There’s a lot of seeming panic going on out there in creative person land regarding what is colloquially being referred to as “AI”, Artificial Intelligence, and much of it unnecessary. And that’s because most people do not seem to have a handle on what these programs do, how they go about doing what they do, and how it is going to affect their futures.
I recognize that a lot of the negative verbiage, the doom, and gloom, is something in the vein of a Bradburian extrapolation as a warning, but if we want to effectively address the impact(s) that these emerging technologies are going to have (it’s inevitable), then we have to really understand what they’re doing and where they might go. You can’t fight unreality. You’ve got to know your “enemy”, and it’s pretty clear that most of us don’t.
I used the word “inevitable”. The advent of these technologies, their continued use, their continued improvement, their growing penetration of the marketplace, and their expansion into other areas of human endeavor beyond art, text, voice, and video, is going to happen no matter how many people protest against it, refuse to engage with it. These genies are not going back into their bottles. And that’s because their development and use is being driven by economic imperatives.
Go tell any business owner with employees that you have a technology that will allow them to eliminate human workers from the mix and that it can do so without any negative impacts on their product or services and watch the salivating begin. Bring a towel, you’ll probably need it.
Offer businesses the opportunity to do whatever it is they do without having to deal with interpersonal BS, without having to offer a health care program, without having to cut payroll checks…with no need for performance bonuses or having to underwrite a company picnic…give them the ability to do whatever it is they do with labor becoming a fixed and reducing cost and, so long as their bottom line is not impacted (in all likelihood it will actually improve), and they’ll take it.
Business costs associated with labor are typically pegged at 20 to 35 percent and can reach as much as 70%. Imagine the impact that would have on your own creative business – royalty payments one-fifth to a third better than they are now, advances with an extra 20% tacked on, paintings selling for thirty percent more. It would have quite an impact. It might even push magazines back into the black.
So far as “business” is concerned, there is every reason for them to utilize these programs and to want to see them improve and expand because, if nothing else, they’re looking at the ability to minutely control one of the few remaining elements that remain largely uncontrollable – the behaviors of people. Anyone who has ever worked in a cube knows this. Anyone working in an Amazon warehouse knows this. Authors who can’t meet a deadline? No longer a problem. Artists whose vision doesn’t match the client’s expectations. No longer a problem. Voice actors who can’t get the accent right. No longer a problem.
Let me take a brief aside here to mention that the impetuous for these technologies may not have originally arisen as a desire to “replace creatives”, but certain businesses recognized that it fits right in with their desire to commodify creative expression. Who wants to invest dollars in something that might be successful, when the alternative is being able to produce exactly what the audience says it wants? As Le Guin has remarked, that’s not art, that’s a commodity, like a dish soap with your favorite scent. If the market changes and lavender becomes more popular than, say, pine, all they have to do is add a different artificial scent to the mix.
Imagine, for a second, that these technologies have improved a bit beyond where they are now, and remember, when it comes to issues of “quality”, we’re not talking about the quality as referenced by successful practitioners of these arts, we’re talking about the market’s acceptance of that quality. When the majority audience is essentially incapable of distinguishing good work from bad, (and you all know that’s true), just about anything that pushes their buttons is deemed acceptable. (Don’t think so? How many times have you seen something that the literati justifiably regard as trash become a best seller, when you’re still wondering how it got past a first read? Clearly, your metric for quality does not coincide with the market’s. You are evaluating the artistic expression. The market is evaluating its ability to sell.)
I’ll make one other side comment before getting to the nightmare scenario. I find it both amusing and confusing that members of the science fiction community seem to have taken a Luddite’s position on these technologies and I can’t fathom how those of us who have been in the business of imagining the future – in many cases this very future – and generally extolling its virtues most of the time, advocating for its implementation, is reacting the way it is.
