You’ve seen the AI art. The hands with six fingers. The melting faces. The backgrounds that defy physics. But have you read the AI stories? They’re surprisingly good. That’s the paradox that’s been gnawing at me since I stumbled on a weird little project on Hacker News.
A developer posted a book of fables about threads and packets. One fable begins: ‘In one process there lived two threads, forked from the same beginning and taught from their first instruction that it is rude to seize what another holds.’ That line is beautiful. It captures the essence of concurrency using a simple social norm. The metaphor works. The story has moral weight. But the illustrations that accompany it? They’re a mess. Wonky perspectives, garbled text, bizarre proportions.
AI can write a fable about thread etiquette, but it can’t draw a coherent circle. That asymmetry is the key to understanding what’s really happening under the hood.
We think AI text is ‘surprisingly good’ because it’s mastered storytelling. The truth is darker: human narrative structures are formulaic. We’re pattern-matching machines. AI exploits that. Every fable follows a template—moral lesson, anthropomorphic characters, a neat resolution. AI has absorbed millions of these templates. It’s a remix machine, not a storyteller.
But visual perception? We’re unforgiving. A single inconsistent shadow ruins the illusion. A misplaced eye triggers our uncanny valley alarm. AI’s failure in art isn’t a weakness—it’s a mirror of our own cognitive strengths. We evolved to detect spatial inconsistencies because falling off a cliff hurts more than a poorly worded sentence.
One Hacker News commenter said: ‘The illustrations are not great but the stories are surprisingly good.’ That’s the consensus. But nobody asks why. The developer who built this didn’t realize they were creating a perfect experiment in cognitive science. They just wanted a book. But they accidentally revealed the boundary of machine creativity.
The real reason AI writes good fables? Because human stories are templates. We’ve been telling the same moral fables for millennia. AI just recombines them. The visual failure proves that true creativity requires spatial reasoning—something AI doesn’t have. Yet.
This isn’t a critique of AI. It’s a revelation about us. We’re easy to fool with words because we’re so eager to find meaning. But we’re hard to fool with images because we’re so good at seeing reality. The next time you scroll past an AI-generated image that looks like a Picasso after a stroke, remember: the text that accompanied it was probably brilliant. That’s not a bug. It’s a feature of how we think.
FAQ
Q: Isn't AI art improving fast? Won't this asymmetry disappear soon?
A: Yes, AI art is improving, but the fundamental asymmetry remains because solving visual consistency requires a different kind of world model—one that understands physics, occlusion, and perspective in a way that text generation does not. The gap may narrow, but it won't vanish until AI can simulate a 3D environment internally.
Q: What's the practical takeaway for developers and creators?
A: Use AI for text generation—it's reliable for narrative, summaries, and formulaic content. But don't rely on it for illustrations without human editing. The visual output is still a crapshoot. If you need coherent images, pair AI text with a human illustrator or use specialized image models that prioritize consistency over creativity.
Q: Could the fables actually be bad, and we're just impressed by novelty?
A: That's a valid contrarian take. The Hacker News commenter called them 'surprisingly good,' but novelty bias is real. The deeper point is that even if the fables are mediocre, they demonstrate a fluency in narrative structure that far exceeds AI's visual ability. The real contrarian view is that this asymmetry reveals more about the formulaic nature of human storytelling than about AI intelligence.