Alakazam!
Huzzah! After a long (2 days) wait in a virtual line, I was finally allowed to use OpenAI’s GPT-4, which includes DALL-E, an image generator. If you have been paying attention, you know that artificial intelligence technology has been the subject of much hand-wringing and fear and strife. Labor negotiations now routinely include attempts by workers to limit the use of AI in specific industries, for fear that AI will take jobs away from people.
While I don’t yet see AI as an existential threat to humans (though of course, War Games and Colossus: The Forbin Project long ago ably imagined some worst-case examples of computers run amok), the concern over AI doing work that is currently done by people is well-founded. It has ever been thus: people have always used technology to automate work, and automation has often been met with resentment and even violence. The media portrays automation via AI as somehow different and inherently more menacing than automation of the paper-folding or car-painting variety, I suppose because it happens in computers, out of sight and not obviously under human control, and people therefore treat it as something other than just another machine created by people, doing work specified by people.
I recognize the uncanny aspect of computer-generated text and images, but I don’t view them as fundamentally different from wheat harvested by large combines rather than by men wielding scythes, or ships unloaded by a few robots instead of armies of Teamsters. AI writing and drawings are the same kind of things people can do, done with less exertion by a machine. People will find a way to automate things, and I can’t imagine a way to stop it. Please don’t talk to me about laws and digital methods to impede the use of automation technology—it is (or they are) to laugh. Increasing productivity at reduced costs will always win.
Now, I like to write and I want my writing to be read and enjoyed by people. I don’t make my living as a writer, so AI does not threaten my livelihood. Still, though, AI could conceivably make my written endeavours redundant in the marketplace by producing, in seconds and in unlimited quantities, work that looks just like mine. Notice the “in the marketplace.” I write mainly for the experience of writing, because it is a way for me to journey through my own imagination and create something from my thoughts. That experience would not be impacted if a machine somewhere was somehow producing the exact same written works. But I probably wouldn’t be able to interest many readers in my work, at any cost (even for free). Yes, losing engagement with readers, even the kind of engagement that happens across time (e.g. people reading my works long after my death) is a real loss, but at the moment, perhaps because I am not all that widely read, it does not bother me.
All of which is to say, yo, let’s draw some giraffes!
I did not intend to step through all those thoughts about AI as it relates to creative work. In fact, I will probably enrage professional artists by demonstrating how I intend to use DALL-E to produce illustrations for my own written (by me) works.
I can explain: the visual artist whom I am depriving of work by using DALL-E is none other than me, myself. Or, more to the point, a hypothetical, more skilled version of me, myself. You see, I am a case of Artistus Interruptus. My goal in life used to be to become a visual artist. I drew all the time from age six onward. I stopped pursuing that goal at age 21. I am now 63, and I have been retired for over 4 years. In those 4 years, I have actively pursued writing, relearning piano, learning bridge, and a little bit of programming. I have not yet made time to work on my drawing and painting skills, and I am not certain that I ever will. But I imagine the works I would produce if I had the skill. In that sense, my thinking about visual art is similar to my writing sessions, in that both engage my imagination. The difference is, I have taken the time and made the effort to hone my writing skills so that I can turn my imagination into something concrete; I have not taken the requisite time nor made the requisite effort to do the same thing with drawing and painting.
AI is a tool that will let me do that, and I feel no compunction against using that tool rather than, say, spending years honing my skills, or hiring an artist to turn my imagination into images.
Where was I? Oh yes: a giraffe, produced at my urging by DALL-E:
Not bad, but rather ordinary. A great thing about AI-generated art is that you can send the AI back to the drawing board, so to speak, and iteratively produce new versions of your picture until it represents what you saw in your imagination. I told it thanks, but could you make the giraffe a tad smug? I feel that giraffes with no attitude are easy targets for predators. Which led to:
Wow! That’s damn good art! But when I said “smug” I didn’t mean “French schoolboy smug.” This ‘raffe is not only predator-bait, he’s a certain target for other boy giraffes. I asked if he could be a bit more English, with perhaps an air of sophistication. Voila:
Better, but still not good. Perhaps “smug” was the wrong word. How about older, with a more confident air:
Much better! But how about even more sophisticated?
OK, I dig the hat, but that little pipe is ridiculous. More masculine, please—I am not exploring gender fluidity at this time. And when did I ever say give me busts of giraffes? May I have the whole animal please?
It took me a long time to notice that “DALL-E” is a pun on “Dali,” but this version brings it home. Note the chain attached to the monocle-free right eye, and the left hoof nestled in the crook of the right elbow, even though the entire left arm and *gulp* human hand is clearly visible holding the pipe aloft. *Shudder*
But still, I asked for a more intimidating giraffe, and boy howdy, I got one! Still not the whole body, though. Never mind. I asked for a bit more ‘tude, and DALL-E delivered:
Double wow! We’re onto something! I was just messing around, but I think I could sell this guy as the mascot for a hip craft brewery! Drink Smug Giraffe Ale, or don’t! If you don’t, cool people will! I would buy a case of that shit!
