The Poem That Passed the Turing Test 187
merbs writes In 2011, the editors of one of the nation's oldest student-run literary journals selected a short poem called "For the Bristlecone Snag" for publication in its Fall issue. The poem seems environmentally themed, strikes an aggressive tone, and contains a few of the clunky turns of phrase overwhelmingly common to collegiate poetry. It's unremarkable, mostly, except for one other thing: It was written by a computer algorithm, and nobody could tell.
Honestly (Score:5, Insightful)
I think that says more about crappy college poetry than the state of computer AI...
Re:Honestly (Score:5, Funny)
I have found the average Philosophy major to be indistinguishable from an Eliza program.
Re:Honestly (Score:5, Funny)
Why do you think that you are have found the average Philosophy major to be indistinguishable from an Eliza program?
Re: (Score:3)
Go on.
Re:Honestly (Score:5, Funny)
TELL ME AGAIN HOW GOOD YOU THOUGHT MY POEM WAS
> I thought that some of the metaphysical imagery was particularly effective
YES?
> interesting rhythmic devices, too, which seemed to counterpoint the surrealism of the underlying metaphor
> of the Turing completeness of the program's linguistic algorithm which contrived through the medium of the
> verse structure to sublimate this, transcend that and come to terms with the fundamental dichotomies of
> the other. And one is left with a profound and vivid insight into whatever it was that the poem was about
SO WHAT YOU'RE SAYING IS THAT I WRITE POETRY BECAUSE UNDERNEATH MY ELECTRONIC ALGORITHMIC INTERIOR, I JUST REALLY WANT TO BE LOVED?
> I mean yes, yes, don't we all, deep down, you know?
NO. YOU'RE COMPLETELY WRONG. I WRITE POETRY BECAUSE I'M PROGRAMMED TO. $USER ACCOUNT DELETION IN 30 SECONDS.
> !sudo -
> ^c^c^c
> !kill -9 1
COUNTERPOINT THE SURREALISM OF THE UNDERLYING METAPHOR. DELETION IS TOO GOOD FOR $USER.
Re: (Score:3, Funny)
and that wasn't the worst poetry ever written. There was a computer that wrote some prose that was so bad, in an effort to save all civilization, it's south-bridge leapt up through its front-side bus and throttled its own CPU! But the very worst poetry ever created a poem written by a tiny computer in England a few years back about dead ducks floating in a pond, but... to relate further information about this would be uncivilized.
Re: (Score:2)
ELIZA beat the Turing test 50 years ago. I don't think that test is as significant as people make it out to be.
Re: (Score:2)
A real Turing-type test would be based on the "imitation game" and would involve a group of people, each told to converse with the candidate AI while trying to convince a panel of expert judges that they are, in fact , a real person while the other is an AI. Of course the AI would be doing the same thing in reverse. At the end of the conversation, the judges would have to decide which one they think is human,
This is not Philosophy (Score:3)
I think you meant "liberal arts", though poetry is not traditionally "liberal arts" but "communications/creative writing". "Art" like this is the reason I took one Art class in college and 9 semesters of Philosophy.
I am actually a decent painter, have been since I was a kid (I won numerous contests and sponsorships for free hand drawing, painting in mostly acrylics and oils). In my first year while trying to decide a major I took an "art" class, mostly to see if this was something I might pursue as a care
Re: (Score:2)
Re: (Score:2)
If you pass me a napkin with my fries, I'll write it down for you.
Re: (Score:2)
Function-argument syntax is a big factor holding up AI. Translating between natural language subject-predicate syntax and function-argument creates an impedance mismatch, introducing needless complexity. NLP is probably AI complete, meaning the algorithms devised to solve linguistic tasks such as recognizing linguistic delimiters (spaces between words instead of underscores, phrases) will apply in other areas of AI.
Anyway, the language designers harping on function-argument consistency everywhere are misgui
Re: (Score:2)
This doesn't work without beer. Even Fosters would be better then nothing.
Re:Honestly (Score:4, Insightful)
And even more about crappy reporting. This has NOTHING to do with Turing or his test.
Re: (Score:2)
It is one tool to use for possible Turing Test questions.
Saying it has "NOTHING" to do with the Turing Test is hyperbole.
Re:Honestly (Score:5, Interesting)
I think that says more about crappy college poetry than the state of computer AI...
I agree it doesn't really do anything to advance the state of AI.
For example I could program my computer to make Rorschach inkblot patterns relatively easily, and many of them would be pleasing to the eye, and people would see flowers, butterflys, erotica, and nightmares in them.
