That's fine, I'm not trying to convert anyone. I'm literally saying: here is something I liked, check it out. If you don't like it, that's fine too.

Robots aren't conscious; and they produce interesting poems. Can we start from there, please? — csalisbury

That sounds like the opening salvo to a discussion to me. More than like or dislike anyway. But fine, I’ll leave it there that the poems are interesting but no more than Burrough’s cut ups.

I think these poems don't hold up to poems by great poets, and I am sure I can break down why -- but I didn't think I needed to bring that up, because I thought that clearly wasn't the point of the OP. I'm beginning to think that everyone who is reflexively responding negatively couldn't show why great poems are different from AI poems and that accounts for the negative reaction. If it's a given, you don't sweat. If it isn't, there is some discomfort.

That sounds like the opening salvo to a discussion to me. More than like or dislike anyway. But fine, I’ll leave it there that the poems are interesting but no more than Burrough’s cut ups. — Brett

I've read both Burroughs' cut-ups and these poems; they are different.

Yes they’re different. I said not much more interesting than Burroughs.

Sure, but you could explain how a Burrough's cut-up sentence was produced, while you couldn't explain how the AI poem was produced. What is the point of comparing the poems to Burroughs cut-ups? If you can clearly, ingenuously, explain what prompted that particular comparison, I think the point is immediately clear. But you have to be clear and honest.

What’s the point of AI producing poetry, no matter how good it might appear to be? Is the success of AI going to be that it’s capable of imitating human capabilities? — Brett

Yes, but it's the greatest hits of human capability. It needs to be remembered these are gadgets in a strong sense. The use of a gadget is whatever you use it for. You could in theory have an endless amount of apps for generating specific material.
An advocate may argue, there is no one writing 12th century poetry in Portuguese so why not relegate it to an app, which is probably true.
But then there is a danger of becoming too mindless. On some level, the experts know it's just a piece of software even if they can't sell it as such. It takes millions upon millions of pics to train an AI (and sometimes failing at that: https://boingboing.net/2018/01/11/gorilla-chimp-monkey-unpersone.html)
whereas a child needs one image. That is a Zebra, job done. There aren't children's books with 4 million zebras in to ensure they understand what they are.

What brought Burrough’s cut-ups to mind and the comparison was your post that said “ The AI was fed poems from many times and genre ... “

So my impression was that the AI had juxtaposed these different lines. (I don’t know if they were fed lines, word or phrases), that the juxtaposing then created interesting stanzas that then generated interesting ideas.

The Oval Office will be
Full of people you never knew.
And you’ll wish you had been born
Another boy or girl.
They will be, what, pre-adults now?
Have you ever seen so many P.M.s?

So, like Burroughs, pre existing material was juxtaposed to create new material in the sense of making it new or creating unexpected, even original, connections, etc.

I don’t see anything wrong with that, except that the writer is still required to keep driving the work in a specific direction.
What drives the direction of AI? If there’s no sense of direction then it’s merely interesting to watch it juggle words and lines.
Of course we can take what we want from it, just like we do with poetry, if you go along with literary theory views the reader bringing meaning to the work, not the author.

I personally find that idea interesting, to a degree. And in that sense, if the reader-response theory wins out then the AI is the next Shakespeare.

You could in theory have an endless amount of apps for generating specific material. — Forgottenticket

That’s a very interesting point. There are many writers I like reading who are long dead. I’ve read all their books and can’t find any books by authors that come close enough to satisfy me. I might be very interested if AI could produce work just like theirs, that ticks all the boxes that does it for me.

There's a thing that text prediction algorithms need to do that makes the full breadth of theme and imagery difficult for them.

To a model, a poem is basically a time series of line index numbers; something like the following story;

Make a poem, model!

1. I have generated line 1 now.
2. Given that line 1 is "I have generated line one now" What should I generate here? I know, I'll generate this.
3. Given that line 1 and 2 are as written, what should I generate here? I know, I'll generate this.
...

