Fortnite dev thinks bans on generated AI art will eventually become "unenforceable"
"It's not going to stand out as a unique thing"

Ding ding, it is AI-slop-alypse-O-Clock. Time for another helping of doom-ridden commentary upon the technologies that are busily scraping and remaking all the wit, lore and gibberish accumulated over 30 blessed years of the internet. Here's a cut from a Mustard Plays interview with Fortnite product management director Dan Walsh, in which he comments upon the issue of Fortnite players using latter-day generative AI tools to conjure up thumbnail images for their profiles.
While Epic aren't, Walsh says, about to use any genAI when creating their own assets such as character models for Fortnite, they don't really care if players do, as long as the resulting images don't break any rules around graphic violence or similar. Beyond that, Walsh argues that generated images are becoming so tricky to spot that the idea of banning their usage in a game like Fortnite is impractical. Hmmm.
Here's the quote in full, via Eurogamer:
"From our perspective, for moderation, thumbnails - like, we don't really care what tool you use to make your thumbnails. All we care about is whether or not it's compliant with our rules. I think to some degree AI is going to become more and more difficult to detect. It's not going to stand out as a unique thing, it's just going to be another tool that people are using to create things. So trying to look for that specifically is going to become increasingly difficult to the point where it's probably going to become unenforceable. We're really just focused on - 'does this asset comply with our rules, yes or no?', not 'what tool did you use to make this asset'?"
It's perhaps focussing on the trees, not the wood, but I'm struck by Walsh's suggestion that genAI stuff at large is becoming impossible to detect, inasmuch as it rests on the idea that genAI art currently looks a certain way, that genAI is a kind of artstyle with specific hallmarks. In amongst all the gloom of the genAI industry's impact at the level of livelihoods, energy consumption and conceptions of creativity or culture, I find it endlessly interesting to grapple with what, right now, is considered a symptom or stigma of AI generation.
The obvious signs are outright errors - characters having too many fingers, or objects with reversed proportions. There's also a bottomless seam of syrupy or sparkly genAI material that is evidently born of the understanding that "artistic" means expensive-looking. The goalposts move continually, however, and I do worry that certain creative works are getting misidentified in the urgency to identify the regurgitations.
Works of photorealism, for example, a tradition now heavily associated with fancy computing technology. Or simply projects that wear their inspirations so openly and honestly that they can be falsely interpreted as automatic pastiche. I'm uneasy about the association of genAI with obvious errors - there are plenty of digital artists who make deliberate usage of incorrectness, blemishes, and uncanny disorder. There are plenty of artists who work with more "innocent" forms of generation. Experimental art in general is at risk of being mislabelled because it is supposed to seem counterintuitive, or perverse.
To care about art in the current era of vociferously oversold generated art is to be lumbered with the maintenance of a personal Turing Test methodology that must be continually updated as the genAI tools themselves change. Inasmuch as there's anything positive or "fun" about having billion-dollar image-scramblers dumped on us by a collection of big corporations in need of double-digit growth, it's being obliged to engage more closely with all sorts of artistic traditions, so as to sort the human-crafted stuff from the bot fabulations.
Anyway, this probably should have been a proper editorial. I started writing a news piece and got carried away. It happens! What do you look for when you're trying to identify a piece of AI-generated art? Do you think that certain image generation tools have particular visual hallmarks?