Friday, January 27, 2023

Why Writers Shouldn't Fear AI


Chat GPT and related AI writing tools are poised to take the writing world by storm. Today, GPT-3 can generate the plot for a novel, complete with character arcs and plot twists. Future iterations may create the plot for a novel based on a user's input, then use the plot it generated to create an entire novel from scratch. All the user has to do is to clean up the prose.

AIs allow content creators to generate vast amounts of content quickly and cheaply. Whale consumers love quick and cheap content. They want a never-ending stream of dopamine, and will throw their money at those who can feed their desires. Businesses with deep pockets are poised to take maximum advantage of AI. They have the resources to license AIs, pair them with a good-enough human for quality control, and churn out oceans of content for their target audience. Prose, music, videos, every creative field will soon be flooded by AI-generated content. I expect to see this within the next five to ten years.

What does this mean for the future?

Mediocre creators will get a leg up. AI will give them the tools to quickly produce content beyond their current skill levels. They can focus on ideation and editing, allowing the AI to do the heavy lifting of producing content.

Content mills will enjoy explosive productivity. AIs will generate the bulk of their content for them. It synergises perfectly with their strategy of quantity over quality. They will hire freelancers and bottom-tier creatives to handle editing and quality control, those willing to work for minimum wage or less. They won't hire excellent creators except for specialist roles, such as artists or formatters, and even then, as soon as there is an AI for that, these creators will be replaced. AI allows content mills to dramatically reduce costs, translating into dramatic profits. They have no incentive to improve the quality of their output.

Higher-quality creators will be drowned in an ocean of homogenous goop.

Signs and Portents

We are already seeing the signs today. The dirty secret in the Japanese light novel industry is that many writers can't write. Editors scour webnovels to identify promising talent. They sign publishing contracts with promising writers, and request polished manuscripts. But those web novelists can't produce work up to par. With deadlines to meet, this forces the editors to do most of their rewrites themselves.

If the talent can't produce quality work, why did the editor sign the contract with them in the first place?

Because publishing is a business. Businesses want to make money. A publisher won't publish a book unless it thinks it can make a profit from it. A highly popular webnovel comes with a ready-made audience, primed to buy the book when it comes out. That's why the publisher signed the contract: not because the writer is talented, but because his product offers low risk and high reward. If the publisher cared about quality, they wouldn't have signed contracts with such writers in the first place.

Immerse yourself in Japanese media and you will see the same tropes over and over again. Ordinary Japanese person transported to another world and gains superpowers. Fantasy monsters drawn from Western mythology. Characters who fit stereotypes to a T. The marketplace does not prize innovation. It prizes familiarity. It rewards those who use familiar tropes but in slightly novel ways. This leads to content homogeneity, in which IPs resemble each other so closely they become almost indistinguishable.

The Western equivalent is writing to market. Writers are advised to study the bestsellers in their field, adopt the tropes they use, and place their own spin on things. This is especially common in romance, LitRPG and cultivation—among others. As with Japanese entertainment, stories in those genres have become indistinguishable goop.

Whales love goop.

They love to see the same tropes over and over and over again. They swallow entire books in days, even hours, then gulp down even more. Craft and care mean little to them. They want more and more and more of the same, as quickly and as cheaply as possible. Those who cater to whales can become whales themselves, growing fat off the profits.

AI writing tools empower people to write for whales.

AI takes input and transforms it into the desired output. The user first feeds the AI prompts, telling it what to create. Armed with these prompts, the AI hunts for samples of content similar to these prompts, then uses this data to generate optimised content. Here's how it works:

A writer feeds the AI a bunch of tropes and a plot outline. The AI studies a large collection of books with similar tropes and plots. Then it generates a novel based on the tropes and outline, reflecting the archetypes, emotions, story beats, tropes and plots used by the books it studied.

This process is identical to writing to market, except that no humans are involved—and the first draft is produced much, much faster. Instead of writing a novel in a month, you can produce a novel in a day. Even hours. All that's left is for a human to go over the draft to ensure it makes sense to a reader, and then the novel will be good to go.

Editing a story is much faster than writing it. If a content mill is capable of publishing a novel a month, it can now publish a novel a week. If it can publish a novel a week, it can publish a novel every two or three days. Or even every day.

This process is already happening in the modern world. Content mills dominate the Amazon bestseller lists. Generic stories top the Western and Asian charts. All AI will do is do exactly what it has been programmed to do: to optimise the process of creation, allowing humans to handle higher-order business. And when the business is based on minimising costs, there will be a sea change in the industry.