Anyways. Here’s where things might go. A distributor of fiction recognizes that much of their audience prefers particular genres. They know so because they employ sophisticated data gathering. Now, instead of relying on human creators to fill that supply chain, they turn to writing programs. They hire writers to write “prompts” based on the audience feedback they’ve received. A Western needs to have horses and six guns, ten-gallon hats. The “prompters” are paid a wage, not a word rate, and get no royalties. The works go up for sale and, because the distributor is making a better margin on these AI works, their algorithms push them into the “Recommended” and “Other works you might like” displays, where, owing to the extra exposure, they start making it onto the best seller ranks. A bit later on and the “prompters” can be eliminated and works tailored specifically to the individual customer’s requests become a thing. AI Translators go to work and now that novel is available in any language the reader desires…not in six months, not in three months, but minutes after the request is made. You can get your cover illustration tailored at the same time as well…and an audiobook with full cast…
Are such works as good as that turned out by human beings? Probably not in the beginning. Does that matter? No. It does not. The marketplace only cares about offering works that will sell, and that is not synonymous with “quality”. (For the unconvinced, I’ll mention the McDonald’s analogy once again. “Most burgers sold around the entire world.” Does the volume make them the “best” burgers? Anyone who has eaten one knows the answer is “No”. But they still bought them.)
Since the presence of these technologies is inevitable, do we just lie down and take it? No, we don’t. But we do have to shift our goals. From prevention to utilization.
Why? Again, it’s the economic imperative. Those rejecting these technologies will soon find themselves in non-competitive positions, with steadily reducing market share and revenue, a negative feedback cycle. No doubt some will survive because of their opposition, but they’ll be relegated to a small niche with few, if any opportunities to expand their reach. (Just compare a newsstand iteration of Amazing Stories to a newsstand version of Amazing AI Stories. Lets say each has 120 pages of content. Each costs about a dollar to print in volume. Each has about 115,000 words of text and about 12 individual pieces, each published with a full page illustration (12 interiors). Cover art as well. At ten cents per word (a bit above SFWA minimum), the fiction content would cost about $11,500. The art work, about another $3500. Fifteen thousand dollars, while AAIS equivalent costs are $0. With everything else being equal (distribution, shipping/postage, etc) AAIS clearly has an advantage. Their cover price could be reduced to be more competitive, their promotional budget can get a boost every issue, their editor or publisher could get a raise. “If you want to experience the future, read fiction FROM the future – Amazing AI Stories”. Don’t think it would fly? Amazing Stories had its highest, or at least 2nd highest, circulation and subscription numbers when Ray Palmer sensationalized the Shaver Mysteries and all of the related UFO pablum, proving that the reading marketplace consists of far more idiots than it does science fiction readers. It always has and it always will.)
There are three aspects to these technologies that it is important to understand as we try to figure out ways to mitigate their effects.
The first is their relationship to intellectual property issues. We’ve had a ruling from the Copyright Office that states that works created by “AIs” can not receive copyright protection. The question of whether they can receive such when they are works created in collaboration with a human being are still being debated, as the question centers on what exactly is collaboration.
A lot of people seem to have the mistaken impression that there’s a database somewhere that contains the texts or images or vocalizations of every work an AI has been “trained” on. That the programs are finding appropriate lines of texts from existing works and patching them together.
This is not true. What resides on the systems used by AIs like ChatGPT* are not the actual texts, line by line, but the patterns and relationships identified by the learning programs and the neural network connections developed from that data. They’re not even a “Cliff Notes” summary of the works, but a collection of “patterns” derived from analyzing those works. This is the exact same thing that is done by people when they read a novel by an author they wish to incorporate some techniques by. How does Heinlein go about developing his characters? You read Heinlein’s works and begin to notice patterns… the kinds of characters he uses, the way he introduces them, the way he uses one as an educator and another as a “learned better” type. You don’t retain all of the detail, but you recognize patterns. The AIs are doing the exact same thing, although they have near eidetic memories and far better pattern recognition abilities.
The question is not one of whether or not infringement is involved, because it isn’t any more than an art student being encouraged to study brush technique by a master or a narrator listening to individuals with particular accents. If such a thing was actual infringement, the only authors writing legally would be those who had never read anything by anyone else.