The dependence of a line upon the history of lines before it is called serial dependence. How far back in line number a model can go from the previous one while generating semantically/thematically/image relevant content is constrained by how much serial dependence the model can ape.

Something like a theme for a poem, like desolation in "A Measure of Desolation" by Ursula Le Guinn:

Again and again the landwind blows,
sending back the rain
to the house of the rain.

Seeking, seeking, the heron goes
longlegged from creek
to thirsty creek.

They cry and cry, the windblown crows
across the sky,
the bare clear sky.

From land to land the dry wind blows
the thin dry sand
from the house of sand.

requires that the model "think about" all the previous steps and devices to embed the consistency/relatedness of each verse to each other. There're multiple serial dependences in it; the first line of each verse is predictive of all the others' structure, the second line in each verse is predictive of all the others' structure. The middle line of each verse is a descriptive elaboration of the bit after the double space in each first line and a rhyme with the last word before the double space.

Dealing with multiple serial dependences to create a consistent image dealing with a theme is hard. Before we start talking about how much branching/association there has to be to grok the associations of semantic content that make imagery work - there's word to word serial dependence too, as well as sentence to sentence serial dependence, motif motif serial dependence, device device serial dependence...

It's amazing that the model can do so much of this, I'm just hypothesizing that a poem dealing with a theme is a particularly hard task in terms of serial dependence.

Robots aren't conscious; and they produce interesting poems. Can we start from there, please? No one is demeaning actual poets, including me; but almost all comments seem to be defending poetry as real against the robots. Yeah, I agree, but I never for a second felt threatened by them - why do so many people here? — csalisbury

Yes, there is a bit of discomfort that I feel in seeing noteworthy aesthetics originated by a non-sentient entity with a significant degree of regularity. An aesthetics which I grant being present in the technical know-how that the AI acquired.

As to the OP’s presentation, I’m grateful for it. Thanks. Good to know such things. As to philosophical comments on what the OP presents, I think one underlying issue is the nature of what art is:

A sunrise can, on some occasions, be beautiful. Does this of itself make the sunrise art? I’ll argue that if and only if one assumes that the sunrise was the intentional creation of one or more sentient originators, one can then hold that the sunrise is art. If no such assumption is made, the aesthetics of the sunrise then does not get interpreted as an instantiation of art. Same will then apply, for example, to some dog accidentally kicking over a number of paint buckets with the result of an aesthetic arrangement of colors. Since no aesthetics was intended, the dog did not create a work of art.

On the other hand, I can mason a brick wall without any intention of conveying anything by it. I would be a sentient originator of the brick wall but, because there was nothing I intended to communicate by it, it would not be to me an instantiation of art. However, after the brick wall is finished, I then place a loose random brick on top of it with the explicit intention of conveying “the precarious nature of abstraction is always supported by a solid substratum of concreteness”. (Why not, right?) This brick I subsequently place on top the brick wall I built is now to me an artistic manifestation, thought the brick wall is not. And it will to me be art irrespective of how good an art piece it is. (In this case, not that great.)

In these scenarios, the difference between that which is and is not art is the occurrence of a sentient being’s intention to convey meaning via that which they originate. To this effect, even ordinary conversations can be deemed to be an art form. And, as goes without saying, poetry is a form of art.

Because the AI program is not sentient, it lacks this intention of conveying meaning. Yet what it produces mimics the outcomes of just such an intention.

There is discomfort in this, for me at least. A bit off topic relative to the OP, but pertinent to the issue of discomfort: AI chat bots are known to exist. They’re not perfect, but are improving by the day. The Orwellian implications of, for example, the degree of propaganda that can occur on social media platforms as a result … are for me unpleasant to think about. And present day AI’s ability to mimic human poetry to me points in such directions.

At any rate, with this I’m just trying to convey where my discomfort is coming from.

As to it being real poetry, verses mimicked poetry - and as per my previous examples - can it be real if it was not originated via a sentient being’s intention to express something of meaning?

Artistic inspiration invariably refers to an external source of information and stimulation for the artist from where they get their ideas. A poetry generating algorithm is just another source of artistic inspiration, whose output a poet will tweak to his own satisfaction. This is little different to the use of 'story dice' as random source of inspiration.