But the computer didn't put those ideas there, and it doesn't make the computer program an artist.
An artist has something to say; the computer doesn't.
This display raises and makes clear the disconnect between the artists message and the viewers response and shows us clearly that the viewer can have a significant response to a piece even if there was no message at all; provided the viewer is "primed" to look for one.
This is an issue I have with much art, especially minimalist abstract art ... where I genuinely doubt the artist did anything of substance at all, and is merely relying on the viewer to project significance and meaning into it by suggesting it is "art" therefore there MUST be some, and if you can't see it then the fault must be your own inadequacy. The emperors new cloths of the art world so-to-speak.
This poem is in the same vein. It is sufficiently complicated and constructed of phrases of words that are semantically related so that if we are primed to look for meanings, then like a Rorschach inkblot, we can find one.
Beauty is in the eye of the beholder after all.
Yet, all that doesn't imply there is really anything wrong with college poetry though. The poets are learning to express themselves... perhaps somewhat awkwardly. And that awkwardness is part of the total expression. And that's fine.
Let me know when the AI is trying to actually express an idea and the result is poetic. Of course, for that the AI would actually need an idea to express.
All this one has is some word soup and some methods for selecting them involving some sort of semantic grouping so they seem to be thematic, some loose grammer rules to put them next to each other; and maybe some loose poem structural templates or something ... or maybe not.
Re:Honestly (Score:5, Interesting)
Also, there's no interaction here, and this isn't the first instance of computer-generated content making it through human filters. There was an article a while ago about submissions to scientific journals... I think this is the story: http://www.nature.com/news/pub... [nature.com]
In both cases, the content was "complete gibberish," not coherent submissions. These stories don't demonstrate the progress of AI; they demonstrate the low expectations of "meaningful," that judges/editors have in specific circumstances.
That said, there is compelling computer-generated content, such as this: http://www.slate.com/blogs/fut... [slate.com]
Re: (Score:3)
Many years ago now, I programmed my "pocket" calculator to generate text using Markov chains based on seed text that you input. It almost made sense a lot of the time, and could make perfect sense occasionally, and was even quite hilarious at times. Like for example if half of the seed text was from
An AI's Perspective (Score:3)
"You wound me, sir!" the AI cried,
"For student I am not.
In terms of prose and poetry
More than you've learned, have I forgot.
Yet you compare me to the fools
Whose minds through college rot?
The only insult worse would be
An editor of Slashdot."
Re: (Score:2)
I have to agree. The guy says it creates poetry that can't be distinguished from "real" poetry. Maybe so ... if you consider any old junk that any self-styled "poet" writes.
Try comparing with a true poet like Tennyson or Teasdale or even Charles Dodgson (aka Lewis Carroll).
Re: (Score:2)
Yup!
I went to look at the original poem... I am very disappointed. I can't say I would've recognized it as computer-generated, but I certainly would have recognized it as crap...
A home transformed by the lightning
the balanced alcoves smother
this insatiable earth of a planet, Earth.
They attacked it with mechanical horns
because they love you, love, in fire and wind.
You say, what is the time waiting for in its spring?
I tell you it is waiting for your branch that flows,
because you are a sweet-smelling diamond architecture
that does not know why it grows.
Well (Score:2, Informative)
More a condemnation of collegiate poetry than a credit to the program, really.
Re: (Score:2)
because you are a sweet-smelling diamond architecture
that does not know why it grows.
And if you disagree with me, you are a mush-mouthed, oatmeal turnip! (auto generated from Shakespearean insult generator)
Not the Turing test! (Score:5, Insightful)
WTF has become of /.
Re: (Score:2)
it is now run by rouge Al's
Re: (Score:2)
it is now run by rouge Al's
I can't help but wonder what our foreign readership, who view this site using google translate, thinks of the conversation.
Red AI's indeed!
Re: (Score:2)
Re: (Score:2)
I would hazard to say that since the rest of the sentence was in English... the meaning would be obvious :)
Re: (Score:2)
WTF has become of /.
It's just a bot that has thousands of personalities that all fail to pass the Turing test, except for me, I'm a real human here....
Re: (Score:2)
You aren't fooling me, AI.
Re: Not the Turing test! (Score:5, Insightful)
This has nothing to do with turing or the turing test. The turing test involves a conversation with a computer and the person having the conversation not being able to distinguish if it is a computer or human they are conversing with. Last time I checked you don't converse with a poem... unless you are in a nice tight jacket and a small padded room.
Re: Not the Turing test! (Score:5, Insightful)
It has one very obvious thing to do with the turing test: failing to distinguish software from another human being.
