it can.
-
3:56 - 3:58But if you're feeling
a little bit uncomfortable -
3:58 - 4:00with this answer, that's OK.
-
4:00 - 4:02If you're having a bunch
of gut reactions to it, -
4:02 - 4:06that's also OK because
this isn't the end of the story. -
4:07 - 4:09Let's play our third and final test.
-
4:10 - 4:12Again, you're going to have to read
-
4:12 - 4:14and tell me which you think is human.
-
4:14 - 4:17Poem 1: Reg flags the reason
for pretty flags. / And ribbons. -
4:17 - 4:22Ribbons of flags / And wearing material /
Reasons for wearing material. (...) -
4:22 - 4:26Poem 2: A wounded deer leaps
highest, / I've heard the daffodil -
4:26 - 4:29I've heard the flag to-day /
I've heard the hunter tell; / -
4:29 - 4:33'Tis but the ecstasy of death, /
And then the brake is almost done (...) -
4:33 - 4:35OK, time is up.
-
4:35 - 4:38So hands up if you think Poem 1
was written by a human. -
4:40 - 4:43Hands up if you think Poem 2
was written by a human. -
4:43 - 4:45Whoa, that's a lot more people.
-
4:46 - 4:49So you'd be surprised to find that Poem 1
-
4:49 - 4:53was written by the very
human poet Gertrude Stein. -
4:54 - 4:59And Poem 2 was generated
by an algorithm called RKCP. -
4:59 - 5:02Now before we go on, let me describe
very quickly and simply, -
5:03 - 5:04how RKCP works.
-
5:05 - 5:09So RKCP is an algorithm
designed by Ray Kurzweil, -
5:09 - 5:11who's a director of engineering at Google
-
5:11 - 5:13and a firm believer
in artificial intelligence. -
5:14 - 5:18So, you give RKCP a source text,
-
5:18 - 5:22it analyzes the source text in order
to find out how it uses language, -
5:22 - 5:24and then it regenerates language
-
5:24 - 5:27that emulates that first text.
-
5:27 - 5:29So in the poem we just saw before,
-
5:29 - 5:32Poem 2, the one that you all
thought was human, -
5:32 - 5:33it was fed a bunch of poems
-
5:33 - 5:35by a poet called Emily Dickinson
-
5:35 - 5:37it looked at the way she used language,
-
5:37 - 5:39learned the model,
-
5:39 - 5:43and then it regenerated a model
according to that same structure. -
5:45 - 5:47But the important thing to know about RKCP
-
5:47 - 5:50is that it doesn't know the meaning
of the words it's using. -
5:50 - 5:53The language is just raw material,
-
5:53 - 5:55it could be Chinese,
it could be in Swedish, -
5:55 - 5:59it could be the collected language
from your Facebook feed for one day. -
5:59 - 6:01It's just raw material.
-
6:01 - 6:04And nevertheless, it's able
to create a poem -
6:04 - 6:07that seems more human
than Gertrude Stein's poem, -
6:07 - 6:10and Gertrude Stein is a human.
-
6:11 - 6:15So what we've done here is,
more or less, a reverse Turing test. -
6:16 - 6:21So Gertrude Stein, who's a human,
is able to write a poem -
6:21 - 6:25that fools a majority
of human judges into thinking -
6:25 - 6:27that it was written by a computer.
-
6:27 - 6:31Therefore, according to the logic
of the reverse Turing test, -
6:31 - 6:33Gertrude Stein is a computer.
-
6:33 - 6:35(Laughter)
-
6:35 - 6:37Feeling confused?
-
6:37 - 6:39I think that's fair enough.
-
6:40 - 6:44So far we've had humans
that write like humans, -
6:44 - 6:47we have computers that write
like computers, -
6:47 - 6:50we have computers that write like humans,
-
6:50 - 6:54but we also have,
perhaps most confusingly, -
6:54 - 6:56humans that write like computers.
-
6:57 - 6:59So what do we take from all of this?
-
7:00 - 7:03Do we take that William Blake
is somehow more of a human -
7:03 - 7:04than Gertrude Stein?
-
7:04 - 7:07Or that Gertrude Stein is more
of a computer than William Blake? -
7:07 - 7:09(Laughter)
-
7:09 - 7:11These are questions
I've been asking myself -
7:11 - 7:13for around two years now,
-
7:13 - 7:15and I don't have any answers.
-
7:15 - 7:17But what I do have are a bunch of insights
-
7:17 - 7:20about our relationship with technology.
-
7:21 - 7:25So my first insight is that,
for some reason, -
7:25 - 7:28we associate poetry with being human.
