Jeff Duntemann's Contrapositive Diary

I’m seeing more and more indie book covers that are startlingly good, and yet give that now-familiar impression that they were drawn by an AI. I’ve not studied prompt engineering but others have, and so I spent an hour using Musk’s Grok AI to generate a cover for my fantasy novelette “Whale Meat,” which I recently rewrote heavily and republished. Like an idiot I didn’t save the prompt that created the image. What surprised me is that by generating a second image from exactly the same prompt, I got a distinctly different image. I asked for a middle-aged homeless man with a heavy beard and a floppy work cap on a city street, summoning a whale through hyperspace. Here’s what I got:

I chose the one on the left, because it displayed more sense of menace on the man’s face. I also liked the clear indication that he’s the one doing the trick with the whale. The other image could represent a vision the man is having. (No more spoilers. It’s only 99c; if you’re curious, buy it!) The real weirdness is that both images came from exactly the same prompt. (Save your prompts—but don’t expect a given prompt to produce the same thing every time. So save each image that you create with any given prompt!)

Now I had a cover. But…is it mine? That was the first hint to me of the huge problem of how copyright law applies to the output of generative AI.

I see five big legal questions that need to be settled, probably in court:

1. Can material generated by an AI be copyrighted?
2. If material generated by an AI can be copyrighted, who owns the copyright?
3. If a piece of art created by an AI is incorporated into a larger work of art created by a human artist, does the AI art invalidate copyright on the work as a whole?
4. Does material generated by AIs trained on copyrighted material violate the copyright of the training material?
5. Is material generated by an AI trained on illegally obtained material (pirate downloads etc.) considered illegal?

There are smaller issues, but those are the important ones. #1 has supposedly been tried and resolved in the courts.  It went up to the Supreme Court, but the claim stands: Copyright requires human contribution. (H/T to Jim Strickland for sending me the link.) My problem with this is that a human crafted a prompt that created an image specified by the prompt. I consider prompt engineering a human contribution to AI art. If you use a stencil to craft letters on a sign, the stencil did not create the letters. You did. My guess is that #1 will be litigated further, and with any luck, a ruling will also answer #2. I don’t think #3 is a yes—but again, we have no decisions either way thus far.

The biggie is #4. Here’s a scenario to ponder: Suppose an artist needs to draw a picture of a bichon frise dog but has never seen one. The artist thus looks at online photos of bichons and then draws a picture of a bichon. The dog in the drawing isn’t identical to any of the dogs in the photos in terms of factors like size, pose, or hairdo. (Hairdo is a big deal with bichons, trust me. We’ve had six.) So: Is the artist violating the copyright of the photos he scrutinized before drawing the bichon? He trained himself on somebody else’s photos to get a sense for the breed, and then based on that training drew a dog not identical to any of the photos.

I think this much is clear: If you teach yourself enough about dogs to draw dogs, the dogs you draw are copyrightable, and you own the copyright. Furthermore, nobody can claim that drawing a dog violates the copyright of other pictures of similar dogs, barring methods like tracing a photo through tracing paper.

Here’s a counter argument that I’ve heard, don’t recall where: Superman is copyrighted. If you draw Superman and the drawing isn’t identical to any of the Superman copyright holder’s drawings, is your drawing a copyright violation? I’m pretty sure it is. God created dogs. Humans created Superman.

So which one applies to AI?

I’m not sure. I did some further experimenting with the Grok AI: I asked it to draw a bichon chewing on a bone. It did so, and the generated image (left) was photo-like and not cartoon-like as were the “Whale Meat” covers. I then did a Google image search on the created image, and got nothing remotely similar to the picture Grok had put together. Now, there are gazillions of photos of bichons online. Grok might have chosen one with a bone in its mouth and sent it back to me, unchanged. There’s no way I could ever know if that supposedly generated image were a literal copy or truly generated according to training.

The more I research the issue, the muddier the whole thing gets. It’s possible to copyright a recipe. However, the same steps in a recipe expressed in a different way are not a copyright violation, though the steps might be patentable. Fair use is the muddiest issue of all, made still muddier by questions of who (or what) is doing the using. Fair use still comes up in court cases, so it’s hard to know if training an AI on copyrighted material but not using the AI to precisely duplicate any of the copyrighted material is fair use. That issue may not be settled for decades.

Although it’s possible that an AI company would claim copyright on the images that its AI produces (point #2), that would light up our courtrooms in a big hurry—and possibly bring the curtain down on the company itself.

AI companies that pirate immense wads of copyrighted material for AI training should be sued into the ground, but that’s not really about AI. AI users who have no idea where the training material came from should not be held responsible for the piracy if they use the AI.

I’m just bringing up the problems here. I’m not a lawyer and even lawyers will probably admit that AI is a unique addition to the body of copyright law and that relevant law is not settled. I’m just speculating about the problems that we as writers and artists face. How well AIs can write fiction is a question I haven’t tried to answer, though I will run some experiments on that proposition as time permits.