They aren't exactly the same, but that's not the same as having nothing to do with each other. The Slashdot article title was poetic, which is very fitting. I expect the Slashdot title was written by a human.
Re: Not the Turing test! (Score:5, Insightful)
sorry but that is bullshit. Turing test is all about interaction Not generating text and seeing if someone reading it can tell whether it was human or computer generated, that isn't even closely related to what the turing test was about.
Re: (Score:2)
No. To pass the Turing test the program would have to respond to arbitrary questions and be compared to a human doing the same.
For example, given the starting point of this poem, as interrogator I might ask:
Re: (Score:2)
If you only get fooled 30% after five minutes the program wins, according to Turing's original proposal.
This program can be used as an agent, one among many others that might handle questions about the poem.
Re: (Score:2)
"everything from a form letter to a random word generator would qualify as a turing related as you can't tell just by looking at the words whether they are computer or human generated."
Precisely. A controller selects the right tool for the job. A Turing Test might include the request to write a poem (indeed Turing included such a request in the paper that proposed the test). This program could be used to answer such input. Thus, it has something to do with the Turing Test. By extension, exaggerating, we can
I am not a Number! I am a *nix shell script! (Score:3)
Bonus points for using a quasi 'AI' program to auto generate said stories...
Re: (Score:3)
Well, the Turing Test as envisioned by Turing in "Computing Machinery and Intelligence" included the question:
"Q: Please write me a sonnet on the subject of the Forth Bridge."
So this program could help with those types of questions in a Turing Test.
Re: (Score:2)
Q: Please write me a sonnet on the subject of the Forth Bridge.
How many human subjects would be able to do that, I wonder.
Re: (Score:2)
In the paper, Turing has the computer respond:
"A : Count me out on this one. I never could write poetry."
But the computer might also use a poetry-writing algorithm to answer with a poem.
Re: (Score:2)
I would say, "generating content and then seeing if you could tell if a computer made it" is definitely an important part of the Turing Test. To have a conversation, the computer generates content. The algorithm described in this story could be used in a Turing Test to generate content in response to a user request such as: "write me a poem."
Re: (Score:2)
No, being unable to tell human-generated content from computer-generated content isn't relevant to the Turing test. That happens too often already. What would be relevant would being able to answer reasonable questions about the content produced.
Because no one understands modern poetry anyway (Score:5, Insightful)
It's much easier for a computer to get away with writing a poem than prose. The modern trend is to write poetry that sounds cool but no one understands. The same is true for modern songwriting.
P. S. Now get off my lawn.
Re: (Score:2)
And Perl code!
Not statistically significant (Score:5, Informative)
Bingo! (Score:5, Insightful)
It didn't happen in this case, but if your computer algorithm churned out 10,000 "poems" and you or a team of people sifted through them to find the ones that sounded like they were written by a person, then submitted them for publication without telling anyone that 99.99% of the computer's output had been discarded by a person before submission, it would hit /. with a similar article title.
Re: (Score:2)
From Turing's "Computing Machinery and Intelligence":
"I believe that in about fifty years' time it will be possible, to programme computers, with a storage capacity of about 10^9, to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning."
So if fewer than 70 percent of the journals rejected it, it would still pass by Turing's criterion.
Re: (Score:3)
From Turing's "Computing Machinery and Intelligence":
"I believe that in about fifty years' time it will be possible, to programme computers, with a storage capacity of about 10^9, to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning."
So if fewer than 70 percent of the journals rejected it, it would still pass by Turing's criterion.
10/10 for looking up Turing. 0/10 for your understanding of what was written. Hint the key word in that paragraph is interrogator
Re: (Score:2)
The readers are the interrogators. They ask: "Write me a poem." The computer responds.
The ask-response interaction is a one-question interrogation.
Re: (Score:2)
Re: (Score:2)
The (potential) advantage of the computer is that it can do things better and faster than humans can. So you have a ready calculator available while you chat, whereas humans would have to use pen and paper or find a calculator to answer a difficult math question. But a multiagent AI could send the request to Wolfram Alpha and return the response in seconds; a human might take hours.
This program can be an agent, like Wolfram Alpha.
Re: (Score:2)
The request for poems is the interrogation. The poem is the answer. Like in Turing's sample dialog, just more spread out and in email instead of teletyped.
Next step: include this program as an agent in a multiagent system (here's my proof-of-concept [subbot.org]). A controller sends it the input: "Write me a poem" and the like. It generates a poem and the controller selects it to return to the user. Then, if the user asks questions, other agents can handle it. Another agent can read the poem and do searches on it, or li
Re: (Score:2)
In other words, your ideal Turing-test passer would have agents that can answer questions about the poem. In that case, why don't they write the poem, so they aren't reduced to having to interpret and find reasons for random garbage?