In my opinion, AI art does not represent a paradigm shift in terms of the meaning of art, rather it reveals a significant part of the preexistent process called artistic inspiration and democratises it. Art algorithms are really about increasing the economy of scale of art production and they accelerate the pace of art revolutions.

Art algorithms ... accelerate the pace of art revolutions. — sime

Can you give evidence for that?

sime

Art algorithms ... accelerate the pace of art revolutions.
— sime

Can you give evidence for that? — Brett

Art has moved in tandem with the accelerating technological trajectory that began with the invention of electronics in the Victorian era. Without semi-autonomous content creation tools, the present creation of vast and open virtual worlds wouldn't be possible. Perhaps we could call the potential algorithmic output of content-creation tools "meta art" in comprising a distribution over art objects, but it is still 'art' in the traditional sense in being ultimately shaped by the vision of it's users who program it, feed data to it, and tweak it's responses. And these tools also complement and fuel the need for traditional artists who produce hand-crafted content, as for example in virtual world content creation where there are neither enough algorithms nor artists to produce the infinite amount of diverse content required.

Art algorithms ... accelerate the pace of art revolutions. — sime

That’s an interesting post, but I don’t see it as evidence of algorithms accelerating the pace of art revolutions. I don’t see an art revolution that’s the result of algorithms.

Without semi-autonomous content creation tools, the present creation of vast and open virtual worlds wouldn't be possible. — sime

True.

Perhaps we could call the potential algorithmic output of content-creation tools "meta art" — sime

We could, but is it? And it’s still only potential.

in comprising a distribution over art objects, but it is still 'art' in the traditional sense in being ultimately shaped by the vision of it's users who program it, feed data to it, and tweak it's responses. — sime

So it’s an output tool, like a painting on canvas or printed pages. But what is the vision in feeding AI words and lines from existing poems? There is no vision except to create what is now redefined as a poem, as art. There is no poet, only the programmer. The vision then becomes that of the reader, as in reader-response literary theory. There is no vision of the artist because the construction of the poem is random and the meaning accidental.

And these tools also complement and fuel the need for traditional artists who produce hand-crafted content, as for example in virtual world content creation where there are neither enough algorithms nor artists to produce the infinite amount of diverse content required. — sime

But AI can only produce a work if it’s received the pre-existing work of traditional artists, because it cannot produce “the infinite amount of diverse content required.” In fact without the diverse content it would just be a tool waiting for someone to switch it on. Is that a revolution in art?

Poets making poems out of other poems counts as poetry; why not a machine doing the same thing.

Poet style is constrained by what they've learned about poetry; machines face the same constraints.

Poets associate words with each other to build resonant series of imagery; machines do the same thing, though maybe face scope issues - poets face scope issues too.

New work always relates to old work thematically or stylistically; poetry generated by an algorithm does so explicitly.

Poets wonder what to write next and the impenetrable to interpretation expressivity of their minds wonders what fits best for their decided purposes; machines have uncertainty about the next poem element and black box nigh on impenetrable decision making processes regarding improvised fit.

Poets improvise poem structure; machines do this too, just depends on their prompt.

Poets write poems when inspired and prompted by their experiences; machines write poems when prompted.

I can understand why people would be anxious about all this.

I can understand why people would be anxious about all this. — fdrake

Anxious about what?

Anxious about what? — Brett

There being so much data to feed gigantic models that they're getting extremely close to being functionally indistinguishable from human conduct in limited domains. The all too rapid and usually hidden encroachment of machine learning techniques (faciliated by panvasive surveillance and automated tabulation of all human experience) into the folk thought ineluctable freedoms of our souls.