Who Will Be Hit Hardest?

Let's start with those who will benefit the most from AI writing tools. At the very top will, of course, be the owners of the content mills. They don't have to get their hands dirty. They can license the AI, oversee the human partners, handle advertising and promotion, and perhaps ideate storylines. There's no need for them to spend more than a half hour on the business on creation. They can focus their time and energy on attracting and retaining customers, and refining business strategy. They will outsource the labour to machines and low-cost labourers. They will free up time for themselves, and rake in the dough.

Those at the bottom will also benefit the most. These are the gig workers and the freelancers with passable language skills, who offer quick content at the cheapest rates. They are the ones who will be hired by the content mills to partner with the AI. They get to enjoy a living wage while doing less work.

These AI partners will not necessarily be native speakers of the language. They will simply need a sufficient command of the language to meet the bar set by the content mills. They will be Pakistanis, Indians, Sri Lankans, Arabs and Africans who studied English as a second or third language and work over the Internet. They will ask for salaries far below any minimum wage set by developed countries. These salaries allow the content mills to save costs, while still allowing the workers to live like barons in their home countries. Native English speakers will have to find some other way to differentiate themselves, or else resort to geoarbitrage. They may even have to resort to working for content mills that seek to compete by offering slightly better quality works.

Consequently, for now, I think the East Asian publishing industry won't be affected as much... yet. Chinese, Japanese and Korean are notoriously opaque languages. It is easier for a content mill in those places to hire a minimum wage worker in their native country than to look for a foreigner with adequate language skills who asks for a lower salary. Foreigners with high-level language skills (or, rather, sufficient language skills) to work on novels may be able to make a living as translators instead—at least until future iterations of machine translation software arrive.

Specialists like artists and formatters will see a boom in demand for their services. More manuscripts equals more jobs available. But with content mills driving this demand, it will be a race to the bottom. The mills don't want excellence, only good enough. They will hire a Nigerian for $10 instead of an American for $500 to handle their formatting if they can. With tools like Atticus and Vellum, book formatting is quick and easy—easy enough that they might simply do it in-house and skip the middleman altogether. Content mills will only hire human artists only as long as AI art generators continue to produce figures with too many eyes and fingers, landscapes with no sense of proportion, and other such quirks. The moment an AI art generator overcomes the uncanny valley effect, demand for artists from content mills will drop dramatically.

Excellent writers will be hard hit by the coming of a next-generation Chat GPT. They produce outstanding work, but that work will be drowned in an ocean of goop. The marketplace is indifferent to quality; the current state of the market demonstrates it. Whales love cheap and fast and good enough books; they do not value high-quality work. Projected into the future, there is no reason to think that excellence alone will attract a sufficiently large audience for a writer to sustain himself. It will become even harder for customers to find their work. It could discourage them from producing more work altogether. But they won't be the ones hardest hit by AI writing tools.

Middling writers will.

These writers occupy the deadly ground between the mediocre and the magnificent. Their work isn't that much better than the stuff produced by the content mills, so they cannot compete for jobs at those mills with low-salary Pakistanis who can do the same work, but cheaper. They are likely to be native English speakers from developed countries, and so even if they could get a job at a content mill, it would barely be enough to pay the bills. They cannot distinguish themselves by quality the way the excellent writers can. Without the ability to compete through cost or quality, where does that leave them?


Raising the Bar

Chat GPT will not replace humans. Instead, it will widen the skill gap between the mediocre and the excellent—and raise the bar for middling writers to overcome.

Competition in the market is fierce. Writing is no different. To distinguish yourself from the competition, you need a unique selling point. Examples of USPs include: high quality, low price, reliability, fast output, versatility.

An AI offers a powerful stack of USPs. It is fast; it can produce content much faster than humans ever can. It is reliable: it won't flake out so long as the machines are properly maintained. It is versatile: it can produce everything from novels to songs to poetry. It is cheap: it saves license holders the cost of hiring skilled writers. This is a stack few humans possess, never mind can compete with.

So don't.

Don't participate in a competition you know that you will lose. Instead, go to where there is no competition. Don't compete against an AI's strengths. Do what it cannot do.

An AI is not sentient. It takes in data and produces optimised output based on that data. That's all it does. It has no notion of beauty, reality, or even common sense. It can sample huge volumes of data and produce content at a rapid pace, but it needs a human to ensure that the output makes sense to a human.

