 |
| The kind of elf you’ll find in any Wizards of the Coast book, but AI-generated |
Once again a post from Monsters and Manuals about AI has prompted… thoughts.
I agree entirely with this statement:
The real ‘threat’ posed by AI is not that it will replace us, but that we will come more closely to resemble robots in our thoughts, behaviours, and opinions.
You can already see similar transformations play out in the worlds of pop music and movies: the ability to (relatively) simply produce anything within the bounds of our imaginations has resulted in a regression to the mean, creating soulless pap to appeal to the lowest common denominator. Generations who grow up knowing nothing other than these mediated mediocre media come to accept them as the norm, which will shape their own expectations, thoughts, and behaviour (this has already happened to expectations of sex as a result of easily-available hardcore pornography).
It’s depressing, but it’s going to happen. And there’s nothing we can do to stop it.
But it won’t happen for a while (and it might make sense for anyone writing about this topic to first have a play with the tools and discover the current state of the art).
Regular readers of this blog will know that I’ve posted a fair bit about AI (I really didn’t set out to do this but… rabbit holes), and I have set up The Mycoleum shop for T-shirts etc. featuring images shamelessly generated by AI.
This is the flowering of a lifelong quest to get computers to surprise me. When I started programming in the early 80s, my best friend was the random number generator. I used it to choose and combine, and was delighted every time the computer gave back something which I hadn’t explicitly asked for…
I lapped up Stephen Levy’s book Artificial Life when it came out in 1992. I played with flocking algorithms and developed variations on the Sorcerer’s Cave board game, generating random labyrinths and caves filled with arbitrary monsters and treasure. Sadly these, and my visual “flashperiments”, are no longer usable since Steve Jobs killed off the App Store’s biggest rival, Adobe Flash. But a couple of years ago I resurrected Flashperiments in the form of a JavaScript music visualiser, used to create this video for my friend Will’s track Ice (inspired by the Donna Kavan novel of the same name):
Most recently, I created the Twitter bot Nanodeities, which populates templates with selections from a long list of possibilities. By nesting templates within templates within template within… turtles, the possibilities are virtually limitless, and yet each is generating by combining words and phrases which I typed out myself.
AI is the realisation of my lifetime’s dream: finally, a computer can awe me with its unexpected creativity. Or rather, “creativity”. Anyway, I can finally get a computer to go to places I truly would never have dreamed of.
And this phrase, “would never have dreamed of”, is key. In the world of the Mycoleum, unpredictability is a feature. For most commercial creators, it is a bug.
The Mycoleum works because it is a world of unimaginable fantasy, inscrutable, understood only by the mushroom-folks who live there. It’s almost impossible to express just how surprised I am by the computer’s responses to my prompts. But what becomes clearer, with each picture that I try to coax out of it, is how little of my prompt it “understands”. This is best demonstrated with examples (all made using Midjourney).
I noticed today that somebody on the Yoon Suin reddit had posted an AI image of a crab-man. So I decided to make one of a slug-man. Below are the images generated, labelled with the prompts that generated them (minus my special sauce that makes the style of a keeping with my other AI images):
 |
|
The decadent slug-man ruler of Yoon Suin
|
The computer seems to have completely ignored the “slug” part of my prompt here, and gone with the slightly racist assumption that “Yoon Suin” must be some Fu Manchu-type character. I’ll ditch that bit, and focus more on making my ruler more slug-like…
 |
| A half-man half-slug emperor |
Still not seeing a lot of slug in this; wearing a slug is not the same as being a slug. Let’s be more specific about the body parts I’m after…
 |
| An emperor with the torso and head of a man and the hindquarters of a slug |
Oh FFS. I’ve no idea what’s going on here. Perhaps I should ditch the “man” part entirely because, come to think of it, I’m not really after a slug-centaur, just some kind of slug-with-personality. Let’s try…
 |
| A slug emperor |
You call those slugs??? I see frog, I see tortoise (at a stretch, if I squint, it could be a snail). I see aphid, iguana… bugger all slug. At this point I have zero expectations of success, but I’ll see what happens if I add a bit of human back into the mix…
 |
| A humanoid slug emperor. |
No, no, no, no, no, no, no. NO!
This “artifical intelligence” simply does not speak my language. It recognises the odd word, but not how sentences work. I am painfully reminded of the time I outsourced the building of a website to India and what I got back was… well, I’ve no idea what it was, everything got lost in translation. (I should add that this was over 20 years ago, things have changed a lot; in fact, I am collaborating with Bangalore-based Jayaprakash Satyamurthy on King Arthur vs Devil Kitty, and I would recommend him without hesitation to anyone looking for editorial assistance; his novel Strength of Water and his short story collection Weird Tales of a Bangalorean are also excellent).
The slug-emperor attempts were portraits, which is actually one of the things that AI does best (see elf-maiden pic above). As soon as other elements are introduced, you’re more likely to roll a 1 on a d10,000 than you are to get what you want. Describing a scene and having the AI render anything even vaguely approaching what you had in mind: forget it. Yesterday I spent an hour trying to get it to draw an oak sapling growing beneath a gorse bush. It couldn’t even draw a gorse bush. Even with very simple bog standard D&D prompts, it fails:
 |
| D&D magic-user casting fireball at an orc |
When I started writing the adventure Ship of Theseus (now Ship of Thorsus) I had planned to use royalty-free art from oldbookillustrations.com. The images I sourced were scarcely representative of the adventure content, and they were also used across many other OSR games and zines, but they were free. Once I started playing with AI art, I figured I could use that to generate images which were also free, but which were original and which more closely matched the adventure content. Could the AI draw a Viking ship whose every inch of surface bristled with vicious thorns? Could it heck as like.
There is certainly an art to crafting prompts for the AI, but even the most artful prompt-crafter in the world would struggle to get an AI to paint a scene like “the slugman emperor of Yoon Suin riding on a palanquin borne by crabmen through dusty streets lined with obsequious crowds of humans; the emperor bears a sceptre and an ivory back-scratcher”.
 |
| the slugman emperor of Yoon Suin riding on a palanquin borne by crabmen through dusty streets lined with obsequious crowds of humans; the emperor bears a sceptre and an ivory back-scratcher. |
This is the current state of play. But it will change. How long before an AI is able to approximate a scene in the manner in which its overlord intends? It could be anything between a year and twenty years, but I would estimate around 5.
Once that happens, the Borg will have arrived.
By the way, over at my personal blog I’ve just posted a look back over the (mostly) non-gaming aspects of my last year. I also have a new email newsletter.
Leave a Reply