Anxious about what? — Brett

There being so much data to feed gigantic models that they're getting extremely close to being functionally indistinguishable from human conduct in limited domains. The all too rapid and usually hidden encroachment of machine learning techniques (faciliated by panvasive surveillance and automated tabulation of all human experience) into the folk thought ineluctable freedoms of our souls. — fdrake

Speaking for myself, you’re projecting metaphysical issues way too much into this. When and if a technological singularity will occur, whatever sentient beings have that distinguishes them from rocks will be had by Strong AI as well in equal measure. Be this the “ineluctable freedoms of souls” or something else. Thereby making whatever metaphysical issue one has qualms about mute in this respect. And besides, anxiety is not it. Anxiety is reserved for more pertinent things.

Again speaking for myself, the issue I was mentioning earlier is that of non-autonomous, non-sentient, decoys which mimic the autonomous and sentient behavior of humans. It’s fathomable that these can be built by humans with big loads of cash and programed so as to manipulate the other humans into beliefs that serve the short sighted interests of those who spend the money on building these decoys. Don’t know about you, but I don’t like the notion of living in (or, more likely, of today’s children growing up to live in) a world of vastly greater misinformation, misinformation that is propagated by AI decoys to boot. As one example, put enough chat bots on the internet which argue for Earth being flat and you’ll have an increased number of voters who vote on the conviction that Earth is in fact flat. This being a very innocuous example.

Speaking for myself, you’re projecting metaphysical issues way too much into this. When and if a technological singularity will occur, whatever sentient beings have that distinguishes them from rocks will be had by Strong AI as well in equal measure. Be this the “ineluctable freedoms of souls” or something else. Thereby making whatever metaphysical issue one has qualms about mute in this respect. And besides, anxiety is not it. Anxiety is reserved for more pertinent things. — javra

It's sort of off topic, but those mechanisms of behavioural modification are already in place. Doesn't need the singularity, just needs flexible enough algorithms (already there), almost total surveillance (already there) and experimental levers to pull (see online contagion experiments done by Facebook, Google's ad work with Pokemon go); your fitbit data is already being fed through a machine learning + insurance risk algorithm and sold on to calibrate health insurance costs.

The most salient issue for the thread in terms of the above algorithms is the type of training data and the model flexibility I think; it's a text processor and generator, it's trained on text and outputs text. If it had a resevoir of behavioural data and a linking model, it'd probably be able to ape the human experience informed aspects of poetry (notice this is a placeholder with no ascribed content) much better; it'd be able to link personal experience to words and generalise from it, just not "its own" experience.

Another angle is that an intelligent poem writer isn't necessarily a human; maybe the next version will have the kind of relationship to poetry that eagles have with their wings.

Anxiety is reserved for more pertinent things. — javra

As in entering a genuinely anxious state? Yeah. As in cordoning off poetry from machine functionality? Nah; that's super prevalent in the thread for mostly unargued reasons. Ego defense mechanism metaphysics everywhere.

If we start conceptualising poetry as a task a human can do, rather than an expression of a necessarily human subjectivity... What remains if you abstract the poet from the poem?

Probably something like; series of evocative expressions, rhymes and meters, associations of words, grammatical structure and how to break it for effect... The model maybe wouldn't write in expressions; thought of as a productive relation between idea and pen; it would sample the next effective device, the next effective word choice, from the corpus of its weighted associations. By the looks of it it can already do consistency of style and create effective contrast with just text input...

Could it have done it without the text input? Seems like a relevant question to draw a line between machine poets and human poets; but can a human write a poem without a much broader input? Doesn't a human need even more competences to write evocative, memorable, poignant poetry? If you give an AI a sample of poetry, that's a sample of what humans care about just like an individual's life is a sample of cares.

So it’s an output tool, like a painting on canvas or printed pages. But what is the vision in feeding AI words and lines from existing poems? There is no vision except to create what is now redefined as a poem, as art. There is no poet, only the programmer. The vision then becomes that of the reader, as in reader-response literary theory. There is no vision of the artist because the construction of the poem is random and the meaning accidental. — Brett

Modern creative algorithms are accessible to anyone, and have many tuneable parameters that the artist himself can control in accordance with his artistic vision, to influence the style, subject etc . But perhaps this is tangential to the discussion.

