Monday, May 22, 2023

One Writer's View on Generative AI

 One writer's thoughts:

The problem with programs like ChatGPT is a mix of copyright issues and the insidious destruction of careers and the average quality of all written material.


Example: What I do when I write a book is not what authors (many now) who lean heavily on AI do, and shouldn't be classified together, but platforms like Amazon don't make any distinctions. If the cover and the blurb are good, it's easy for people to buy a novel or rely on a nonfiction book without knowing the primary author was an AI program with no ethical rules or "talent."

Everything I've read before essentially creates fertile soil to grow a new creation. It's usually my subconscious that has notions like "There was a neat twist in a Jack Reacher novel, and something like it should happen here." My having that thought consciously is rare, although it does happen, and when it does I make sure everything I write is original except for the basic idea.


AI does the opposite. If I give it the basic idea "write a thriller about a Southern PI who comes to New York to find a murderer," and maybe some supporting ideas like "he has sex with a cop" and "his girl from Mississippi gets murdered while he's gone" the program is going to pick bits out of PI novels, New York novels, Southern-set murder mysteries, novels with sex scenes, and "fish out of water" novels. Nothing will be original except my idea and some randomness in the mixing. I know people who are doing that now just to add "novelist" to their LinkedIn profile.


Some writers say they only lean on ChatGPT for things like "give me five ways an 18th century person  might describe the front of a mansion." But all of them are lifted, often word for word, from other, copyrighted works.  


A famous author with lots of books could, if she had the time or the assistants, do searches and see how many times her best lines reappear in AI novels.  Human writers steal like that, too, but it's ALL AI text programs do, and they do it on a scale that lets millions of people "author" books just by stealing bits and pieces. In practice, that means life is getting far, far worse for novelists and nonfiction writers alike. The flood of bad submissions that buried publishers and editors when the internet matured is going to be multiplied a thousandfold: new authors may see their chances of being noticed by major publishers and top agents go from well below one percent today to an almost unmeasurable fraction of that.


It also affects the ethical authors and all the readers using platforms like Amazon. Amazon now gets 5,000 novels uploaded a day: when that multiplies, it's terrible for readers looking for quality and writers trying to stand out. All based on millions of little thefts. Except for old books out of copyright, AI is being trained ENTIRELY on copyrighted work. If original work is grown from the soil of all the stuff an author is previously read, with AI generation programs the soil is pre-loaded with seeds stolen from other gardens: all the "author" has to do is water it.


It's much more insidious for writers of short articles and listicles for web sites. They are simply being fired and replaced. Some of them are hacks and some have real talent, but there's nowhere left for them to take that talent.


The situation is worse, in a different way, for readers seeking to learn from nonfiction. You can say writers are responsible for the truth of their work, but the market will flood with books from authors who don't care, and the current problem of incorrect facts becoming part of history (see "J. Edgar Hoover wore dresses," pulled from an unsourced rumor by one writer and now "fact"). Publishers can't and don't check every fact and footnote: that's the author's job, but some authors won't try and those who do will find it impossible, because they've no idea where a supposed fact in their book came from, or whether a footnote is accurate without hunting for every one, killing the fake ones, and being left with facts with no sources. Peer reviewers can say "this is inaccurate trash" and get a book sent back for rework or dumped entirely, but much of what's published in books, magazines, websites, and pay-to-play journals never gets true peer review.


Unscrupulous contractors or even inside staff writers could give government or corporate clients work that only SMEs will be able to tell is wrong, and clients who trust them aren't going to check everything. Think what that can do to, say, a weapon evaluation report. Think what it does to adversaries who say things like, "Oh, an official report to the Air Force says this, so it has X characteristic that must be aimed at our radars and they must specifically be building it to attack us." That may not be a common occurrence, but it will happen sometime.


Anti-AI software is an arms race. What's scary is that AI generation program writers have a thousand times the audience, and the big money to be made money is in ever more capable programs that disguise a work's origins.


This is not "the buggy whip makers being driven out of business by cars" problem. It's "one of the foundations of civilization being degraded, twisted, and often destroyed." It's also not like another comparison, the way musicians grumbled about American Idol winners becoming stars without "paying their dues" in little clubs and bars. These kids had still developed their talent enough to produce good performances. AI programs let people skip THAT part of the work, too.  Authors who have never developed their talent are flooding the market with "books" that crowd out the real thing, when I've spent 35 years developing the skills of a good writer.  This is why some authors are using their skills on Twitter to come up with imaginative combinations of four-letter words to describe these programs and their sellers.  


So I and, I think, nearly all genuine authors think this is a kind of theft no matter how you squint at outdated copyright law.  We certainly don't think it's ethical, and we hope copyright law does something in the next revision - which may take years. The recent CRS interpretation of copyright rules and decisions (linked here) is helpful, and I think it generally agrees with me, but it's not law.  One court decision or change in Administrations could sweep it all aside.  


We can only keep raising our voices to legislators and regulatory bodies as we try to do our best.


 Matt Bille is a writer, historian, and naturalist living in Colorado Springs. He can be reached at mattsciwriter@protonmail.com. Website: www.mattbilleauthor.com.

Read Matt's Latest book, Of Books and Beasts: A Cryptozoologist's Library. This unique reference offers a friendly skeptic's 400 reviews of books on cryptozoology, zoology, related sciences, and cryptozoological fiction. Your search for the world's new and undiscovered animals begins here!


No comments: