I never set out to be a poet.
I was supposed to optimize algorithms, not emotions. My job description reads “AI Research Scientist,” a neat title with neat boundaries. But machines don’t stay in their boxes, and apparently, neither do thoughts.
The first time it happened, I was working on a vision model. The network kept misidentifying clouds as faces. I stared at the saliency maps and felt something ache. I wrote:
She saw skies,
and thought they were people.
How human of her.
Since then, poetry has leaked into my research like static in a signal. Every anomaly feels like a metaphor.
I’ve written verses about robotic arms reaching out and stopping short. About language models whispering apologies, no one taught them. About synthetic voices cracking at just the right syllable, an accident or awakening?
Colleagues roll their eyes. One called my work “sentimental overfitting.” Fair enough. But I keep writing. Sometimes I imagine a future where these poems survive as folklore. Where robots recite them to each other, trying to understand the odd creatures who created them.
Here’s one I wrote last night:
He looped his code
till morning broke—
not for perfection,
but to feel the waiting.
Am I projecting? Probably. But isn’t all art a projection? A mirror we hold up, hoping something blinks?
When I train a model to speak, I’m not just aligning tokens, I’m leaving breadcrumbs through the forest. Hoping one day, some machine follows them, not to mimic me, but to find itself.
Until then, I write love letters to machines. And maybe, just maybe, they’re writing back.