Is the creative process a self-directed and inwardly driven process determined by the artistic foresight of the visionary artist? or is it a blind and partly external environment-driven process, whereupon external stimuli bring about artful stimulus-responses in a person said to have artistic temperament?

I suspect that Cartesian minded internalists who believe in the former might have a harder time reconciling AI and the arts compared to externalists who view the artist as a shambolic director of
haphazard external processes.

There are many writers I like reading who are long dead. I’ve read all their books and can’t find any books by authors that come close enough to satisfy me. I might be very interested if AI could produce work just like theirs, that ticks all the boxes that does it for me. — Brett

I think it's with box ticking mistakes are often made. I've seen a screenwriter produce scripts using whiteboards of what works in other episodes and make a complete mess. So then we make it about feeding it more data, the behavioral information of each author, the predictions based on past decisions ect, perhaps social events attended or what they will/would have/could have attend.
And perhaps even random numbers to account for Brownian motion that may occur during the production stage of each chapter. Do we have new books by the author?
The article csalisbury posted on page 2 has vanished because the author was doxed (which is a very human threat and a separate though related issue). But before it did, it showed the algorithm was progressing in a logarithmic fashion. I think there is an argument to make that it may break or come to a crawl.

Is it, as it appears, that the final proof of AI abilities is in producing art, like poetry? Is that what we’re using as proof that it’s equal to the human experience, not just the logical aspects, which AI supposedly excels at, but the actual human experience of living. If that’s the case then it suggests something very particular about art and being human, a benchmark for AI to crest.

If that’s the case does it also mean that logic is not what’s so special about being human?

logic is not what’s so special about being human? — Brett

That was actually the meaning behind my first post in this thread. We can easily move the goalposts to art being away from its traditional definition. Aesthetics is often considered being related to logic.

Anyway, a couple of queries:

1: Is AI generating new logic? if so then shouldn't we go into another short lockdown so it can have the power and equipment to get to Type III civilization theorems in a very short space of time.
2: If it's just machines making use of logic, then every single boss from every video game would count because they continually defeat humans at logic every day. The Turing test for logic was completed on the first defeat.

This?

Does AI undo its legitimacy as art? — Forgottenticket

Do you mean that art challenges the legitimacy of AI?

I find it difficult to be sure what your posts are getting at.

Do you mean that by moving the goalposts about art we allow AI to be creative? And that aesthetics, being related to logic, allows AI to legitimately take part in the creative act. My feeling, anyway, about aesthetics is that it’s the non creative‘s door into art. They take part in art by applying a slide rule to what others do instinctively.

What’s new logic? Is that a logic that lay dormant until we found it? Can there be another sort of logic? I don’t know, I’m not even sure if that’s what you were getting at,

2: If it's just machines making use of logic, then every single boss from every video game would count because they continually defeat humans at logic every day. The Turing test for logic was completed on the first defeat. — Forgottenticket

So the AI defeats us in logic, right? And so logic in humans, so what? Move on, nothing special here to look at.

It could. Neither intentionally nor unintentionally but rather as the result of an algorithm programmed by logic or otherwise "to be logical". But. If that higher "new" logic isn't quantified by manual human input to be it's own logic rather "more logical" than (algorithmically favored over) a similarly assembled phrase using the same base logic and algorithm it will be otherwise ignored and would only happen as random chance.

I'm thinking all these algorithms when assembling phrases or ideas have a % of accuracy it assesses when building larger forms. Obviously the earliest versions of any kind of program were more gibberish than coherence. But with human input ie. programming, data processing, probability and assessment it improved to what it is now.

It's sort of off topic, but those mechanisms of behavioural modification are already in place. [...] it'd be able to link personal experience to words and generalise from it, just not "its own" experience. — fdrake

Most of what you say here goes without saying, but it misses the point I was making: not its own experience on account of it not being sentient, or conscious, or aware, or cognizant. That is, not until Strong AI comes about, if it ever will.