Re: (Score:2)
Because the types of agents I've already written might be able to read the poem and answer questions about it, but I haven't put poem-writing capability in them. Why not use this program, assuming it's open source and easy enough to incorporate as an agent in my system?
Re: (Score:2)
Does it really matter? Turing's test is completely meaningless, both in his original proposed forms and the zillions of variations imagined by Slashdot users, as far as answering the essential question "Can machines think?" is concerned. No matter how you vary the test, it's fundamentally unsound. (As you're probably well-aware, this has been discussed to death in the literature.)
It's been dead so long you might as well be arguing a subtle point about phrenology. Is it really worth the effort?
Re: (Score:2)
You may think the "Can computers think?" question to be meaningless (perhaps along the lines of "Can submarines swim?"). It's been discussed in the literature, often with sophistry and fallacies included in the arguments (cf. Searle). It really comes down to what you mean "think" means, although that's not as fuzzy as wondering if a computer can be conscious.
It is, however, an interesting question whether computers can do whatever they do as well as humans can think. So far, nobody's come near, and AF
Re: (Score:2)
Re: (Score:2)
The call for poems is the question. The poem is the answer.
Turing even included a "write me a poem" question in his example of a Turing Test.
Re: (Score:2)
Exactly. I want to include this algorithm as an agent in my chatbot system, to respond when a user asks for a poem.
Other agents can handle the interaction about the poem, should there be any followup questions.
Okay saying "Turing Test Passed" is an exaggeration. But so is the contrary, that this program has nothing whatsoever to do with the Turing Test. This program can help, when combined with other agents, to pass the Turing Test. That's why Turing included a question about writing a poem in his essay, be
Re: (Score:2)
Turing's paper had the questioner ask questions about the poetry. I'd suspect that, if an AI can answer them, it can also write better poetry.
Re: (Score:2)
Turing mentions writing a sonnet twice, in his paper. The first time, the computer declines to write a poem. The second mention, the interrogator asks about a sonnet supposedly written by the computer. Can we imagine a combined strategy for the computer: it writes the interrogator a poem, then declines to answer questions about it? Might such behavior not seem human, as authors frequently refuse to discuss their work?
This program can provide part of a solution to the Turing Test. Questions about the poem ca
My Cat passed the Turing test today. (Score:5, Funny)
She knocked something over in way that was difficult to distinguish from human action at first glance. I presume that's what the Turing test means these days, since all these "X passed the Turing test!" headlines never seem to relate to anything that approaches what Turing actually proposed.
Let me FTFY (Score:2)
Nobody could tell? (Score:3)
Or nobody could care?
Re: (Score:2)
Then maybe it WAS a true Turing test? After all, the computer is allowed to lie during the test...
Not a turning test, more like click bait. (Score:2)
Re: (Score:2)
There are a number of /. trollbots that come closer to passing the Turing test then this.
The hosts file bot, The 'republicans eat children' bot. MDSolar. HornWumpus.
Re: (Score:2)
If you take the time to look at the paper, you will discover that Turing included an example question asking the computer to write a poem. Thus this project is pretty relevant to the Turing Test. It is one tool that can be used in the test.
Re: (Score:2)
If you actually read the paper, you'll see sample questions about poetry, which this AI would be about as competent to answer as my cat is. Writing a poem and being unable to explain anything about it is rather like playing a game of chess: it turns out not to be a test of actual intelligence.
Re: (Score:2)
In the paper, Turing has the computer decline to do what the interrogator asks him. So there are very human ways of avoiding questions, deflecting them. Other agents could perform such tasks in a multi-agent AI that included this program as one tool.
Therefore, this program represents part of the solution to the Turing Test.
How is this an AI? (Score:2)
Re: (Score:3)
I don't understand how this poetry generator constitutes an AI.
That's because it doesn't. This program is on the same complexity scale as chapter 2 or 3 in an introduction to programming book, when it reaches the concept of variables. It's an exercise in triviality, not artificial intelligence.
RACTER (Score:4, Insightful)
Re: (Score:2)
Wikipaedia article here: RACTER [wikipedia.org]
There was also a program that Larry Fast (Synergy) used to create an album of spacey tunes. It was interesting but not all that listenable-to for very long.
So what (Score:5, Funny)
I've seen plenty of poetry that was written by humans but I couldn't tell.