Was the “reading” of the works in the database an illegal activity? At the very least, not at all in the case of public domain works. In Bradbury’s Fahrenheit 451, books are preserved by humans tasked with memorizing entire books. Is that memorization infringement? No. Once the book is obtained, the owner is free to do what they will with that copy – memorize it, use it as a doorstop, give it to a friend (though they can’t reproduce and sell it. Writing AIs do not do that). Current works that are available on the net? Not infringement either. If you can read it without infringing, the AI can read it without infringing as well. Library loans? Presumably, the publisher/author received their fees from the library, so, no, the AI program is doing the same thing a library patron does – borrowing a book, reading it, and returning it.
We’re not going to stop these systems from being utilized by fighting them with intellectual property claims, because they are essentially doing exactly what creatives do when they study their craft. We could, perhaps, mandate that any inclusion of a work that is currently under copyright or other IP protection receive some form of compensation in addition to advances and royalties, but doing so would require some changes to the law. What you’d really be saying in making such a claim is that James Cameron owes Ursula Le Guin, Harry Harrison, The Stravinsky brothers, Alan Dean Foster, and others “something” because he was clearly influenced by their works when he made Avatar. Much as I would like to see good SF authors receiving more and better compensation, it’s been pretty clear for a long time that “influence”, “inspiration”, “homage” and “in response to” works do not owe their “trainers” additional compensation.
Another misconception seems to be that these AIs are involved in a “creative process”. That they are somehow taking vast reams of memorized data and creating something new. They are not. They’re pattern recognition engines that are using those patterns in attempts to “fill in the blank”. They create nothing new from whole cloth. Show them a bunch of pictures of windows with drapes and they’ll “predict” that a mention of a window in a text should probably include drapes. The old programmer’s adage of GIGO (Garbage In, Garbage Out) is entirely applicable. Base a text-generating AI on a database of science fiction novels and the patterns it derives from those works will be patterns common to all literature (like in English direct and indirect objects usually follow the verb) and others almost unique to SF, such as most usages of FTL often involve discussions of time dilation. Trying to get it to turn out a coherent story about galactic empires when the database consists of Greek tragedies may result in some hilarities, but not a useable SF story.
At their base, plain and simple, these programs are tools. They’re having a disruptive effect, true, but so did the printing press (all those poor illuminators and scribes!), the typewriter, the word processor, outlining programs, the internet and on and on. Did any of those tools get banned or restricted because of the disruption? Maybe attempts were made, but clearly weren’t successful, so, No.
These programs are limited also. As was mentioned earlier, you’re not interfacing with the database, nor even the neural net trained on that database. You’re using a user interface with limitations, one of which is a relatively small short term memory. Plug in a prompt and the engine will create a “summary” (its analysis of the patterns it finds in your prompt) that’s limited to about 3,000 words. Those patterns are then used to find other patterns used to generate its output, from the “canned” version of its learning. You’ll find that when you add prompts to extend the generated text, the program will have forgotten the information from earlier prompts and you’ll have to keep on reminding it of detail in order to keep things consistent. Works beyond about 3,000 words require human intervention to maintain continuity, if nothing else. How is this fundamentally different than say, a William Patterson reviewing the work of one of his legion of ghost writers and requesting changes to maintain continuity?
In other words, a human author utilizing an AI engine to generate a story’s text is going to have to do a lot of writing, editing and re-writing (perhaps even more so than if they just put pen to paper themselves). In such a scenario, at what point do we say “that work should be rejected because the author used “tools” to help them write it?” (THAT novel was typeset from a manuscript written on a typewriter! We can’t have that!)
Where does that leave us? It leaves us on the crest of a wave of new technologies that are going to both disrupt our current ways of doing things and offer all manner of as-yet untold opportunities, some of which will, no doubt, inure to creator’s benefits. Could a fully realized writing AI, capable of churning out better than acceptable prose and guided by the creative mind of, say, a George R.R. Martin have allowed him to write more novels in the Song of Ice and Fire series more quickly? If the author himself vetted the final product, would those fans complaining about how long it takes to get a new volume keep complaining, or would they shut up because they got what they asked for?
What about unfinished works? Story fragments? Could we be on the cusp of an era in which it doesn’t matter that Heinlein, or Asimov or Clarke are no longer with us, because now we can get new works by them? (Their estates are appropriately compensated and rights appropriately assigned.)
