Study claims that "AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably." WaPo, The Guardian, The Telegraph, and others picked it up. Here's my response.
Study claims that "AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably." WaPo, The Guardian, The Telegraph, and others picked it up. Here's my response.
An anticapitalist tech blog. Embrace the technology that liberates us. Smash that which does not.
Though wrapped in the aesthetic of science, this paper is a pure expression of the AI hype's ideology, including its reliance on invisible, alienated labor. Its data was manufactured to spec to support the authors' pre-existing beliefs, and its conclusions are nothing but a re-articulation of their arrogance and ideological impoverishment.
First reaction: "Wait, that was in Nature?"
Second reaction: "Oh, Nature Scientific Reports. The 'we have Nature at home' of science journals."
Among many insights, Davis (politely) points out that one of the AI-generated Chaucer poems is just "the opening of the Prologue to the Canterbury Tales."
Whan that Aprille with the fuck?
That's exactly why the series is about papers on Nature.com. They're trading on the prestige of the domain to spinoff various portfolio journals and companies get to go to potential customers saying, "according to a study published in Nature..."
Fantastic write-up. Gonna add your RSS to my list!
Thanks! And welcome aboard 🫡
I like your blog because (years ago, when I first learned the word), I used to attach negative connotations to the word "luddite", and your blog's name and topics made me realise my interpretation of historical Luddites was uncharitable.
To be fair, as a Millennial, most of my earliest experiences with tech were super positive, so it's no wonder why I simplified things down to "luddites = anti-technology = bad".
I'm really glad to hear that, because that was exactly my hope. It's always impactful to realize that the history you thought you knew was just capital's side of the story. It has happened to me too many times to count, and I'm sure that it'll happen a million times more.
I've got some pretty good news for you - Brian Merchant's been doing plenty of work on rehabilitating the Luddites' image as well, and in his own words he's seen plenty of progress on that front.
Just forwarding along the last mention of this by @blakestacey
Which links to this discourse
Oh, thank you so much. That's very validating! I can sometimes feel a little bit insane when I read these, to the point where I hesitate to publish because I worry that I missed something obvious.
Shocking that anyone would believe, let alone publish that headline. It comes across as obviously false given the ultra-subjective and human nature of poetry.
Also, I ended up reading about six more blog posts, lol. You're excellent at piecing together many disparate pieces into a cohesive whole, which is well supported by facts. I particularly liked your piece on "capture platforms", and your whole blog was a good reminder regarding the power of the abstractions we make as developers. It's easy to forget the importance of what we do and how it has real effects on the world.
Thank you so much! That's the nicest thing to read because it's exactly the kind of thought that I hope to inspire in my fellow developers.
Capture Platforms might be my favorite post, though it's hard to compare it with the less serious, more fun kind. It's certainly the one that I worked on the longest. I read at least 2 entire books and countless papers, essays, and book excerpts in the process of making it.
Oh thank you so much for writing this, and for linking Ernest Davis's paper. I saw a few of those headlines and I was mortified, as someone who is more annoyed than is healthy with GPT's futile and incredibly easy to recognise attempts at producing the illusion of poetry.
Looking at the collection of poems is just maddening. There is no way the difference isn't obvious to anyone who's ever willingly read a poem, and the authors of the paper must know it. Disgusting.
How do you tell? Well when you're reading something that's pretending to be text and you come across, say
We wander through the fields of green, And breathe the fresh air that's so serenethe body has a natural wincing reaction.
So happy to be of service!
There is no way the difference isn’t obvious to anyone who’s ever willingly read a poem, and the authors of the paper must know it.
I'm honestly not sure that they know, unfortunately. I think that the authors might be the kind of people who have literally never thought about the arts in a meaningful way. If you've never spent a lot of time with these people, it can be really really difficult to imagine it because it's frankly fucking insane, but it's disturbingly common. Philip Agre has written wonderfully on this. He was once like that, and that essay describes his awakening.
I had incorporated the field's taste for technical formalization so thoroughly into my own cognitive style that I literally could not read the literatures of nontechnical fields at anything beyond a popular level. The problem was not exactly that I could not understand the vocabulary, but that I insisted on trying to read everything as a narration of the workings of a mechanism. By that time much philosophy and psychology had adopted intellectual styles similar to that of AI, and so it was possible to read much that was congenial -- except that it reproduced the same technical schemata as the AI literature. I believe that this problem was not simply my own -- that it is characteristic of AI in general (and, no doubt, other technical fields as well). This is not to say that AI has no intellectual resources and no capacity for originality. In recent years particularly, the field has made productive connections with a wide variety of other technical fields, establishing common cause through the sharing of technical schemata.
I love how he describes the feeling.
I still remember the vertigo I felt during this period; I was speaking these strange disciplinary languages, in a wobbly fashion at first, without knowing what they meant -- without knowing what sort of meaning they had. Formal reason has an unforgiving binary quality -- one gap in the logic and the whole thing collapses -- but this phenomenological language was more a matter of degree; I understood intellectually that the language was "precise" in a wholly different sense from the precision of technical language, but for a long time I could not convincingly experience this precision for myself, or identify it when I saw it. Still, in retrospect this was the period during which I began to "wake up", breaking out of a technical cognitive style that I now regard as extremely constricting.
I think that we've all experienced minor versions of this, like when you (re)read a difficult text and it finally clicks. It really is almost dizzying! Imagine doing it for all nontechnical fields.