3 ms·

Yes I agree entirely. Poetry is one of the purest forms of human expression. By human expression I mean the combination of ideas and emotions. Even after the

Yes I agree entirely. Poetry is one of the purest forms of human expression. By human expression I mean the combination of ideas and emotions. Even after the AI has acquired a language model and knowledge of rhyme and meter , and even perhaps a crude world model, A robot writing poetry would need to know what emotions something would typically evoke in order to write about it. For eg: to write a poem about a rotting pile of carcasses , or cockroaches it would require an innate understanding of how humans react to these things. What they might feel , the physical response and the visceral aspects of the reactions evoked. You could machine learn your way through thousands of poems and understand how lots of things feel but to produce true poetry the machine would have to completely imbibe our ideas of beauty and self. What objects are beautiful vs not beautiful , what sounds are harmonious vs dissonant and jarring , what kinds of faces are attractive vs ugly and so on and on and on... For example , what would it feel like to fly ? I'm not talking about the air rushing past , or looking down upon the world you whoosh past overhead. The one word that comes to my mind is a sense of exhilaration. Freedom. Perhaps a fear of heights.And I bet that's what came to your mind as well. But would our AI poet come to the same conclusion ? So this is not just AI Complete but an AI Insane problem. To appreciate and create poetry is to be human.

bhickey 11y ago
Why do you believe this? I have some familiarity with natural language processing and think that generating passable computer poetry is a fairly rudimentary task. Tell your model about meter, tell it about rhyme, learn word associations. Pick a language, give it every line of poetry you can find and hit go. What makes you think you could distinguish computational verse from something written by human hands? Try these examples. Which ones were written by people? a) A home transformed by the lightning the balanced alcoves smother this insatiable earth of a planet, Earth. b) For it matters not, how much we own; the cars ... the house ... the cash. What matters is how we live and love and how we spend our dash. c) It is really hard to break bad news without crying or falling apart.
- bpicolo 11y ago
  (a) seems inherently robotic via the phrase "insatiable earth of a planet, Earth.". That said, there are certainly poems that could go both ways. But art is a weird environment: It's not necessarily a piece in and of itself that creates human interest in there. I'm sure there are a lot of `great poems` out there that nobody cares about because they weren't written by some great poet.
  - bhickey 11y ago
    (a) is indeed written by a machine, (b) is excerpted from a famously awful poem, and (c) is an unintentional haiku found in the New York Times. I agree with what you're saying about art, but it's far afield of the point thewarrior was making.
- eli_gottlieb 11y ago
  > Why do you believe this? I have some familiarity with natural language processing and think that generating passable computer poetry is a fairly rudimentary task. Tell your model about meter, tell it about rhyme, learn word associations. Pick a language, give it every line of poetry you can find and hit go. What makes you think you could distinguish computational verse from something written by human hands? See, people always say that the trouble is getting your ML model to have some semantics to what it produces, to produce texts which somehow manage to genuinely correspond to the real world rather than just to the use of language. With poetry they might say, we don't just want something that fits the forms, but something with content, with soul. I say, yes, that's all true, and have you ever tried getting a randomly sampled human being to produce prose or poetry with meaningful content and soul? Untrained, we're possibly even worse than the neural nets!
visarga 11y ago
I appreciate your argument about ML lacking the embodied experience of humans and our feelings, but I think we can attack this problem. Anything we can get data on, can be learned. Emotions and physical sensations are accessible for collection in example datasets. ML is capable of detecting emotions in text, speech and human faces. In a similar way to learning to predict sentiment in reviews, we could learn to predict sentiment in photos and videos. On the other hand, ML can learn a thought vector (meaning) for every concept and how they relate. So it can learn both emotion and meaning. Combining the two, it could make poetry. It can also compose music pretty convincingly and make a painting in the style of Van Gogh from a simple picture, or make completely original style combinations (example: https://dreamscopeapp.com/ https://dreamscopeapp.com/). ML can't reach human level in music and painting yet, and surely not in poetry, but it might one day even surpass us. I think in the future humans will become the "children" of these artificial minds and one of their problems will be to explain their thoughts on our level. The tables will turn on us eventually.
kolinko 11y ago
You're making two assunptions: - that the process behind machine writing a poem would be the same as with human's - that one needs to understand what's being written to write well

bhickey 11y ago

Why do you believe this? I have some familiarity with natural language processing and think that generating passable computer poetry is a fairly rudimentary task. Tell your model about meter, tell it about rhyme, learn word associations. Pick a language, give it every line of poetry you can find and hit go. What makes you think you could distinguish computational verse from something written by human hands? Try these examples. Which ones were written by people? a) A home transformed by the lightning the balanced alcoves smother this insatiable earth of a planet, Earth. b) For it matters not, how much we own; the cars ... the house ... the cash. What matters is how we live and love and how we spend our dash. c) It is really hard to break bad news without crying or falling apart.

bpicolo 11y ago
(a) seems inherently robotic via the phrase "insatiable earth of a planet, Earth.". That said, there are certainly poems that could go both ways. But art is a weird environment: It's not necessarily a piece in and of itself that creates human interest in there. I'm sure there are a lot of `great poems` out there that nobody cares about because they weren't written by some great poet.
- bhickey 11y ago
  (a) is indeed written by a machine, (b) is excerpted from a famously awful poem, and (c) is an unintentional haiku found in the New York Times. I agree with what you're saying about art, but it's far afield of the point thewarrior was making.
eli_gottlieb 11y ago
> Why do you believe this? I have some familiarity with natural language processing and think that generating passable computer poetry is a fairly rudimentary task. Tell your model about meter, tell it about rhyme, learn word associations. Pick a language, give it every line of poetry you can find and hit go. What makes you think you could distinguish computational verse from something written by human hands? See, people always say that the trouble is getting your ML model to have some semantics to what it produces, to produce texts which somehow manage to genuinely correspond to the real world rather than just to the use of language. With poetry they might say, we don't just want something that fits the forms, but something with content, with soul. I say, yes, that's all true, and have you ever tried getting a randomly sampled human being to produce prose or poetry with meaningful content and soul? Untrained, we're possibly even worse than the neural nets!

visarga 11y ago

I appreciate your argument about ML lacking the embodied experience of humans and our feelings, but I think we can attack this problem. Anything we can get data on, can be learned. Emotions and physical sensations are accessible for collection in example datasets. ML is capable of detecting emotions in text, speech and human faces. In a similar way to learning to predict sentiment in reviews, we could learn to predict sentiment in photos and videos. On the other hand, ML can learn a thought vector (meaning) for every concept and how they relate. So it can learn both emotion and meaning. Combining the two, it could make poetry. It can also compose music pretty convincingly and make a painting in the style of Van Gogh from a simple picture, or make completely original style combinations (example: https://dreamscopeapp.com/ https://dreamscopeapp.com/). ML can't reach human level in music and painting yet, and surely not in poetry, but it might one day even surpass us. I think in the future humans will become the "children" of these artificial minds and one of their problems will be to explain their thoughts on our level. The tables will turn on us eventually.

kolinko 11y ago

You're making two assunptions: - that the process behind machine writing a poem would be the same as with human's - that one needs to understand what's being written to write well