I have to admit, I didn’t really plumb the depths of my imagination before setting out on this journey, but I’m kind of thinking, in a way totally not controlled by a menacing machine in The Cloud somewhere, that this is the exact image I had in mind!
I feel, though, that this guy should have some obvious visual signal that he is the object of mass admiration—a visual way of luring the viewer in to buy whatever he is selling:
Whoa. The FTC might have something to say about this. Perhaps like this, with admirers, but less of the in-your-face giraffe bosoms:
By which DALL-E thought I meant to move the giraffe bosoms further into the background, I guess. I wonder, is that Giraffe smugness the smugness I was going for, or is it DALL-E looking out at me, relishing his inhuman power? And what’s with the gigando flame on the cig? I asked for a bit of context to justify all this attitude and the bikinis:
OK, sure, when twin tornadoes strike, summon your herd of non-cartoon giraffes and a bevy of now-human women in pornographic poses, and fire up your—what is that now, a backscratcher, or a mini-squeegee? Oh, and sure, don’t forget the balloons.
Just a few minutes in, and I’m starting to feel a twinge of the aforementioned fear that the media is blabbing on about. Look at how that guy is looking at you! I asked if perhaps my giraffe could be engaged in something more serious or historically relevant:
Well alright then! But we’re not going to sell any beer, are we? How about historically important, but with a tad more action?
Bold move! The salvation of all Mankind after The Flood was pretty damn historic! Skipped over a lot of minor history to get there, didn’t we? But even an AI loses the thread now and then. Where’s the giraffe? Please put him back in, and make it obvious that he is cognizant of the astounding nature of today’s historic events.
Thar he blows, and looking quite cognizant indeed, if not stunned into imbecilic catatonia! Is that a Dodo Bird back there? Let’s not be cultural colonists; don’t assume the viewer knows that this is (probably) the Ark. Make it more obvious.
OK, asshole. Back off just a little, and lose the ark-ified station wagon (though now I want to own one—diabolical this AI). And please restore the cartoonish nature of my giraffe.
Wetcove To Faduicas He Gas Nevads! Ark!! And wired for telegraph…nice!
Let me remind you, I asked for smug and not subject to being eaten and/or beaten up, and this is what I got. Clearly, guiding an AI is an art unto itself, one perhaps not amenable to automation. So there is hope for continued employment for artists, but employment of a different nature than wielding the pencil and brush oneself. HELP WANTED: People who can explain things in excruciating detail to a quite literal-minded AI, without suffering undue extremes of high blood pressure.
Now that we’ve had a bit of practice, let’s see how AI can help me with an actual project: producing illustrations for my story, Nice Ned the Squirrel, which I would like to publish as a children’s book. This is a project that I might never get around to if it depends on me developing my artistic skills. Come on, DALL-E, let’s do it!
First, I told DALL-E that I wanted a cartoon squirrel drawn in the style of Sir John Tenniel, who so brilliantly illustrated Alice in Wonderland. I asked that it be made obvious that Nice Ned is a helpful fellow. Boom:
Now, the message tee-shirt is a bit on the nose, but I find this drawing delightful. If I could just tweak it a bit…how about we put an acorn in his mouth, making one cheek bulge noticeably?
Or, you know, make his face look like a sheep’s ass with a head-sized nut jammed in the crack. Oh, and let’s make Ned morbidly obese instead of pleasingly muscular. Either/or, DALL-E. Let’s lose the acorn but add Moskowitz Wolf and Taliaferro Tortoise:
GAH!! You know, I’m liking the tortoise. But I didn’t say give me two squirrels, and I didn’t say make them look Satanic, and I didn’t say lose the charming Tenniel-like quality. Oh, and I said the wolf should be cartoonish and snarling whimsically as though he has a crick in his neck. I did not say give me White Fang about to eat all the squirrels.
My experiences with the giraffe and now with Nice Ned illustrate the fact that instead of trying to prompt DALL-E by referring to tweaks to prior art, you might need to work up a rather long, wordy and complete specification. Then if it doesn’t give you the desired result, tweak the spec but then submit it as a brand new work, rather than as changes to an existing one. I’m thinking maybe DALL-E doesn’t keep track of prior art very well, since it always seems to start nearly from scratch.
But I should be able to at least just say “make the wolf less scary,” right?
Sigh. Maybe not.
Once again, as with the giraffe whose gaze began to unnerve me, I find this wolf is making me fear DALL-E, and so I shall call it quits for now.
Boing!