In Latimer, Iowa, the elementary school has a little animal farm1. Fifth and sixth grade students are responsible, under the guidance of community volunteers, to raise and care for chickens, goats, and the occasional pig. The students feed the animals, change out bedding, monitor health, and scoop manure. But this isn’t a petting zoo. In November, the pig is sent to be butchered. The meat is returned to the school, and BBQ pork ribs are served in the cafeteria for lunch.
Maybe you’ve been to a Midwest county fair, strolled through the livestock barns, and observed the 4-H kids grooming their prize market cows—cows sold at auction after the fair and “converted” into hamburger. These kids appreciate where their protein comes from.
Or perhaps you follow one of those families (living in places like upstate New York or northern Michigan) who raise their own chickens, sheep, and goats. These families teach their children the facts of life—by which I mean that chicken soup requires butchering chickens.
I, on the other hand, am glad I can purchase chicken, pork, or beef that has been slaughtered in a place far away. Someone I don’t know bones and trims the carcasses, after which the cuts are trucked to my butcher shop and arranged neatly in a refrigerated grocery case. I find this very convenient, a real time-saver, no mess, no fuss. As I eat my stir fry, or my BLT sandwich, or my hamburger, I do not give credit to the very particular chicken, pig, or cow whose life was sacrificed for my supper. I don’t give any thought at all to the people who raised or butchered my meat. I do not ponder how the sausage is made.
Which makes me think about Google’s AI Overview.
Sometime during the past year, I noticed that Google responded to my questions with an answer, instead of with a list of links. This is very convenient!
- “What’s the best ice cream shop in St. Clair, Missouri?”
- “How do I transplant a rose bush?”
- “How do I make a frittata?”
Just like shopping for meat, I can pick up a pre-packaged answer to my question, and accept it, as is, with no further thought.
Google’s AI Overview, like ChatGPT and other Large Language Models (LLM’s), have been trained by systematically browsing through everything on the world wide web. These models crawl (really, really fast) through the internet and scrape up information. Their code has been written to look for connections, resemblances, causes and effects, similarities and differences. We could say the LLM corrals vast quantities of discreet ideas, sorts them by type, cuts and trims them, and packages them for consumption. Just like meat.
Google’s AI Overview provides links for the summaries, and links are also attached for the more detailed breakdown of each abstract. This, too, resembles the livestock industry, where, in the United States, at least theoretically, a cut of meat can be traced back to a specific animal.
But let’s face it. When I wanted to know the best ice cream shop in St. Clair, Missouri, the AI Overview response was sufficient. I plugged it into my GPS and never linked back to any clever blog about St. Clair and the “best places to eat”. Similarly, when I really needed to transplant that rose bush before it rained, the AI Overview was straightforward and to the point. I didn’t spend time linking back to any charming gardening website filled with beautiful photos carefully edited by the website’s author. That frittata? The AI Overview was all I needed. I was in a hurry, and I only wanted to know how long to cook the thing. The chefs who have experimented (broken a lot of eggs, so to speak) and published their results on the web—results that the LLM corralled and butchered and packaged? I didn’t need to acknowledge them or their efforts. I just picked up my metaphorical meat and consumed it.
When I shop at my literal butcher store, I pay for the meat I will consume. The market has established prices for chicken breasts, porkchops, and steak. Everyone in the production chain—except for the animals themselves—is compensated. When I pick up an answer at the Google AI Overview site, everyone in the production chain—except for the content creators—is compensated, albeit indirectly through ads placed in and near the AI Overviews.
The LLM’s consume immeasurable amounts of content—recipes, poems, scientific papers, gardening tips, photos, essays, travelogues, research results, music, novels, jokes, designs, movies—without ever asking permission, let alone compensating, the people who created those things. The advent of the world wide web had given these creators a huge stage—a platform—to showcase their creations. If they thought at all about protecting those creations, they probably assumed that copyright laws would be the mechanism to do so.
We have always recognized the value to society of allowing creators to benefit from their creations, at least for a set period of time. The first copyright law in the United States was adopted a mere three years after we adopted our Constitution. But copyright law, like all the laws in our common law system, builds on precedents and is dependent on specific circumstances and cases. In general, courts consider “fair use” when judging copyright violation. The judges consider the nature of the creation, the purpose of the use, and whether that use transforms the creation into something else entirely, as well as whether that use harms the market for the original creation2. Since the advent of LLM’s, creators of different stripes have filed lawsuits against the AI companies who consume their creations to train their LLM’s. Recently, two decisions came down from a District Court3, 4. In both cases, the judges ruled against the creators and for the AI companies, saying that the AI companies transformed the creations into something new. Therefore, they did not violate any copyrights.
The stage, or the platform, of the world wide web turned out to be a space without protection for the creations. The creations were like a litter of piglets that a mama pig could not own or shield from those who eventually consumed them.
Maybe these District Court decisions will be appealed, perhaps on the grounds that this “something new” is decimating the market for these creations5. Fewer people are visiting news sites, travelogues, how-to sites, and food blogs. Consequently, journalists, travel writers, craftsmen, and chefs are receiving less revenue. Meanwhile, the LLM’s continue to gobble, even as society attempts to untangle this snarl6. So, it’s all a little late, even if a higher court eventually rules in favor of the creators. The cow is not only out of the barn, but it’s been eaten.
So, I wonder what will happen next? Savvy writers, artists, musicians, photographers, comedians, and designers will either find a space safe from the LLM’s, which come like ravenous wolves in the night and consume without compensating, or they will quit producing. And then what will the LLM’s feast upon? As the best and brightest, and the most clever and original artists and creators figure out how to protect their work, the material left for the LLM’s to corral, butcher, and package might be lower quality stuff—glue-factory horses instead of pasture-fed cows. And if the quality of the LLM “diet” degrades, won’t the quality of the Overview answers degrade, too?
Change comes. Consequences follow. Laws are passed. Cases are judged. We adapt. It has always been so, and it will ever be thus. But maybe the elementary students in Latimer, Iowa, can provide a good example for those of us on the consumer side of the equation. The creators upon whose sacrifices we feast should not be taken for granted.
_______________
2 https://www.masslawblog.com/copyright/copyright-and-the-challenge-of-large-language-models-part-2-2/
5 https://www.wsj.com/tech/ai/google-ai-news-publishers-7e687141
6 https://www.wsj.com/tech/ai/hollywood-ai-copyright-a582502c
photo credit: Christina Maiia