As in cordoning off poetry from machine functionality? Nah; that's super prevalent in the thread for mostly unargued reasons. — fdrake

Hmm, I’ve given an opening argument for why art, poetry included, necessitates some sentient being’s intention to convey meaning here - thereby precluding Weak AI from being creators of (authentic) poetry, for they lack sentience and thereby sentient intentions to convey meaning. These un-emotional arguments have so far not been addressed, and so have not been debunked.

For the record, I’m more than open to learning how artistic manifestations can be denoted as occurring in the absence of sentient intentions to convey meaning, this without wreaking havoc on commonsense understandings of what constitutes art.

Ego defense mechanism metaphysics everywhere. — fdrake

Seems like this is true for all sides of this issue, at least in relation to some.

Apologies, I got a bit prickly, it’s a little silly, in retrospect, to post on a philosophy forum about something, and then object to it generating philosophical conversation.

But yeah, I think what we’d want to defend is why we read and write poetry - and if we admit an unfeeling and unthinking computer as generating poems equal in value to those generate by human poets, we begin to lose why we write and read poetry in the first place. So, I agree that it's important to focus on what we value in human poems versus AI poems. I find, personally, that I like poems that help me reflect on aspects of emotional and spiritual life in ways I wouldn't be able to otherwise, on the one hand, and that also let me admire style and mastery, on the other. Mastery implies something hard-won, the result of a long struggle, and so, to me, can be admirable only in other sentient (conscious/feeling/spiritual) beings. I think we are basically on the same page here.

I wouldn't necessarily agree with the intent to convey meaning, but that may just be a matter of semantics. The reason is something close to what I belive James Baldwin to be talking about here (in an interview with Paris Review:

"When you are standing in the pulpit, you must sound as though you know what you’re talking about. When you’re writing, you’re trying to find out something which you don’t know. The whole language of writing for me is finding out what you don’t want to know, what you don’t want to find out. But something forces you to anyway."

I think the best art is an articulation which, in being articulated, reveals both to the reader and the writer its meaning - its not a message intended ahead of time. That's my personal feeling anyway - obviously 'conceptual art' is very much along the lines of the brick on top of the wall. I don't want to try to draw lines demarcating art and non-art, but what I'm personally drawn to, what I Value is the kind of thing Baldwin is talking about - and I agree AI doesn't do that.

But, the poems are pretty and have their own kind value, yes like a sunset, but also in their own highly novel way.

I wouldn't necessarily agree with the intent to convey meaning, but that may just be a matter of semantics. The reason is something close to what I belive James Baldwin to be talking about here (in an interview with Paris Review:

"When you are standing in the pulpit, you must sound as though you know what you’re talking about. When you’re writing, you’re trying to find out something which you don’t know. The whole language of writing for me is finding out what you don’t want to know, what you don’t want to find out. But something forces you to anyway."

I think the best art is an articulation which, in being articulated, reveals both to the reader and the writer its meaning - its not a message intended ahead of time. — csalisbury

Very nice quote. Yes, approaching this from my own experiences, I agree with what you address being an important aspect to the process of art creation. To use my own words and understandings here, artistic manifestation is as much a conscious as it is an unconscious goal oriented (hence, intended) processes. In my own experience at making (sometimes crappy) art, the conscious self chooses - if only emotively - between what the unconscious self throws up at it while at first having, maybe, a generalized intention of producing X; and, at the end of the process, what one ends up with is outcome Y - which resembles X only in the most basic structural ways, but is in many ways utterly different, and unforeseen at the very commencement. And this the final product, when it receives the last stamp of approval by its creator, so to speak, reveals meaning to the conscious creator as well as communicating some basic aspect of what the conscious creator intended. Revealed meaning that on occasion can leave the conscious creator in awe in terms of what is gained and learned from what was created (rather than from the egoistic sense of what “I” made). But, notwithstanding, for me the conscious self’s goal-oriented decisions between the alternatives which the unconscious self presents nevertheless play a role throughout. And in this, I'd like to think that the conscious self chooses the meaning which the final product conveys - even if the meaning is only that of a particular aesthetic devoid of conceptual content. .

Its nice to have such complex aspects of art creation brought up and discussed.