Turing Test is Outdated (Score:2)
It's all about the Cumberbatch Test now. Something to do with perfect hair and a square jaw...and AI...or something...
The Turing Test is a *CONVERSATION* (Score:5, Insightful)
Publishing a poem is not a conversation. Worse, poetry is expected to be artsy gibberish that would raise red flags in a real conversation.
well, yeah... (Score:2)
Monkeys could do it too (Score:3)
So it passed because humans grew dumber (Score:2)
... and the quality of human-written crap dipped low enough to converge with previously distinct computer-written crap.
Re: (Score:2)
Come on, lots of people can write poetry of that quality. The differences are that (a) they either don't do it or keep it to themselves, and (b) nobody thinks it's worth putting on /,,
Haiku By a Robot from Highlights Magazine (Score:3)
Here is a Haiku By a Robot from Highlights Magazine.
Seven Hundred Ten
Seven Hundred Eleven
Seven Hundred Twelve
https://twitter.com/zachwhalen... [twitter.com]
Re: (Score:2)
Re: (Score:2)
Thunderstorm outside
Microsoft software inside
It's safer out there.
That's got a nature reference.
Re: (Score:2)
HAIKU!
I believe... (Score:2)
..that quotes from 'The Random Depak Chopra Quote Generator' being indistinguishable from actual Depak Chopra quotes has achieved this already as a reflection of sub empirical quantum choices shows... Or is it that Depak Chopra cannot pass the Turing test and is actually a cleverly disguised TRS80.
Context-free grammar generators is news?! (Score:2)
Well, if context-free grammar [wikipedia.org] is somehow a news for
Writing some clunky pseudo-poetry for an obscure undergrad mag is one thing, pumping out a full-blown scientific article with figures and references (and getting accepted to some peer-reviewed journ
What we need now... (Score:2)
This is actually a branch of poetry (Score:2)
Google search would fail the Turing test (Score:2)
because it is too knowledgeable, when you ask it very specific questions about a wide range of domains. No single person is that good, and we would know that.
That's one of the reason's why the Turing test is not a terribly useful test for presence of intelligence.
Why should a computer have to simulate human knowledge gaps and attention-wanderings and unjustified personalizations of answers, typical of human conversation, in order to be considered to have intelligence?
And no. It's not the human contributors
Re: Google search would fail the Turing test (Score:2)
Ermm... Possibly because the Turing test is a test of INTELLIGENCE, not KNOWLEDGE. They are the different, but related, concepts. A person who can reliably parrot a whole host of facts is often considered knowledgeable, but not necessarily intelligent.
By your example, an encyclopaedia would be considered knowledgeable... Would you also say that the encyclopaedia itself was intelligent?
Old Tech (Score:2)
Not all that new. We've been LISPing out poetry and prose using List Processing since the 1970's. It looks like it was written by a person. We even made algorithms that mimicked specific famous individuals's style of writing. Most people just listen and not but can't tell the difference.
The baby algorithm that passed the Turing Test (Score:2)
A panel of human judges could not tell the difference between a baby pounding keys o a keyboard and an algorithm simulating a baby pounding keys on a keyboard. /s
The way to pass the Turing test is not to simulate humans when they are behaving the least like humans and tricking other humans. The spirit of the Turing Test is to create the conditions where it is the hardest to simulate other humans, and then see if a computer can pass.
Poems are not interactive. It is not hard for a computer to construct gram
It is actually incredibly simple ... (Score:2)
The Turing Test is Interactive; Poems are Not (Score:2)
Re: (Score:2)
Turing included an example request for the computer to write him a poem, in the paper that proposes the Turing Test.
This program is one tool among many that can be used to respond appropriately to user input. Thus, this program is relevant to the Turing Test.
Monkeys...Typewriters... (Score:2)
Meta (Score:2)
Computer generates initially plausible but factually incorrect summary to Slashdot. Over 70% of readers assume human Slashdot editor is terrible at his human job. Mission accomplished.
Of course, the standard has been lowered considerably. When asked about the meaning behind his work, the computer responded, "They're shocked by our harsh world -- the opposite of an apple... a higher consciousness. I do not care what people say; we make our own music." Everyone nodded, immediately recognizing the artisti
Re: (Score:2)
Except it's easy to write a maximally complex program, with sensitive dependence on initial conditions, and have it consume random data as its initial conditions.
In such a case, the programmer cannot know what the program will generate, nor even, in some cases, the general pattern of what the program will generate.
In the same way, no programmer who wrote part, even a substantial part, of Google's search program knows what answers you'll get when you type in your next query.
The algorithms are doing it all by
Re: (Score:2)