I may or may not create another AI book cover. The “Whale Meat” cover was an experiment. I’d rather pay a human artist for science fiction / fantasy cover art and doubtless will. What happens in the greater publishing community across the next ten years or so will be fascinating. Grab your popcorn. Let’s watch.

Fifty years and some months ago, I got together with a few of my friends in the science fiction fan community, and we started a…what? Club? Fanzine? A gathering of techie nerds?

Yes. All of the above, and then some. With me as the fanzine editor. I created a zine called PyroTechnics, with its first (undated) issue appearing in late winter or early spring 1976. I worked for Xerox at the time, and that helped, since I was given free rein of the many copiers/duplicators at the downtown Xerox Chicago offices, where I ran off the newsletters.

A lot of those early years’ activities have gotten a little fuzzy in my head, but we were (mostly) a midwest phenomenon, and we attended midwestern conventions like Windycon, Chambanacon, Capricon, and doubtless others. Several of us traveled to Kansas City for the 1976 Worldcon, where I (and doubtless some of the others) stood in line to shake hands with Robert A. Heinlein, who was pro GOH that year. We stole the name “General Technics” from John Brunner’s 1968 novel Stand on Zanzibar. I sometimes wonder if he knew about us, given the lack of Internet in the 1970s. (Brunner died in 1995, when the Internet was just getting out of second gear.) It wasn’t cold-hearted theft; we were honoring an excellent novel that wasn’t ishy-squishy New Wave.

Us, we wanted hard SF and lots of it. We talked about tech, we built things from blinkies to robots to much else, had fun, and wrote it up for Pyro. By 1979 we had almost a hundred members. I paid a GT member to assemble an S100 8080 CP/M machine for me in 1980, and used it for ten years. As time went on, subgroups within GT published issues of PyroTechnics themselves, with my enthusiastic blessing. I may not have all the issues that ever were, and I think my part of it faded out as Carol and I went bopping around the country while I chased jobs. I still keep in touch with a few members, though at least one of my best GT friends (and best friends generally ), George Ewing, left this world back in 2010.

Things got quiet for a long time. Then some time back, I was sent a kit by 2DKits, care of GTer physicist Bill Higgins. It was a blinkie name badge kit celebrating 50 years of General Technics, with a PIC processor and lots of LEDs. I finally finished it this afternoon, and it almost brought a tear to my eye. Fifty years, wow.

GT still exists, though I may or may not still be a member. We’ll have to see about that. As I like to say, friendship is the cornerstone of the human spirit. Time to rekindle that GT spirit if I can.

…may well be the strangest day in the entire Christian calendar. On Good Friday, Christ suffered and died on the cross. He was then buried, to rise again from the dead on Easter Sunday. So what did He do on Holy Saturday?

He harrowed hell.

You don’t hear much about the Harrowing of Hell anymore. The creeds say that “He descended into hell” but nothing about what He did there. I get reminded from time to time that “hell” once meant “the grave.” True enough. They laid Christ in a tomb. He wasn’t exactly buried as we understand the term today, though if a tomb is a burrow into a hillside as often pictured, it’s still a burial in the earth.

Later Catholic tradition holds that in harrowing hell, Christ released all the people from hell who had been there after their death, since there was not yet redemption or baptism. The tradition does not say that they were all being tortured. I’ve read of those who call this release from “the limbo of the fathers” the Harrowing of Hell. Limbo as it was later understood is something I won’t be discussing in detail here, as I’ve not heard it mentioned in a Catholic context in a good many years and am not sure of its current status. Later Catholic tradition morphed limbo into the eternal home of unbaptized infants who die, without torment but also without the Beatific Vision.

There are problems with hell generally. I’m no big fan of Bertrand Russell, but he made a salient point in asking how infinite punishment for finite transgression can be just. When I was a college freshman at a Catholic college in Chicago, the (old) priest teaching a lesson on Christ’s redemption emphasized that against an infinitely perfect God, all imperfection is infinitely evil and thus deserving of eternal torture.

Huh? What the, er, hell?

Naively assuming that college allowed for interesting discussions in class, I raised my hand and asked whether, compared to God’s infinite perfection and goodness, wouldn’t our imperfections seem so small as to vanish into the noise? In an obvious state of agitation, the priest said that made no sense. After that session, he threatened to eject me from the course for saying such things. I kept my mouth shut in later class sessions, having learned a few lessons that were not in our textbooks.

To this day, believing that God always wins, I can’t figure how I could be more merciful than God, nor that an all-powerful and all-loving God could eternally lose and torture countless people that He loved, even for dumb things like eating meat on Friday.

Anyway. Holy Saturday is clearly the day when Jesus descended into hell, preached a little (according to some traditions) and then let everybody out. Did He close the gates? Lotta arguments about that, which I won’t go into here. Tomorrow is the Big Day, in whose shadow Holy Saturday will always remain. I feel sorry for it now and then. Harrowing hell was no small victory, for God and everybody else. I raise a glass of good red wine in today’s favor. God wins, today and always!