-
7:28 - 7:32So that when we ask,
"Can a computer write poetry?" -
7:32 - 7:33we're also asking,
-
7:33 - 7:35"What does it mean to be human
-
7:35 - 7:38and how do we put boundaries
around this category? -
7:38 - 7:42How do we say who or what
can be part of this category?" -
7:42 - 7:46This is an essentially
philosophical question, I believe, -
7:46 - 7:48and it can't be answered
with a yes or no test, -
7:48 - 7:49like the Turing test.
-
7:50 - 7:53I also believe that Alan Turing
understood this, -
7:53 - 7:56and that when he devised
his test back in 1950, -
7:56 - 7:59he was doing it
as a philosophical provocation. -
8:01 - 8:07So my second insight is that,
when we take the Turing test for poetry, -
8:07 - 8:10we're not really testing
the capacity of the computers -
8:10 - 8:13because poetry-generating algorithms,
-
8:13 - 8:18they're pretty simple and have existed,
more or less, since the 1950s. -
8:19 - 8:22What we are doing with the Turing
test for poetry, rather, -
8:22 - 8:27is collecting opinions about what
constitutes humanness. -
8:28 - 8:31So, what I've figured out,
-
8:31 - 8:34we've seen this when earlier today,
-
8:34 - 8:37we say that William Blake
is more of a human -
8:37 - 8:38than Gertrude Stein.
-
8:38 - 8:41Of course, this doesn't mean
that William Blake -
8:41 - 8:42was actually more human
-
8:42 - 8:45or that Gertrude Stein
was more of a computer. -
8:46 - 8:50It simply means that the category
of the human is unstable. -
8:51 - 8:54This has led me to understand
-
8:54 - 8:56that the human is not a cold, hard fact.
-
8:57 - 9:00Rather, it is something
that's constructed with our opinions -
9:00 - 9:03and something that changes over time.
-
9:05 - 9:09So my final insight is that
the computer, more or less, -
9:09 - 9:13works like a mirror
that reflects any idea of a human -
9:13 - 9:15that we show it.
-
9:15 - 9:17We show it Emily Dickinson,
-
9:17 - 9:19it gives Emily Dickinson back to us.
-
9:20 - 9:22We show it William Blake,
-
9:22 - 9:24that's what it reflects back to us.
-
9:24 - 9:26We show it Gertrude Stein,
-
9:26 - 9:28what we get back is Gertrude Stein.
-
9:29 - 9:31More than any other bit of technology,
-
9:31 - 9:37the computer is a mirror that reflects
any idea of the human we teach it. -
9:38 - 9:40So I'm sure a lot of you have been hearing
-
9:40 - 9:43a lot about artificial
intelligence recently. -
9:45 - 9:48And much of the conversation is,
-
9:48 - 9:49can we build it?
-
9:50 - 9:54Can we build an intelligent computer?
-
9:54 - 9:56Can we build a creative computer?
-
9:56 - 9:58What we seem to be asking over and over
-
9:58 - 10:01is can we build a human-like computer?
-
10:02 - 10:04But what we've seen just now
-
10:04 - 10:07is that the human
is not a scientific fact, -
10:07 - 10:10that it's an ever-shifting,
concatenating idea -
10:10 - 10:13and one that changes over time.
-
10:13 - 10:16So that when we begin
to grapple with the ideas -
10:16 - 10:18of artificial intelligence in the future,
-
10:18 - 10:20we shouldn't only be asking ourselves,
-
10:20 - 10:22"Can we build it?"
-
10:22 - 10:24But we should also be asking ourselves,
-
10:24 - 10:27"What idea of the human
do we want to have reflected back to us?" -
10:28 - 10:31This is an essentially philosophical idea,
-
10:31 - 10:34and it's one that can't be answered
with software alone, -
10:34 - 10:39but I think requires a moment
of species-wide, existential reflection. -
10:39 - 10:40Thank you.
-
10:40 - 10:43(Applause)
- Title:
- Can a computer write poetry?
- Speaker:
- Oscar Schwartz
- Description:
-
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.
- Video Language:
- English
- Team:
closed TED
- Project:
- TEDTalks
- Duration:
- 10:56
![]() |
Brian Greene edited English subtitles for Can a computer write poetry? | |
![]() |
Retired user commented on English subtitles for Can a computer write poetry? | |
![]() |
Maricene Crus commented on English subtitles for Can a computer write poetry? | |
![]() |
Brian Greene edited English subtitles for Can a computer write poetry? | |
![]() |
Brian Greene edited English subtitles for Can a computer write poetry? | |
![]() |
Brian Greene edited English subtitles for Can a computer write poetry? | |
![]() |
Brian Greene approved English subtitles for Can a computer write poetry? | |
![]() |
Brian Greene edited English subtitles for Can a computer write poetry? |
Maricene Crus
Just a question:
shouldn't the subtitles for the poems be written between square brackets since they are shown in slides and not spoken?
Thank you!
Retired user
A typo at 04:13 It should read "Red" instead of "Reg"