An oft-heard complaint in the Japanese entertainment industry is that content is produced by otaku with no real life experience. Characters are nothing more than stereotypes. Relationships are fake, weird or outright toxic. Movements look unnatural, and thought processes are unrealistic. When the AI revolution hits, we will see a lot more of this. Someone who can create a work grounded in reality will immediately stand out from the herd.

An AI does not understand cultural nuance. Neither does a minimum-wage foreigner. It may be able to articulate the difference between temee and anata in the Japanese language, but it takes a human to imbue evolving personalities and relationships with human feeling, signaled by the word the characters use for 'you'. An AI can tell you all about the Chinese custom of giving red packets during Chinese New Year; it cannot explain to you why someone would joke about giving $44.44 to someone, or the context that would prompt someone to make that joke to begin with. All an AI knows about culture is what it can sample from open source texts; there is a huge disconnect between what an AI can parse and what an insider has lived.

This becomes especially prominent in English language works featuring characters who speak in multiple languages. The prose is written in English, but here and there are snippets in foreign languages. When is the right moment to write something in a foreign language? Should the text be explained or left untranslated? When should you not write in that language? Does using a foreign language carry greater cultural nuance than simply writing it in English? This is a judgment call only humans can make.

An AI does not know human psychology. It can assemble a psychology textbook from other psychology texts. It does not mean that it can use this information to craft believable characters. If the tropes it trained on call for outlandish emotional reactions or robotic stoicism, that is what it will produce. Having no conception of the human mind, it does not know how to produce well-rounded characters without human input. A low-wage AI trainer won't care about rounded characters; he only cares about ensuring that his assigned texts meets the minimum quality standards.

Trope-driven books appeal to whales. Whales want all the tropes they are used to. But before the coming of Kindle Unlimited, the stories written by bestsellers were driven by characters. By placing characters as the drivers of the story, the story becomes focused on the human experience. Stories that place tropes in the driver's seat are all about the tropes and nothing more. The latter becomes homogenous goop. The former become memorable epics.

AIs cannot think laterally. They merely sample huge quantities of data and produce output with an optimised concentration of the critical components of that data. They cannot tell you why a martial artist may take up dancing as a hobby, the connection between bugei and bundo, whether there is a connection between magic and quantum physics and perception. Humans have to make that inference—and a low-wage AI partner will only care about hitting all the tropes he is supposed to hit.

Not many humans can do this either. This is the measure of true genius. When tasked with identifying multiple uses for a hammer, someone with an IQ of 180 may only be able to think of it as a tool for hitting things with, whereas someone with an IQ of 160 may immediately brainstorm a dozen different ideas. True innovation lies not in repeating the same old thing over and over again, but in finding novel ways to logically connect seemingly disparate fields. Apple melded Zen with computing, Japanese light novels incorporated manga artwork with pulp-style writing, J. R. R. Tolkien united his studies into language and European mythology within a Christian framework to create a fantasy epic that lasts into the ages.

These are just some ways a writer can use to compete against the coming of the AI. From a craft perspective, all an AI can do is to set a bar for standards, and not a particularly high bar at that. When stories all look and feel alike, when they are all equally bland, a distinctive story crafted to high standards immediately stands out.

This is not easy. If anything, it is extremely risky. The easy option is to license the next Great Automatic Grammatizator and have it churn out books like everyone else, then sit back and rake in the dough. It's simply an accelerated version of writing to market, and writing to market sells. The only people who would do this possess a rare combination of traits: risk appetite, writing skills, disdain for wealth and/or machine writing, and a stack of USPs that is uniquely their own. And by that very rarity, they shall be known by those who encounter their works.

If you are a middling creator, now is the decision point. Will you hone your talents to the razor's edge? Or will you consign yourself to join the masses of AI trainers? Do you aspire to heights of creative genius? Or are you content with a regular salary producing mass-produced goop? That is a question only you can answer.

As for myself, I spent my entire writing career distinguishing myself from the crowd. No one else brings the unique combination of elements I do to my stories. No one else can. This blend of deep worldbuilding, authentic martialism, cultural nuance, and metaphysical experience is entirely my own. No AI will be able to do what I can do. An AI with a human partner can copy the tropes and characters and story beats, they may even be able to copy the style and the vocabulary and the cadence, but they will never be able to transmit the essence of what I do to someone else, an essence that can only be grasped by direct personal experience. And that is by design.

In the face of the machine ascendent, become more human.

An AI wouldn't be able to write this story.

No comments:

Post a Comment