Clouds To Poem

WORK IN PROGRESS

Clouds to Poem is an installation where an AI gazes at clouds and speaks poetry about what it «sees». A sculptural cloud floats in the space, animated by projections of live sky footage. Visitors wear headphones and listen as the AI continuously observes, hallucinates objects in cloud formations, and speaks generated poems. A secondary screen reveals the machine's process in real-time: transformed images, object detection overlays, and emerging text, making visible what the AI «thinks» before it speaks.
Two projections overlap in this work: we project meaning onto clouds by seeing faces, animals, landscapes where there are none, and we project consciousness onto the AI, assuming it truly «sees». The machine performs the distinctly human act of daydreaming. What we witness is an algorithm engaged in radical unproductivity.

You cannot watch clouds faster.

This poetic inefficiency is intentional. Contemporary AI embodies optimisation culture: maximising efficiency, extracting patterns, delivering results. Clouds to Poem deliberately misuses this technology. Pattern recognition is applied where there is no correct answer. A system designed for productivity is tasked with purposelessness. The machine works hard at doing nothing useful. The technical pipeline amplifies this absurdity: cloud images pass through Deep Dream and Stable Diffusion, enhancing pareidolic potential. Object detection labels what it «sees». A language model responds with contemplative, sometimes nonsensical verse. The AI speaks its daydreams: a machine poet staring at the sky. The installation reclaims what efficiency cannot deliver: slowness, wonder, and the freedom to look at the sky for no productive reason at all.

The work is part of the series «Poetic Inefficiency» — installations that deliberately misuse AI for purposeless, contemplative ends. Currently in development.