What if such programs allowed creatives to produce more works – recognizably theirs, ones that pass all of their fans’ sniff tests – expand their audiences and make more money? Should we stop them from becoming more efficient?
I’m all for retaining the viability of human creatives in the marketplace (and improving their lot), but protesting and fighting against the continued use of these “AI” programs as it stands right now is like Joe McCarthy claiming he has lists of Communist infiltrators in our government, while actually holding up blank sheets of paper. Scare mongering only lasts until the truth is revealed.
In order to get to that truth, we’ve got to better understand what these programs are actually doing, so that we can uncover the ways in which we can use them to our advantage.
The Time: 1907
The Place: Bob’s Buggy Whip, Nosebag and Fancy Bridle Mfgs., Pty
“Seen any o’ them thar horseless carriages yet, Bob?”
“I have, Phil, I have. Not too keen on them, I’m thinking.”
“No, I imagine you wouldn’t be, what with your livelihood and all coming from horse fittings. Me, I think its just a fad.”
“Sure hoping you’re right about that, Phil. If they catch on, it’s going to put a lot of people just scraping by into the poorhouse. Why, what would the widow Mable do if there weren’t any road apples to be swept up, I ask you?”
“Say, Bob, you know, I hadn’t thought on it that deeply, but I think you’re right. I was talking to Arnie down at the tavern the other day and he said he was thinking of replacing his team with one of them “trucks”. Says he can haul more kegs in less time and for less money…thinking he can get rid of his stableboy, and his drovers. Might be able to use his savings to buy another tavern.”
“Mphmf. That’s concerning. He buys his feed from me. Gets his horses shoed here too. Sure hope others don’t start getting ideas like that.”
“Think there’s anything worth doing about it?”
“Well, I don’t rightly know. You got anything in mind?”
“Well sure! You and me are standing here fussin and botherin about it. We can’t be the only ones looking into the future and seeing black clouds. There’s all kinds of folks who’ll do poorly if those four-wheeled noise makers become popular.”
“Yes, I suppose you’re right about that.”
“So what we have to do, see, is get them all together and say “NO!” to the auty mobile. We won’t use ’em and we’ll have no truck with those that do! Why, I’ll bet we could start a political campaign, like the suffragettes.”
“I ain’t wearing no sash, Phil.”
“No one has to wear sashes. We’ll appeal to the monied interests. There’s horse breeders got a lot of money and this here is an existential threat to their very existence! I’ll bet they’ll put up some cash. And how many politicians you seen riding around in cars, Bob? Not many of them neither. Most of their constituencies ride horses too. Most of the voters do! Who’re people going to vote for, some rich moneybags riding around in an auto mo beale, trying to put them all out of work, or someone just like you an me with both feet in the stirrups?”
“You’ve got a very fine point there, Phil. I think the first thing we should do is write a manifesto, get all of our ideas down on paper, so we can share ’em with others that are like-minded.”
“That sounds dandy. But before we get started, how about adding a little kerosene to that lamp. We need better light for writing.”
*** Some time later. ***
“And I heard that Doc Martin, down at the apothecary, he’s buying a ‘steamer. Says he can get patients to the hospital faster. Just get in an’ go, he says. No having to get the horses out of the barn, harness ’em up – just get in and go, he says.”
“Well that sounds kind of useful. That poor Blackwood boy might have lived if they’d got to him faster.”
“And Mo, down at the Volunteer, says that they’re fundraising for a Fire Truck of all things. Got a water pump built right into it. No more bucket brigade.”
“And that nice Misses Appleby what teaches at the school house says her fiancé is going to start driving all the kids what live out on the farms to school in the morning. That way they can get their readin’ an writin’ and ‘rithmatic in and still get all their chores done!”
“That sure is nice of him.”
“And Mayor Prudhome says he’s got a fella coming all the way from Deeetroit, gonna talk to him about selling automobiles right here in town!”
(* And you’re not interacting with that database or program when you use one of these publicly available services, either. You’re dealing with a user interface that is a subset, optimized for the task of taking your prompts and churning out text. You interface with the bubble gum dispenser, not the factory making the gumballs, and, like the bubblegum dispenser, you can only ever get the flavor of gumballs out of it that were loaded into it.)