Didactic poetry might come back. One of the consequences of computer creativity is lowering fixed costs of every sort, which can have very large elasticities and enable an even greater long tail of outputs.
My usual example is Vocaloid: you might think singers are a dime a dozen, and finding or being a singer cannot possibly be a bottleneck of any sort to would-be musicians; yet, turns out, if you can provide even terrible-sounding singing voice synthesizers, you unleash tens of millions of songs and create countless careers. (Even the singers can do pretty well out of it as they can cover songs or work with new composers to redo ‘professionally’, as it were, their prototypes.) The attempt to power through bottlenecks also has costs—I’ve mused this about single-creator works like Kill Six Billion Demons or A Practical Guide to Evil: the creators aspire to working in many genres or formats, like both poetry and fantasy epic, or Western comic and religious parable, but no one is gifted at everything, and it’s pretty clear which one the creator is best at and the attempt to be more of a Gesamtkunstwerk does not always help them. If they had bigger budgets, perhaps they could afford to hire a collaborator, but they don’t and they won’t. (Assuming they could even locate them. Who do you hire to write didactic poetry on graph theory? Seriously. Even if you’re a billionaire with unlimited budget, how do you find “a great, talented poet willing to write on that topic”, rather than simply take your money and do a poor job? Surely there is someone out there who can do it. But do you know who? I sure don’t. But I do know GPT-3 does surprisingly good math-themed poetry already, and it’ll do so 24⁄7 for as long as you want to pay the API bills, and that future models will only get better.)
But with the march of generative models in particular, both of those change. A composer may not be able to sing, but he can recognize what he wants and work with a Vocaloid well enough, and it’s far easier than working with a real singer. A multi-genre creator may be only really good at worldbuilding and narrative, and be unable to write decent poetry to sprinkle in his fictional world, but he can recognize a good sample of poetry if he plays around with enough generated samples on the necessary theme. And this goes for whatever others he might need: generative models are by nature generalists, and adding in additional genres or styles is no big deal.
The creator of the future can look much more like an editor than a creator, organizing gangs of APIs churning out dozens of samples until one proves adequate. (“Six lines, cleverly rhymed, and every word beginning with the letter ‘s’!” Now all in g! A sonnet, trochaic hexameter, about an old cyclotron who kept sixteen artificial mistresses, blue and radioactive, had four wings, three purple pavilions, two lacquered chests, each containing exactly one thousand medallions bearing the likeness of Czar Murdicog the Headless! Now all in the letter ‘a’. About the mating habits of the radioactive white-faced chinchilla. Not easy to compose, I assure you, but very funny.”) It only takes 1 person to then generate what they wanted to read.
Didactic poetry might come back. One of the consequences of computer creativity is lowering fixed costs of every sort, which can have very large elasticities and enable an even greater long tail of outputs.
My usual example is Vocaloid: you might think singers are a dime a dozen, and finding or being a singer cannot possibly be a bottleneck of any sort to would-be musicians; yet, turns out, if you can provide even terrible-sounding singing voice synthesizers, you unleash tens of millions of songs and create countless careers. (Even the singers can do pretty well out of it as they can cover songs or work with new composers to redo ‘professionally’, as it were, their prototypes.) The attempt to power through bottlenecks also has costs—I’ve mused this about single-creator works like Kill Six Billion Demons or A Practical Guide to Evil: the creators aspire to working in many genres or formats, like both poetry and fantasy epic, or Western comic and religious parable, but no one is gifted at everything, and it’s pretty clear which one the creator is best at and the attempt to be more of a Gesamtkunstwerk does not always help them. If they had bigger budgets, perhaps they could afford to hire a collaborator, but they don’t and they won’t. (Assuming they could even locate them. Who do you hire to write didactic poetry on graph theory? Seriously. Even if you’re a billionaire with unlimited budget, how do you find “a great, talented poet willing to write on that topic”, rather than simply take your money and do a poor job? Surely there is someone out there who can do it. But do you know who? I sure don’t. But I do know GPT-3 does surprisingly good math-themed poetry already, and it’ll do so 24⁄7 for as long as you want to pay the API bills, and that future models will only get better.)
But with the march of generative models in particular, both of those change. A composer may not be able to sing, but he can recognize what he wants and work with a Vocaloid well enough, and it’s far easier than working with a real singer. A multi-genre creator may be only really good at worldbuilding and narrative, and be unable to write decent poetry to sprinkle in his fictional world, but he can recognize a good sample of poetry if he plays around with enough generated samples on the necessary theme. And this goes for whatever others he might need: generative models are by nature generalists, and adding in additional genres or styles is no big deal.
The creator of the future can look much more like an editor than a creator, organizing gangs of APIs churning out dozens of samples until one proves adequate. (“Six lines, cleverly rhymed, and every word beginning with the letter ‘s’!” Now all in g! A sonnet, trochaic hexameter, about an old cyclotron who kept sixteen artificial mistresses, blue and radioactive, had four wings, three purple pavilions, two lacquered chests, each containing exactly one thousand medallions bearing the likeness of Czar Murdicog the Headless! Now all in the letter ‘a’. About the mating habits of the radioactive white-faced chinchilla. Not easy to compose, I assure you, but very funny.”) It only takes 1 person to then generate what they wanted to read.